R Internals - with modifications for pqR

Table of Contents

R Internals

This is a guide to the internal structures of R, with modifications for pqR, plus coding standards for the core team working on R itself.

This document is based on the version for R 2.15.0 (2012-03-30).

ISBN 3-900051-14-3

Modifications for pqR are for pqR version 2.15.1 (2020-07-23).

1 R Internal Structures

The documentation here about R internal structures is a work-in-progress, and should be checked against the current version of the source code.

1.1 Configuration

Designing efficient layouts for objects requires knowledge of the sizes of various C data types. Accordingly, pqR is designed for a limited set of platforms, which however does encompass virtually all those that are current. Specifically, pqR can be configured for systems with either 32-bit C pointers or 64-bit C pointers, with all systems assumed to have both 32-bit and 64-bit versions of integers (with two’s complement representation of negative values), specifiable with the C99 types int32_t and int64_t and corresponding unsigned versions uint32_t and uint64_t. The C int type is also assumed in places to be identical to int32_t.

The R_len_t type is defined as int. For compatibility with code written for R-3.0.0 and later R Core versions, R_xlen_t and related symbols are defined, but are the same as the non-x versions (as is the case with R-3.0.0 for 32-bit platforms).

For all platforms, the --enable-compressed-pointers option to ./configure can be used to enable use of 32-bit compressed pointers. This makes sense even when uncompressed pointers are 32 bits in size, though the benefits may then be less than for 64-bit platforms. For 64-bit platforms, the --enable-aux-for-attrib option can be used to change how attributes are stored. For more information on these options, see the Compressed and uncompressed pointers in R Installation and Administration: manual.

1.2 Segmented Generational Garbage Collection

Memory allocation and garbage collection in pqR are done using the Segmented Generational Garbage Collection (SGGC) facility written by Radford M. Neal, and stored in src/extra/sggc. See the sggc-doc, sggc-imp, and sbset-doc files there for documentation on this facility. This directory also contains the files sggc-app.h and sggc-app.c, which specify how SGGC is used by the pqR interpreter.

The SGGC facility allocates objects in “segments”, each of which can contain several small objects, or a single big object. An object can be identified by a 32-bit “compressed pointer” consisting of the index of a segment and an offset within that segment. pqR may be configured so that compressed pointers are used throughout the interpreter, with the SEXP type being a compressed pointer. Alternatively, a SEXP can be an ordinary (uncompressed) pointer, with compressed pointers used only for interfacing to the SGGC facility.

The SGGC “type” of an object is not used to represent the R type, since this is incompatibile with the occassional use of SET_TYPEOF to change the R type. Instead, the SGGC type is used to specify what references need to be followed during garbage collection, and to control which objects are uncollected (currently SYMSXP, SPECIALSXP, and BUILTINSXP objects).

Five configurations are supported, as follows:

• 32-bit C pointers with SEXP a 32-bit compressed pointer
• 64-bit C pointers with SEXP a 32-bit compressed pointer
• 32-bit C pointers with SEXP a 32-bit uncompressed pointer
• 64-bit C pointers with SEXP a 64-bit uncompressed pointer
• 64-bit C pointers with SEXP a 64-bit uncompressed pointer, with attributes stored as “auxiliary information”, in order to allow tighter packing of the main data area.

The first two configurations have almost identical data layouts, with only external pointers and primitives differing due to the different sizes of C pointers.

1.3 SEXPs

The values of R variables and other objects can be thought of as either a SEXP (a pointer of some sort), or the structure it points to, a SEXPREC (and there are alternative forms used for vectors pointing to VECTOR_SEXPREC structures, and symbols and primitives). So the basic building blocks of R objects are often called nodes, meaning SEXPRECs, or VECTOR_SEXPRECs, or other such.

Note that the internal structure of the SEXPREC is not made available to R Extensions: rather SEXP is an opaque pointer, and the internals can only be accessed by the functions provided. Depending on how pqR is configured, a SEXP may be an actual memory address (an uncompressed pointer), which will be 32 or 64 bits in size, depending on the platform, or it may be a compressed pointer, 32 bits in size, that specifies the location of an object via a segment index and offset within a segment

Node structures of all types contain, in some fashion, a 32-bit sxpinfo header and a pointer to a pairlist of attributes. Additional information varies by type (a SEXPTYPE), which is specfied by a five-bit field in the sxpinfo header.

1.3.1 SEXPTYPEs

Currently SEXPTYPEs 0:10 and 13:25 are in use. Values 11 and 12 were used for internal factors and ordered factors and have since been withdrawn. Note that the SEXPTYPE numbers are stored in saved objects and that the ordering of the types is used, so the gap cannot easily be reused.

no	SEXPTYPE	Description
0	NILSXP	R_NilValue
1	SYMSXP	symbols
2	LISTSXP	pairlists
3	CLOSXP	closures
4	ENVSXP	environments
5	PROMSXP	promises
6	LANGSXP	language objects
7	SPECIALSXP	special functions
8	BUILTINSXP	builtin functions
9	CHARSXP	internal character strings
10	LGLSXP	logical vectors
13	INTSXP	integer vectors
14	REALSXP	numeric vectors
15	CPLXSXP	complex vectors
16	STRSXP	character vectors
17	DOTSXP	dot-dot-dot object
18	ANYSXP	make “any” args work
19	VECSXP	list (generic vector)
20	EXPRSXP	expression vector
21	BCODESXP	byte code
22	EXTPTRSXP	external pointer
23	WEAKREFSXP	weak reference
24	RAWSXP	raw vector
25	S4SXP	S4 classes not of simple type

Many of these will be familiar from R level: the atomic vector types are LGLSXP, INTSXP, REALSXP, CPLXSP, STRSXP and RAWSXP. Lists are VECSXP and names (also known as symbols) are SYMSXP. Pairlists (LISTSXP, the name going back to the origins of R as a Scheme-like language) are rarely seen at R level, but are for example used for argument lists. Character vectors are effectively lists all of whose elements are CHARSXP, a type that is rarely (never?) visible at R level.

Language objects (LANGSXP) are calls (including formulae and so on). Internally they are pairlists with first element a reference¹ to the function to be called with remaining elements the actual arguments for the call (and with the tags if present giving the specified argument names). Although this is not enforced, many places in the code assume that the pairlist is of length one or more, often without checking (though note that taking the CAR or CDR of R_NilValue gives R_NilValue without error, so disaster may be avoided).

Expressions are of type EXPRSXP: they are a vector of (usually language) objects most often seen as the result of parse().

The functions are of types CLOSXP, SPECIALSXP and BUILTINSXP: where SEXPTYPEs are stored in an integer these are sometimes lumped into a pseudo-type FUNSXP with code 99. Functions defined via function are of type CLOSXP and have formals, body and environment.

The SEXPTYPE S4SXP is for S4 objects which do not consist solely of a simple type such as an atomic vector or function. Prior to R 2.4.0 these were represented as empty lists.

1.3.2 Rest of header

The sxpinfo header is defined as the 32-bit C structure below:

struct sxpinfo_struct {

    /* First byte:   Whole of it allows quick checks for, eg, simple scalars. 
                     Parts give type and other information */

    unsigned int type_et_cetera : 8;    /* Bit fields as defined, 0 if use 
                                           is not specified below */

#   define TYPE_ET_CETERA_TYPE 0x1f /* Type of object, narrower than SEXPTYPE */

#   define TYPE_ET_CETERA_BEING_COMPUTED 0x20 /* Obj being computed in helper?*/

#   define TYPE_ET_CETERA_VEC_DOTS_TR 0x40 /* Symbol:   is ... or ..1, ..2, etc.
                                              Vector:   not scalar (length != 1)
                                              Function: is being traced 
                                              CHARSXP:  may need translation
                                              PROMSXP:  has been forced */

#   define TYPE_ET_CETERA_HAS_ATTR 0x80 /* Vector type, LISTSXP, or PROMSXP: 
                                             1 if ATTRIB != R_NilValue */

    /* Second byte.  Various flags, some of whose meaning varies with the
       type of the object. */

    unsigned int nmcnt : 3;   /* Count of "names" referring to object */

    unsigned int in_use: 1;   /* whether contents may be in use by a helper */

    unsigned int debug : 1;       /* Function/Environment: is being debugged
                                     Symbol: maybe use fast subassign */
    unsigned int rstep_pname : 1; /* Function: is to be debugged just once
                                     Symbol: subassign counterpart follows it
                                     CHARSXP: is used as a symbol's printname
                                     Envir: store gradient with variables
                                     Promise: evaluate with VARIANT_GRADIENT
                                     Vec type: full jacobian cached as attrib */
    unsigned int base_sym_env : 1;/* Symbol: has base binding in global cache,
                                     Envir: R_BaseEnv or R_BaseNamespace
                                     VECSXP: Holds gradients wrt list variable,
                                        as opposed to gradients _of_ list var */

    unsigned int obj : 1;     /* Set if this is an S3 or S4 object */

    /* The "general purpose" field, used for miscellaneous purposes */

    unsigned int gp : 16;     /* The "general purpose" field */

};

The debug bit is used for closures and environments. For closures it is set by debug() and unset by undebug(), and indicates that evaluations of the function should be run under the browser. For environments it indicates whether the browsing is in single-step mode.

The rstep_pname bit is used for closures to indicate "debugonce". It is used for CHARSXP objects to mark those that are used as the printname of a symbol, and hence should never be garbage collected (since symbols are never collected).

For environments, a flag (accessed with IS_BASE) set only for the base environment and namespace, and used to slightly speed up the check for whether an environment is one of these, which is done in time-critical code. For symbols, it is used for the BASE_CACHE flag, that indicates that the symbol is in the global cache with a binding in the base environment.

The field called nmcnt in pqR replaces the two-bit named field in R-2.15.0. The nmcnt field is accessed and set by the NAMEDCNT and SET_NAMEDCNT macros, by other macros for testing, incrementing, and decrementing the field, and also by the NAMED and SET_NAMED macros provided for compatibility with the previous scheme using named.

The nmcnt field is three bits in size, allowing for values from 0 to 7, which are interpreted as the number of bindings to variables or other references to the object, though in many cases, this is simply an upper bound on the number of references, since the count is not always decremented when a reference disappears.

This count of references allows pqR to more efficiently maintain the appearence that when an object is assigned to a variable, passed as the argument of a function, or stored as an element of a list, it is duplicated, so that subsequent changes to the original object do not modify the copy, and vice versa. This duplication can be deferred and sometimes avoided by instead simply incrementing the nmcnt field for the object.

See Reference counts and the nmcnt field for details.

The gp bits are called that because they are supposedly ‘general purpose’. This term is misleading, however, since at least one (bit 4) is not at all general purpose, but is instead reserved in all objects. The only meaning of ‘general purpose’ seems to have been that people using them felt no obligation to document their usage.

We label these bits from 0 to 15. Some uses of these bits are described below:

The bits can be accessed and set by the LEVELS and SETLEVELS macros, which names appear to date back to the internal factor and ordered types now used in only a few (no?) places in the code. The gp field is serialized/unserialized for the SEXPTYPEs other than NILSXP, SYMSXP and ENVSXP.

Bits 14 and 15 of gp are used for ‘fancy bindings’. Bit 14 is used to lock a binding or an environment, and bit 15 is used to indicate an active binding. Bit 15 is used for an environment to indicate if it participates in the global cache. See the next section on Environments and variable lookup for further details.

Bits 8-15 in a node in a pairlist of gradient values holds the index of the variable whose gradient is recorded, within the list in the relevant with gradient, track gradient, or active back gradient construct, starting with 1.

Bit 8-15 in a REALSXP or EXPRSXP representing an internal Jacobian matrix (not a user visible one, as for example from gradient_of) are used to specify whether a compact form of the matrix is used, and if so of what type. Zero means a full Jacobian matrix.

Bit 0-3 in an EXPRSXP representing a matrix product form of Jacobian matrix specify the order of products, and whether each is to be transposed first.

The macros MISSING and SET_MISSING are used for pairlists of arguments. Four bits are reserved, but only two are used. Bit 0 is used by matchArgs to mark missingness on the returned argument list, and bit 1 is used to mark the use of a default value for an argument copied to the evaluation frame of a closure. Note that this information is not captured by the use of R_MissingArg for the bound value in completely missing (no default) arguments, so this use is not redundant (though it could be reduced to one bit).

Bit 0 is used by macros DDVAL and SET_DDVAL. This indicates that a SYMSXP is one of the symbols ..n which are implicitly created when ... is processed, and so indicates that it may need to be looked up in a DOTSXP.

Bit 0 is used for PRSEEN, a flag to indicate if a promise has already been seen during the evaluation of the promise (and so to avoid recursive loops).

Bit 0 was formerly used for HASHASH, on the PRINTNAME of the TAG of the frame of an environment. (This bit is not serialized for CHARSXP objects.)

Bits 0 and 1 are used for weak references (to indicate ‘ready to finalize’, ‘finalize on exit’).

Bit 0 is used by the condition handling system (on a VECSXP) to indicate a calling handler.

Bit 4 is turned on to mark S4 objects.

Bits 1, 2, 3, 5 and 6 are used for a CHARSXP to denote its encoding. Bit 1 indicates that the CHARSXP should be treated as a set of bytes, not necessarily representing a character in any known encoding. Bits 2, 3 and 6 are used to indicate that it is known to be in Latin-1, UTF-8 or ASCII respectively.

Bit 5 for a CHARSXP indicates that it is NA_STRING or is in the CHARSXP cache; it is not serialized. Only exceptionally (never?) is a CHARSXP not set, and this should never happen in end-user code.

Bits 5 to 15 of the gp field are used for the table offset in a BUILTINSXP or SPECIALSXP. This offset is found again from the name when restoring saved data.

1.3.3 The ‘data’

A SEXPREC is a C structure containing the 64-bit header as described above, a pointer to the attributes, and the node data, a union

union {
    struct listsxp_struct listsxp;
    struct closxp_struct closxp;
    struct promsxp_struct promsxp;
} u;

All of these alternatives consist of three SEXP values.

The data for vectors, environments, symbols, and primitives (builtin and special) is held in different structures.

The vector types are RAWSXP, CHARSXP, LGLSXP, INTSXP, REALSXP, CPLXSXP, STRSXP, VECSXP, EXPRSXP and WEAKREFSXP. Such are a VECTOR_SEXPREC, which contains a sxpinfo header, an integer giving the length of the vector, an integer “truelength”, and the actual data. The data is aligned as required for a double and/or a pointer (even when the actual data is integer, or some other type not requiring this alignment).

The ‘data’ for the various types are given in the table below. A lot of this is interpretation, i.e. the types are not checked. In particular, the garbage collector assumes that all the “three-pointer” types have pointers that can be acessed as CAR, CDR, and TAG, even when they aren’t in a LISTSXP object, and that pointers in all vector types can be accessed with STRING_ELT even when they are not in a STRSXP.

NILSXP

There is only one object of type NILSXP, R_NilValue, with no data. However, asking for the CAR, CDR, or TAG of R_NilValue will actually return R_NilValue without error. Note that since there is only one NILSXP, the tests s==R_NilValue and isNull(s), which checks whether TYPEOF(s) is NILSXP, should be equivalent.

SYMSXP

Contains a pointer to the PRINTNAME (a CHARSXP) and the value in the base environment (SYMVALUE). Internal functions were previously held here also, but are now recorded elsewhere. The “attribute” for a symbol is used to hold its binding in the global cache, if this binding is not for the base environment, or R_NilValue if no such binding is currently recorded.

Note that currently symbols are never garbage collected, and this is sometimes relied upon in C code, making it difficult to change the scheme so that space for unused symbols can be recovered. Accepting this, pqR now treats symbols as uncollectable objects in the garbage collector.

There are three further pointers, used to record the last binding of the symbol that was found in an unhashed environment, and the last environment in which the symbol was not found (in a lookup for a function). See the next section on Environments and variable lookup.

The symbits field stores a 32-bit pattern with only a few 1s, fixed when the symbol is created, which is used to sometimes quickly determine that the symbol is not in an environment, in the style of a “Bloom filter”.

Finally, the hash value of the printname is redundantly recorded (except when compressed pointers are used, or uncompressed pointers are used with attribute in auxiliary information), just to make access to it be faster.

CLOSXP

Pointers to the formals (a pairlist), the body and the environment.

ENVSXP

Contains pointers to the frame, enclosing environment, and hash table (R_NilValue or a VECSXP). A frame is a tagged pairlist with tag the symbol and CAR the bound value. Also contains the logical OR of the symbits fields of all symbols that have ever been stored in the environment, allowing quick determination that some symbols cannot be in the environment. Finally, if the environment is for the body of a with gradient, track gradient, or active back gradient construct, it contains a vector list of symbols (and other info) for the variables whose gradient will be tracked, accessed by GRADVARS. This list is not seen by the garbage collector, and hence must be protected otherwise (by with gradient or track gradient).

PROMSXP

Pointers to the value, expression and environment (in which to evaluate the expression). Once an promise has been evaluated, the environment is set to R_NilValue. The attribute of a promise may hold gradient information, which is set when the promise is forced if requested.

LISTSXP

Pointers to the CAR, CDR (usually a LISTSXP or R_NilValue) and TAG (usually a SYMSXP or R_NilValue). For LISTSXP nodes used to record gradient values, the TAG is the environment for the relevant with gradient, track gradient, or back gradient construct; LISTSXP nodes used as binding cells may have such gradient information stored as their ‘attributes’.

LANGSXP

A special type of LISTSXP used for function calls. (The CAR references the function (perhaps via a symbol or language object), and the CDR the argument list with tags for named arguments.) R-level documentation references to ‘expressions’ / ‘language objects’ are mainly LANGSXPs, but can be symbols (SYMSXPs) or expression vectors (EXPRSXPs).

DOTSXP

A special type of LISTSXP for the value bound to a ... symbol: a pairlist of promises.

SPECIALSXP

BUILTINSXP

An integer (in the gp field) giving the offset into the table of primitives/.Internals, plus various information copied from that table for fast access. This extra information is set up when the offset is set with SET_PRIMOFFSET.

CHARSXP

length followed by a block of bytes (allowing for the nul terminator). The truelength field holds a hash value.

LGLSXP

INTSXP

length followed by a block of C ints (which are 32 bits on all R platforms).

REALSXP

length followed by a block of C doubles

CPLXSXP

length followed by a block of C99 double complexs.

RAWSXP

length followed by a block of bytes.

STRSXP

length followed by a block of pointers (SEXPs pointing to CHARSXPs).

VECSXP

EXPRSXP

length followed by a block of pointers. These are internally identical (and identical to STRSXP) but differ in the interpretations placed on the elements.

BCODESXP

For the ‘byte-code’ objects generated by the compiler.

EXTPTRSXP

Has three pointers, to the pointer, the protection value (an R object which if alive protects this object) and a tag (a SYMSXP?).

WEAKREFSXP

A WEAKREFSXP is a special VECSXP of length 4, with elements ‘key’, ‘value’, ‘finalizer’ and ‘next’. The ‘key’ is R_NilValue, an environment or an external pointer, and the ‘finalizer’ is a function or R_NilValue.

S4SXP

two unused pointers and a tag.

ANYSXP

This is used as a place holder for any type: there are no actual objects of this type.

1.4 Environments and variable lookup

What users think of as ‘variables’ are symbols which are bound to objects in ‘environments’. The word ‘environment’ is used ambiguously in R to mean either the frame of an ENVSXP (a pairlist of symbol-value pairs) or an ENVSXP, a frame plus an enclosure.

Some bindings of variables are ‘active bindings’, for which fetching and storing activate user-supplied functions. They are created by makeActiveBinding. Active bindings contain a function in their value cell. Getting the value of an active binding calls this function with no arguments and returns the result. Assigning to an active binding calls this function with one argument, the new value. Active bindings may be useful for mapping external variables, such as C variables or data base entries, to R variables. They may also be useful for making some globals thread-safe.

Bindings and environments can be locked against change. See the help for lockBinding and lockEnvironment.

Finally, there are additional places that ‘variables’ can be looked up, called ‘user databases’ in comments in the code. These seem undocumented in the R sources, but apparently refer to the RObjectTable package at http://www.omegahat.org/RObjectTables/. In pqR, S3 methods should not be stored in such user databases, since it is assumed that if a symbol fun.cls does not exist, then there is no fun method for class cls, which may not be true if the method exists in a user database.

The base environment is special. There is an ENVSXP environment with enclosure the empty environment, R_EmptyEnv, but the frame of that environment is not used. Rather its bindings are recorded in the SYMVALUE field of a symbol (which is R_UnboundValue if it has no value in the base environment). This environment contains heavily-used symbols such as + and for. Note that the “internal” functions are not bindings in any environment, but are recorded in a separate table. When R is started the primitive and internal functions are installed (by C code). Then .Platform and .Machine are computed and the base package is loaded into the base environment followed by the system profile.

The frames of environments, other than local environments of functions, are normally hashed for faster access (including insertion and deletion). The frame field is unused for hashed environments.

R maintains a global cache of ‘variables’ (that is symbols and their bindings) which have been found, and this refers only to environments which have been marked to participate, which consists of the global environment (aka the user workspace), the base environment plus environments² which have been attached. When an environment is either attached or detached, the names of its symbols are flushed from the cache. The cache is (usually) used when searching for variables when the global environment is reached, but not when directly searching some other environment that participates in the cache (since then bindings in the environment masked by others in the global cache should be visible).

Variables in the global cache whose binding within it is its binding in the base environment are marked so that they can be accessed quickly. (Since access to the global cache itself is now quick (the value is the ‘attribute’ for the symbol), this may now be an obsolete optimization.)

Symbols and environments have ‘symbits’ bit vectors that often allow for quick determination that an environment does not contain a symbol.

Every symbol contains a pointer to the environment in which the last unhashed binding for the symbol was found (provided it is not an active binding), along with a pointer to the binding cell that was found for that lookup. (These can also be set up in anticipation of their being needed, such as when function arguments are bound.) If a binding for the symbol is looked for again in this environment, the binding cell can be returned without a search (unless it has been set to R_UnboundValue due to deletion of the variable). Similarly, every symbol contains a pointer the the last unhashed environment in which a lookup for the symbol failed, provided that lookup was for a function, and either the lookup later succeeded, or the lookup was for an S3 method (for which lookups often fail).

These fields are cleared to R_NoObject (rather than being followed) by the garbage collector, so that they will not result in memory being occupied after the environment or the object bound is no longer used. Note that it is possible that although these pointers are set only for unhashed environments, such an environment may have become hashed later.

1.4.1 Search paths

S has the notion of a ‘search path’: the lookup for a ‘variable’ leads (possibly through a series of frames) to the ‘session frame’ the ‘working directory’ and then along the search path. The search path is a series of databases (as returned by search()) which contain the system functions (but not necessarily at the end of the path, as by default the equivalent of packages are added at the end).

R has a variant on the S model. There is a search path (also returned by search()) which consists of the global environment (aka user workspace) followed by environments which have been attached and finally the base environment. Note that unlike S it is not possible to attach environments before the workspace nor after the base environment.

However, the notion of variable lookup is more general in R, hence the plural in the title of this subsection. Since environments have enclosures, from any environment there is a search path found by looking in the frame, then the frame of its enclosure and so on. Since loops are not allowed, this process will eventually terminate: it can terminate at either the base environment or the empty environment. (It can be conceptually simpler to think of the search always terminating at the empty environment, but with an optimization to stop at the base environment.) So the ‘search path’ describes the chain of environments which is traversed once the search reaches the global environment.

1.4.2 Namespaces

Namespaces are environments associated with packages (and once again the base package is special and will be considered separately). A package pkg with a namespace defines two environments namespace:pkg and package:pkg: it is package:pkg that can be attached and form part of the search path.

The objects defined by the R code in the package are symbols with bindings in the namespace:pkg environment. The package:pkg environment is populated by selected symbols from the namespace:pkg environment (the exports). The enclosure of this environment is an environment populated with the explicit imports from other namespaces, and the enclosure of that environment is the base namespace. (So the illusion of the imports being in the namespace environment is created via the environment tree.) The enclosure of the base namespace is the global environment, so the search from a package namespace goes via the (explicit and implicit) imports to the standard ‘search path’.

The base namespace environment R_BaseNamespace is another ENVSXP that is special-cased. It is effectively the same thing as the base environment R_BaseEnv except that its enclosure is the global environment rather than the empty environment: the internal code diverts lookups in its frame to the global symbol table.

1.4.3 Hash table

Environments in R usually have a hash table, and nowadays that is the default in new.env(). It is stored as a VECSXP where length is used for the allocated size of the table and truelength is the number of primary slots in use—the pointer to the VECSXP is part of the header of a SEXP of type ENVSXP, and this points to R_NilValue if the environment is not hashed.

For the pros and cons of hashing, see a basic text on Computer Science.

The code to implement hashed environments is in src/main/envir.c. Unless set otherwise (e.g. by the size argument of new.env()) the initial table size is 29. The table will be resized by a factor of 1.2 once the load factor (the proportion of primary slots in use) reaches 85%.

The hash chains are stored as pairlist elements of the VECSXP: items are inserted at the front of the pairlist. Hashing is principally designed for fast searching of environments, which are from time to time added to but rarely deleted from, so items are not actually deleted but have their value set to R_UnboundValue.

1.5 Attributes

As we have seen, every SEXPREC has a pointer to the attributes of the node (default R_NilValue). The attributes can be accessed/set by the macros/functions ATTRIB and SET_ATTRIB, but such direct access is normally only used to check if the attributes are R_NilValue or to reset them. Otherwise access goes through the functions getAttrib and setAttrib which impose restrictions on the attributes. One thing to watch is that if you copy attributes from one object to another you may (un)set the "class" attribute and so need to copy the object and S4 bits as well. There is a macro/function DUPLICATE_ATTRIB to automate this.

Note that the ‘attributes’ of a CHARSXP are used as part of the management of the CHARSXP cache: of course CHARSXP’s are not user-visible but C-level code might look at their attributes. Also, ‘attributes’ on binding cells in environments, on promises, and on internal argument pairlists are used to record gradient information (see Gradient information for details). The ATTRIB field of a REALSXP vector representing a Jacobian matrix for a gradient may be used for purposes such as caching an expanded version of the Jacobian.

The code assumes that the attributes of a node (except for special uses noted above) are either R_NilValue or a pairlist of non-zero length (and this is checked by SET_ATTRIB). The attributes are named (via tags on the pairlist). The replacement function attributes<- ensures that "dim" precedes "dimnames" in the pairlist. Attribute "dim" is one of several that is treated specially: the values are checked, and any "names" and "dimnames" attributes are removed. Similarly, you cannot set "dimnames" without having set "dim", and the value assigned must be a list of the correct length and with elements of the correct lengths (and all zero-length elements are replaced by R_NilValue).

The other attributes which are given special treatment are "names", "class", "tsp", "comment" and "row.names". For pairlist-like objects the names are not stored as an attribute but (as symbols) as the tags: however the R interface makes them look like conventional attributes, and for one-dimensional arrays they are stored as the first element of the "dimnames" attribute. The C code ensures that the "tsp" attribute is an REALSXP, the frequency is positive and the implied length agrees with the number of rows of the object being assigned to. Classes and comments are restricted to character vectors, and assigning a zero-length comment or class removes the attribute. Setting or removing a "class" attribute sets the object bit appropriately. Integer row names are converted to and from the internal compact representation.

Care needs to be taken when adding attributes to objects of the types with non-standard copying semantics. There is only one object of type NILSXP, R_NilValue, and that should never have attributes (and this is enforced in installAttrib). For environments, external pointers and weak references, the attributes should be relevant to all uses of the object: it is for example reasonable to have a name for an environment, and also a "path" attribute for those environments populated from R code in a package.

When should attributes be preserved under operations on an object? Becker, Chambers & Wilks (1988, pp. 144–6) give some guidance. Scalar functions (those which operate element-by-element on a vector and whose output is similar to the input) should preserve attributes (except perhaps class, and if they do preserve class they need to preserve the OBJECT and S4 bits). Binary operations normally call copyMostAttributes to copy most attributes from the longer argument (and if they are of the same length from both, preferring the values on the first). Here ‘most’ means all except the names, dim and dimnames which are set appropriately by the code for the operator.

Subsetting (other than by an empty index) generally drops all attributes except names, dim and dimnames which are reset as appropriate. On the other hand, subassignment generally preserves such attributes even if the length is changed. Coercion drops all attributes. For example:

> x <- structure(1:8, names=letters[1:8], comm="a comment")
> x[]
a b c d e f g h
1 2 3 4 5 6 7 8
attr(,"comm")
[1] "a comment"
> x[1:3]
a b c
1 2 3
> x[3] <- 3
> x
a b c d e f g h
1 2 3 4 5 6 7 8
attr(,"comm")
[1] "a comment"
> x[9] <- 9
> x
a b c d e f g h
1 2 3 4 5 6 7 8 9
attr(,"comm")
[1] "a comment"

1.6 Gradient information

The value obtained from evaluating an expression may be accompanied by gradient information, if the evaluation is done with the VARIANT_GRADIENT variant (see The eval function for details on variants). After evaluation, the presence of variant information is indicated by R_variant_result having the VARIANT_GRADIENT_FLAG bit set, in which case the gradient information is in R_gradient. Note that R_gradient is not a root for the garbage collector, and so must be protected when necessary if it is being used.

VARIANT_GRADIENT is used when evaluating the body of a with gradient, track gradient, or active back gradient construct, and passed on from there, for example to the true or false branch of an if expression (but not when evaluating the condition in the if expression). VARIANT_GRADIENT is also used when evaluating the right-hand side of an assignment to a variable in an environment with the STORE_GRAD flag, which will be the case for an environment associated with with gradient, track gradient, or an active back gradient construct, and for environments of functions called from there.

Values assigned to variables in an environment with STORE_GRAD set have their gradient information stored as the attribute field of their binding cell, accessed in this context as GRADIENT_IN_CELL(x), and set by SET_GRADIENT_IN_CELL(x,g), with its presence tested by HAS_GRADIENT_IN_CELL(x). Gradient information can also be stored in the attribute field of a promise, when it is forced, if it has the STORE_GRAD flag, and this informatiion is accessed by the same macros. The internal evalList_gradient function evaluates a pairlist of arguments and returns a pairlist of values in which gradient information may be stored in the attribute fields of nodes, again for access with GRADIENT_IN_CELL.

The number of references in binding cells or promises to the pairlists, vector lists, and numeric vectors holding gradient information is indicated by their NAMEDCNT fields. This may sometimes be just an upper bound on the number of references, not the precise number. When gradient informations is stored in a binding cell or promise, NAMEDCNT is incremented for all parts of the gradient information, recursively, and correspondingly, NAMEDCNT may be decremented for all parts when a reference is removed. An attempt is made to avoid duplication of gradient information on subassignment operations when NAMEDCNT is one.

Gradient information takes the form of a pairlist of gradient values, in which the tag is not a symbol, but rather the environment of the with gradient, track gradient, or back gradient construct in which the variable for which the value is the gradient was defined. The particular variable within that environment is identified by GRADINDEX (the top 8 bits of the general purpose field) for the pairlist node, which indexes a table pointed by the GRADVARS field of the environment.

The gradient information for the gradient of a scalar quantity with respect to a scalar variable is just a scalar real. If the gradient is for a list, the gradient is a list of gradients for the list elements, with arbitrary depth of nesting. If the gradient is with respect to a list variable, the gradient information is a list with the GRAD_WRT_LIST flag set, with elements of this list being the gradients with respect to the elements of the variable (which may itself be a list, to arbitrary depth). For example, the gradient information for a list of 10 scalars with respect to a list of 20 scalars will be a list of length 20 with the GRAD_WRT_LIST flag set, whose elements are lists of length 10 without the GRAD_WRT_LIST flag set, which have scalar reals as elements.

Internal gradient values do not have names (they are instead added when the gradient information is converted to user-visible form), but the lists with GRAD_WRT_LIST set may have names.

Gradient information may be R_NilValue when the gradient is zero (or all zeros for lists). R_NilValue may also appear as a list element in a gradient value, indicating that the gradient for that portion of the variable is zero.

The gradient of a real vector with respect to a real vector is a real "Jacobian" matrix (which for internal versions may be missing its dim attribute), whose rows correspond to elements of the vector it is the gradient for, and whose columns correspond to elements of the vector with respect to which this is the gradient. Internal Jacobian matrices will have their TRUELENGTH field set to the number of numeric values the gradient is with respect to (ie, the number of columns in the Jacobian matrix) minus one (subtracting one allows use of constant objects, for which TRUELENGTH is 0). This value should be accessed with GRAD_WRT_LEN and set with SET_GRAD_WRT_LEN, which handle adding/subtracting one.

Jacobian matrices may be stored in a compact form (eg, just the diagonal elements if the matrix is diagonal), with the type of compact representation being in JACOBIAN_TYPE (which is top 8 of the gp bits), with zero if it is not a compact form. The full form of a compact Jacobian may be needed, and if computed is stored in the attribute field of the compact form (the compact representation may still be used when doing so is faster). A sxpinfo bit accessed with JACOBIAN_CACHED_AS_ATTRIB indicates if a cached form exists. Some compact forms also use the attribute field (accessed with ATTRIB_W to bypass normal checks), but if so, they must not be used after this is replaced by the cached full Jacobian. The GRAD_WRT_LEN field for the representation of a Jacobian matrix is always the number of columns in the full Jacobian matrix (which may not be the number of columns in some matrix used as part of a compact represention of the Jacobian matrix).

A Jacobian matrix may also be represented by a function closure, stored in an EXPRSXP vector, which has GRAD_WRT_LEN and other fields as for REALSXP vectors representing Jacobian matrices, including JACOBIAN_TYPE. Other compact forms may also be represented with EXPRSXP vectors.

1.7 Contexts

Contexts are the internal mechanism used to keep track of where a computation has got to (and from where), so that control-flow constructs can work and reasonable information can be produced on error conditions (such as via traceback), and otherwise (the sys.xxx functions).

Execution contexts are a stack of C structs. The exact declaration as below is used by RStudio, which breaks (when debugging) if it is changed (except by adding fields at the end):

typedef struct RCNTXT {
    struct RCNTXT *nextcontext;	/* The next context up the chain */
    int callflag;		/* The context "type" */
    JMP_BUF cjmpbuf;		/* C stack and register information */
    int cstacktop;		/* Top of the pointer protection stack */
    int evaldepth;	        /* evaluation depth at inception */
    SEXP promargs;		/* Promises supplied to closure */
    SEXP callfun;		/* The closure called */
    SEXP sysparent;		/* environment the closure was called from */
    SEXP call;			/* The call that effected this context */
    SEXP cloenv;		/* The environment */
    SEXP conexit;		/* Interpreted "on.exit" code */
    void (*cend)(void *);	/* C "on.exit" thunk */
    void *cenddata;		/* data for C "on.exit" thunk */
    void *vmax;		        /* top of R_alloc stack (should really be SEXP) */
    int intsusp;                /* interrupts are suspended */
    SEXP handlerstack;          /* condition handler stack */
    SEXP restartstack;          /* stack of available restarts */
    struct RPRSTACK *prstack;   /* stack of pending promises */
    SEXP *nodestack;
#ifdef BC_INT_STACK
    IStackval *intstack;        /* BC_INT_STACK seems to never be defined */
#endif
    SEXP srcref;	        /* The source line in effect */
    const struct R_local_protect *local_pr;  /* linked list of protected vars */
    SEXP scalar_stack;          /* Next unused position on scalar stack */
} RCNTXT, *context;

The ‘types’ are from

enum {
    CTXT_TOPLEVEL = 0,  /* toplevel context */
    CTXT_NEXT     = 1,  /* target for next */
    CTXT_BREAK    = 2,  /* target for break */
    CTXT_LOOP     = 3,  /* break or next target */
    CTXT_FUNCTION = 4,  /* function closure */
    CTXT_CCODE    = 8,  /* other functions that need error cleanup */
    CTXT_RETURN   = 12, /* return() from a closure */
    CTXT_BROWSER  = 16, /* return target on exit from browser */
    CTXT_GENERIC  = 20, /* rather, running an S3 method */
    CTXT_RESTART  = 32, /* a call to restart was made from a closure */
    CTXT_BUILTIN  = 64  /* builtin internal function - or .C, etc. */
};

where the CTXT_FUNCTION bit is on wherever function closures are involved.

Contexts are created by a call to begincontext and ended by a call to endcontext: code can search up the stack for a particular type of context via findcontext (and jump there) or jump to a specific context via R_JumpToContext. R_ToplevelContext is the ‘idle’ state (normally the command prompt), and R_GlobalContext is the top of the stack.

Calls to closures set a context, as do calls of builtins when profiling is being done. Calls of the foreign functions (.C, .Fortran, .External, and .Call) also set a context, with type CTXT_BUILTIN (even though in pqR these are now special primitives).

Dispatching from a S3 generic (via UseMethod or its internal equivalent) or calling NextMethod sets the context type to CTXT_GENERIC. This is used to set the sysparent of the method call to that of the generic, so the method appears to have been called in place of the generic rather than from the generic.

The R sys.frame and sys.call functions work by counting calls to closures (type CTXT_FUNCTION) from either end of the context stack.

Note that the sysparent element of the structure is not the same thing as sys.parent(). Element sysparent is primarily used in managing changes of the function being evaluated, i.e. by Recall and method dispatch.

CTXT_CCODE contexts are currently used in cat(), load(), scan() and write.table() (to close the connection on error), by PROTECT, serialization (to recover from errors, e.g. free buffers) and within the error handling code (to raise the C stack limit and reset some variables).

1.8 Argument evaluation

As we have seen, functions in R come in three types, closures (SEXPTYPE CLOSXP), specials (SPECIALSXP) and builtins (BUILTINSXP). In this section we consider when (and if) the actual arguments of function calls are evaluated. The rules are different for the internal (special/builtin) and R-level functions (closures).

For a call to a closure, the actual and formal arguments are matched and a matched call (another LANGSXP) is constructed. This process first replaces the actual argument list by a list of promises to the values supplied. It then constructs a new environment which contains the names of the formal parameters matched to actual or default values: all the matched values are promises, the defaults as promises to be evaluated in the environment just created. That environment is then used for the evaluation of the body of the function, and promises will be forced (and hence actual or default arguments evaluated) when they are encountered. (Evaluating a promise sets NAMED = 2 on its value, so if the argument was a symbol its binding is regarded as having multiple references during the evaluation of the closure call.)

If the closure is an S3 generic (that is, contains a call to UseMethod) the evaluation process is the same until the UseMethod call is encountered. At that point the argument on which to do dispatch (normally the first) will be evaluated if it has not been already. If a method has been found which is a closure, a new evaluation environment is created for it containing the matched arguments of the method plus any new variables defined so far during the evaluation of the body of the generic. (Note that this means changes to the values of the formal arguments in the body of the generic are discarded when calling the method, but actual argument promises which have been forced retain the values found when they were forced. On the other hand, missing arguments have values which are promises to use the default supplied by the method and not by the generic.) If the method found is a primitive it is called with the matched argument list of promises (possibly already forced) used for the generic.

The essential difference³ between special and builtin functions is that the arguments of specials are not evaluated before the C code is called, and those of builtins are. Note that being a special/builtin is separate from being primitive or .Internal: quote is a special primitive, + is a builtin primitive, cbind is a special .Internal and grep is a builtin .Internal.

Many of the internal functions are internal generics, which for specials means that they do not evaluate their arguments on call, but the C code starts with a call to DispatchOrEval. The latter evaluates the first argument, and looks for a method based on its class. (If S4 dispatch is on, S4 methods are looked for first, even for S3 classes.) If it finds a method, it dispatches to that method with a call based on promises to evaluate the remaining arguments. If no method is found, the remaining arguments are evaluated before return to the internal generic.

The other way that internal functions can be generic is to be group generic. Most such functions are builtins (so immediately evaluate all their arguments), and all contain a call to the C function DispatchGeneric. There are some peculiarities over the number of arguments for the "Math" group generic, with some members allowing only one argument, some having two (with a default for the second) and trunc allows one or more but the default method only accepts one.

1.8.1 Missingness

Actual arguments to (non-internal) R functions can be fewer than are required to match the formal arguments of the function. Having unmatched formal arguments will not matter if the argument is never used (by lazy evaluation), but when the argument is evaluated, either its default value is evaluated (within the evaluation environment of the function) or an error is thrown with a message along the lines of

argument "foobar" is missing, with no default

Internally missingness is handled by two mechanisms. The object R_MissingArg is used to indicate that a formal argument has no (default) value. When matching the actual arguments to the formal arguments, a new argument list is constructed from the formals all of whose values are R_MissingArg with the first MISSING bit set. Then whenever a formal argument is matched to an actual argument, the corresponding member of the new argument list has its value set to that of the matched actual argument, and if that is not R_MissingArg the missing bit is unset.

This new argument list is used to form the evaluation frame for the function, and if named arguments are subsequently given a new value (before they are evaluated) the missing bit is cleared.

Missingness of arguments can be interrogated via the missing() function. An argument is clearly missing if its missing bit is set or if the value is R_MissingArg. However, missingness can be passed on from function to function, for using a formal argument as an actual argument in a function call does not count as evaluation. So missing() has to examine the value (a promise) of a non-yet-evaluated formal argument to see if it might be missing, which might involve investigating a promise and so on ….

Special primitives also need to handle missing arguments, and in some case (e.g. log) that is why they are special and not builtin. This is usually done by testing if an argument’s value is R_MissingArg.

1.8.2 Dot-dot-dot arguments

Dot-dot-dot arguments are convenient when writing functions, but complicate the internal code for argument evaluation.

The formals of a function with a ... argument represent that as a single argument like any other argument, with tag the symbol R_DotsSymbol. When the actual arguments are matched to the formals, the value of the ... argument is of SEXPTYPE DOTSXP, a pairlist of promises (as used for matched arguments) but distinguished by the SEXPTYPE.

Recall that the evaluation frame for a function initially contains the name=value pairs from the matched call, and hence this will be true for ... as well. The value of ... is a (special) pairlist whose elements are referred to by the special symbols ..1, ..2, … which have the DDVAL bit set: when one of these is encountered it is looked up (via ddfindVar) in the value of the ... symbol in the evaluation frame.

Values of arguments matched to a ... argument can be missing.

Special primitives may need to handle ... arguments: see for example the internal code of switch in file src/main/builtin.c.

1.9 Autoprinting

Whether the returned value of a top-level R expression is printed is controlled by the global boolean variable R_Visible. This is set to TRUE for evaluation of constants and variables, and also on entry to every primitive or internal function, with subsequent behaviour depending on the setting in the eval column of the function’s specification (obtained with the macro PRIMPRINT). For primitive functions, R_Visible should be modified if necessary in the code for that function, though for non-special primitives if PRIMPRINT is 0 it will be set back to TRUE in any case. For internal functions, it may be modified, and this modification will be allowed to stand if PRIMPRINT is 2, but not if it is either 0 (set to TRUE) or 1 (set to FALSE).

The R primitive function invisible makes use of this mechanism: it’s PRIMPRINT value is 1, and it just sets R_Visible = FALSE and returns its argument.

Primitive or internal functions may call other functions which may change the visibility flag. For example, left brace (a primitive) will leave R_Visible as it was set by the call of eval for the last expression in the braces. Note that since eval is called when evaluating promises, even object lookup can change R_Visible.

The actual autoprinting is done by PrintValueEnv in file print.c. If the object to be printed has the S4 bit set and S4 methods dispatch is on, show is called to print the object. Otherwise, if the object bit is set (so the object has a "class" attribute), print is called to dispatch methods: for objects without a class the internal code of print.default is called.

1.10 The eval function

The eval function is the central routine in the interpreter. It takes as arguments an expression to evaluate and an environment (both of which must be protected by the caller) and returns the value that the expression evaluates to.

In pqR, eval calls evalv (or is a macro expanding to a call of evalv internally), which takes one additional argument, that may specify that a “variant” result is allowed. This argument should be 0 if no variant is allowed, as for plain eval. Symbols for other variants are defined in Defn.h.

For example, a caller of evalv might specify (by VARIANT_NULL) that the value will not be used (evaluation being done only for side effects), and may therefore be returned as R_NilValue, or (by VARIANT_SUM) that if the value is a numeric vector, only the sum of vector elements is needed, so that this sum may be returned rather than the vector (which will then not need to have space allocated for it). However, the caller of evalv must always be prepared to receive an ordinary value, rather than the variant asked for.

For some variants (eg, VARIANT_SUM), there is no need to specially identify that the value returned is the variant. For others (eg, VARIANT_TRANS, which may return the transpose of the result), a variant result is identified by the global variable R_variant_result being set to a non-zero value after the return of evalv. The caller of evalv should note the contents of R_variant_result and then reset it to zero before it could taken as indicating that a later return is of a variant. (But note that evalv will itself start by setting R_variant_result to zero, so there is no need to set it just before calling evalv.)

If the expression evaluated is a call of a primitive funciton, the variant argument of evalv will be passed on to the C function implementing this primitive if the appropriate flag is set in the primitive’s specification (see below). Variant arguments may be passed onwards from the expression passed to evalv — for example, from an “if” statement to the evaluation of the chosen branch, or from a function call to the evaluation of the body of the function — with any variant result propagated back, as controlled by the VARIANT_PASS_ON macro.

Primitive functions (objects of type BUILTINSXP and SPECIALSXP) are implemented using C functions that by convention have names of the form do_XXX. These functions are associated with symbols by tables in the source files that define the functions, which are copied to a single table for all such functions by initialization code in names.c. This table also has other information, such as whether the primitive is BUILTIN or SPECIAL, and whether the variant argument of evalv is passed on to the do_XXX function. (Currently, SPECIAL primitive functions must be called with a variant argument, so that a check for this in time-critical code can be avoided.)

Arguments are passed to the do_XXX function as a pairlist. This is inefficient, since it requires that a CONS cell be allocated for every argument passed to a primitive. In pqR, a faster interface is provided for simple cases of unary primitives, in which the argument is unnamed, and are not S3 or S4 objects. This interface calls a do_fast_XXX function to handle such simple cases, which receives the argument of the primitive directly as a C argument. These functions are also set up with tables in the various source files and initialization code in names.c. For details, see names.c, the documentation in the code for evalv, and the definitions for PRIMFUN_FAST and related macros in Rinternals.h.

Alternatively, creation of a pairlist of evaluated arguments may be avoided by marking the primitive as SPECIAL, even if it will always eventually evaluate its arguments, so that the arguments may be stored in local variables without creating a pairlist.

Some primitives use a “scalar stack” that may contain INTSXP or REALSXP vectors of length one, with no attributes. Values may be returned on this stack if the VARIANT_SCALAR_STACK_OK bit in the variant is set, and if so returned, must be popped off by the caller. Such values must never be allowed to appear in persistent data structures. It is permitted, but unnecessary, for them to be “protected” from garbage collection. See the source file src/main/scalar-stack.h for details.

1.11 Reference counts and the nmcnt field

There is evidence in the code that R at one point used a single-bit named field in an object, to indicate whether the object was ‘named’ — that is, was the value of some variable. Later, this field was expanded to two bits, though only values 0, 1, and 2 were possible. A value of 2 for named indicated that the object is (or at least might be) the value of more than one variable (or otherwise could be referenced in more than one way), and therefore would need to be duplicated before being modified when accessed via any of these references. The value of named was never decremented in this scheme, leading to many unnecessary duplications.

In pqR, a true reference counting scheme is implemented, using a nmcnt field (currently three bits in size) that records how many variables, function arguments, list elements, or other references an object has. If this number is greater than the maximum that is possible (MAX_NAMEDCNT, currently 7), the maximum value is stored. Currently, nmcnt is not always decremented when a reference ceases to exist, so nmcnt is merely an upper bound on the number of references. In particular, note that once nmcnt reaches MAX_NAMEDCNT, it can never be decremented, since it is not known how many references beyond this number actually exist.

For compatibility with C code written for use with the previous scheme, the NAMED and SET_NAMED macros are still defined. SET_NAMED sets nmcnt to the value given if it is 0 or 1, and to MAX_NAMEDCNT if the value given to SET_NAMED is 2. NAMED delivers the value of nmcnt if it is 0 or 1, and the value 2 if nmcnt is any value greater than 1. Furthermore, if NAMED delivers the value 2, it changes nmcnt to MAX_NAMEDCNT. This is necessary because existing code may assume that if NAMED is 2, it will never decrease, and hence the object will always be duplicated before being modified, regardless of subsequent operations.

The nmcnt field can be accessed by the macro NAMEDCNT, but when appropriate, macros for testing its value called NAMEDCNT_EQ_0, NAMEDCNT_GT_0, and NAMEDCNT_GT_1 should be used, since they may be more efficient than a comparison with the value returned by NAMEDCNT.

When use of helper threads or task merging is enabled, an object that is being used by a task that has not yet completed must not be modified, even if its NAMEDCNT field is such that it could otherwise be modified. This constraint is implemented by having the NAMEDCNT macro wait until no task is using the object before returning the value. (But macros like NAMEDCNT_EQ_0 can sometimes avoid a wait.) See Helper threads and task merging for details.

The nmcnt field can be set with the SET_NAMEDCNT macro, but when possible, it is better to use the SET_NAMEDCNT_0, SET_NAMEDCNT_1, SET_NAMEDCNT_MAX, INC_NAMEDCNT, DEC_NAMEDCNT, and INC_NAMEDCNT_0_AS_1 macros, since they may be more efficient. The INC_NAMEDCNT and DEC_NAMEDCNT macros increment and decrement nmcnt, except that they do not change its value if it is MAX_NAMEDCNT, and DEC_NAMEDCNT does not change a value of zero. The INC_NAMEDCNT_0_AS_1 macro first changes nmcnt to 1 if it is 0, and then increments the value (unless it is MAX_NAMEDCNT).

In detail, pqR aims to follow (but may not yet fully follow) the (tentative) conventions below regarding how NAMEDCNT should be set, and when an object may be modified:

• Pairlists, language constructs, and closures (LISTSXP, LANGSXP, DOTSXP, CLOSXP, and BCODESXP objects) should never be modified — all changes must be implemented by duplicating all or part of the object — except for internal uses of such objects that are never visible to user R code (eg, lists of bindings in environments and lists of attributes), and when such objects are in the process of being constructed. This immutability is the easiest way of, for example, ensuring that expressions being evaluated are not changing during the course of their evaluation. Accordingly, NAMEDCNT is not meaningful for such objects.

  An atomic or non-atomic vector that is referenced from an object of one of these types (eg, a numeric constant in a language construct) should have its NAMEDCNT set to MAX_NAMEDCNT whenever it becomes visible elsewhere (eg, as the result of eval) to ensure that it is never changed.

• Environments (ENVSXP objects) may always be modified (ie, bindings within them created, deleted, or changed), since their specified semantics is that the same environment may be referenced in several ways, with changes to the environment seen via all references. Accordingly, NAMEDCNT is not meaningful for environments. (It is possible that in future NAMEDCNT will be maintained for environemnts so that when it is decremented to zero the NAMEDCNT in all values bound in the environment can be decremented, but this is not attempted at present.)
• Promises (PROMSXP objects) should have NAMEDCNT set to the number of bindings of the promise in an environment. This count will usually be one, but may be greater as a result of method dispatch. (This count is not yet properly maintained in pqR, so at present it must always be assumed to be greater than one.)
• NAMEDCNT for atomic and non-atomic vectors (objects of type LGLSXP, INTSXP, REALSXP, CPLXSXP, STRSX, RAWSXP, VECSXP, and EXPRSXP) and for objects of type S4SXP should be set to the count how many references exist to the object in any of the following ways:
  • as the value bound to a name in an environment.
  • as the value of an attibute of an object for which NAMEDCNT is non-zero.
  • as an element in a list (VECSXP or EXPRSXP) for which NAMEDCNT is non-zero.
  • as the value of a promise for which NAMEDCNT is non-zero.

  However, NAMEDCNT for a vector or S4SXP objects is allowed to be zero if there is only a single reference to it as an attribute or a list element. If such a vector is made visible otherwise (eg, as the value returned by eval), and NAMEDCNT for the object referencing it is still non-zero, its NAMEDCNT must be set to 1, so that it will be correctly seen and maintained later. This convention is convenient because it sometimes avoids the need to set NAMEDCNT to 1 for all elements when creating a list, and because it sometimes automatically results in NAMEDCNT being zero for an element of a list that has NAMEDCNT of zero (either because it was always zero, or because it became zero after NAMEDCNT was decremented, for example in get_rm).

  Note that this convention applies recursively. For example, when a list with a non-zero NAMEDCNT contains a list with zero NAMEDCNT which in turn contains a real vector with zero NAMEDCNT, the list and real vector with zero NAMEDCNT are considered to really have NAMEDCNT of 1, and must have NAMEDCNT set to 1 if they become visible elsewhere.

  Note also that an object may reference another object more than once, in which case NAMEDCNT for the referenced object must account for all such references. For example, an object that appears twice in a list, and is otherwise unreferenced, should have its NAMEDCNT field set to 2.

• NAMEDCNT is not meaningful for the remaining types of objects (NILSXP, SYMSXP, SPECIALSXP, BUILTINSXP, CHARSXP, EXTPTRSXP, and WEAKREFSXP). These objects are either never modified, or modifications are intended to be visible everywhere.

It is always permissible (though undesirable) to set NAMEDCNT to a value greater than it needs to be according to the above conventions. Also, note that setting and accessing NAMEDCNT is permitted for all objects, even those for which it is not meaningful,

1.12 The write barrier and the garbage collector

The write barrier enforces use of (for example) SETCAR(x,y) rather than CAR(x)=y, which will produce a compiler error. The write barrier is essential for generational garbage collection to work correctly.

The write barrier is overridden only in src/main/memory.c, src/main/access.c, and src/extra/sggc/sggc-app.c, in order to perform low-level memory management operations and to implement SETCAR, etc. Overriding it is done by redefining the NOT_LVALUE macro.

For background papers see http://www.stat.uiowa.edu/~luke/R/barrier.html and http://www.stat.uiowa.edu/~luke/R/gengcnotes.html.

1.13 Serialization Formats

Serialized versions of R objects are used by load/save and also at a slightly lower level by saveRDS/readRDS (and their earlier ‘internal’ dot-name versions) and serialize/unserialize. These differ in what they serialize to (a file, a connection, a raw vector) and whether they are intended to serialize a single object or a collection of objects (typically the workspace). save writes a header at the beginning of the file (a single LF-terminated line) which the lower-level versions do not.

save and saveRDS allow various forms of compression, and gzip compression has been the default (except for ASCII saves) since R 2.4.0. Compression is applied to the whole file stream, including the headers, so serialized files can be uncompressed or re-compressed by external programs. Since R 2.10.0 both load and readRDS can read gzip, bzip2 and xz forms of compression when reading from a file, and gzip compression when reading from a connection.

R has used the same serialization format since R 1.4.0 in December 2001. Earlier formats are no longer supported in pqR. The current serialization format is called ‘version 2’, and has been expanded in back-compatible ways since R 1.4.0, for example to support additional SEXPTYPEs.

save works by writing a single-line header (typically RDX2\n for a binary save: the only other current value is RDA2\n for save(files=TRUE)), then creating a tagged pairlist of the objects to be saved and serializing that single object. load reads the header line, unserializes a single object (a pairlist or a vector list) and assigns the elements of the object in the specified environment. The header line serves two purposes in R: it identifies the serialization format so load can switch to the appropriate reader code, and the linefeed allows the detection of files which have been subjected to a non-binary transfer which re-mapped line endings. It can also be thought of as a ‘magic number’ in the sense used by the file program (although R save files are not yet by default known to that program).

Serialization in R needs to take into account that objects may contain references to environments, which then have enclosing environments and so on. (Environments recognized as package or name space environments are saved by name.) There are ‘reference objects’ which are not duplicated on copy and should remain shared on unserialization. These are weak references, external pointers and environments other than those associated with packages, namespaces and the global environment. These are handled via a hash table, and references after the first are written out as a reference marker indexed by the table entry.

Version-2 serialization first writes a header indicating the format (normally ‘X\n’ for an XDR format binary save, but ‘A\n’, ASCII, and ‘B\n’, native word-order binary, can also occur) and then three integers giving the version of the format and two R versions (packed by the R_Version macro from Rversion.h). (Unserialization interprets the two versions as the version of R which wrote the file followed by the minimal version of R needed to read the format.) Serialization then writes out the object recursively using function WriteItem in file src/main/serialize.c.

Some objects are written as if they were SEXPTYPEs: such pseudo-SEXPTYPEs cover R_NilValue, R_EmptyEnv, R_BaseEnv, R_GlobalEnv, R_UnboundValue, R_MissingArg and R_BaseNamespace.

In the current version of pqR, objects stored as shared constants (probably in read-only memory) are written with a flag bit saying they are constants (unless this is disabled with the nosharing argument of serialize). This flag will be ignored by earlier versions of pqR or R that didn’t have such constants, but will cause the current version of pqR to unserialize them as the same constants. (A check is done that other aspects of the object match the constant, for compatibility with future versions that might have a wider variety of such constants.)

For all SEXPTYPEs except NILSXP, SYMSXP and ENVSXP serialization starts with an integer with the SEXPTYPE in bits 0:7⁴ followed by the object bit, two bits indicating if there are any attributes and if there is a tag (for the pairlist types), an unused bit and then the gp field⁵ in bits 12:27. Pairlist-like objects write their attributes (if any), tag (if any), CAR and then CDR (avoiding tail recursion on CDR): other objects write their attributes after themselves. Atomic vector objects write their length followed by the data: generic vector-list objects write their length followed by a call to WriteItem for each element. The code for CHARSXPs special-cases NA_STRING and writes it as length -1 with no data.

Environments are treated in several ways: as we have seen, some are written as specific pseudo-SEXPTYPEs. Package and namespace environments are written with pseudo-SEXPTYPEs followed by the name. ‘Normal’ environments are written out as ENVSXPs with an integer indicating if the environment is locked followed by the enclosure, frame, ‘tag’ (the hash table) and attributes.

In the ‘XDR’ format integers and doubles are written in bigendian order: however the format is not fully XDR (as defined in RFC 1832) as byte quantities (such as the contents of CHARSXP and RAWSXP types) are written as-is and not padded to a multiple of four bytes.

The ‘ASCII’ format writes 7-bit characters. Integers are formatted with %d (except that NA_integer_ is written as NA), doubles formatted with %.16g (plus NA, Inf and -Inf) and bytes with %02x. Strings are written using standard escapes (e.g. \t and \013) for non-printing and non-ASCII bytes.

1.14 Encodings for CHARSXPs

Character data in R are stored in the sexptype CHARSXP. Until R 2.1.0 it was assumed that the data were in the platform’s native 8-bit encoding, and furthermore it was quite often assumed that the encoding was ISO Latin-1 or a near-superset (such as Windows’ CP1252 or Latin-9).

As from R 2.1.0 there was support for other encodings, in particular UTF-8 and the multi-byte encodings used on Windows for CJK languages. However, there was no way of indicating which encoding had been used, even if this was known (and e.g. scan would not know the encoding of the file it was reading). This lead to packages with data in French encoded in Latin-1 in .rda files which could not be read in other locales (and they would be able to be displayed in a French UTF-8 locale, if not in non-UTF-8 Japanese locales).

R 2.5.0 introduced a limited means to indicate the encoding of a CHARSXP via two of the ‘general purpose’ bits which are used to declare the encoding to be either Latin-1 or UTF-8. (Note that it is possible for a character vector to contain elements in different encodings.) Both printing and plotting notice the declaration and convert the string to the current locale (possibly using <xx> to display in hexadecimal bytes that are not valid in the current locale). Many (but not all) of the character manipulation functions will either preserve the declaration or re-encode the character string.

Strings that refer to the OS such as file names need to be passed through a wide-character interface on some OSes (e.g. Windows), which is to a large extent done as from R 2.7.0.

When are character strings declared to be of known encoding? One way is to do so directly via Encoding. The parser declares the encoding if this is known, either via the encoding argument to parse or from the locale within which parsing is being done at the R command line. (Other ways are recorded on the help page for Encoding.)

It is not necessary to declare the encoding of ASCII strings as they will work in any locale. ASCII strings should never have a marked encoding, as any encoding will be ignored when entering such strings into the CHARSXP cache.

The rationale behind considering only UTF-8 and Latin-1 is that most systems are capable of producing UTF-8 strings and this is the nearest we have to a universal format. For those that do not (for example those lacking a powerful enough iconv), it is likely that they work in Latin-1, the old R assumption.

This was taken further in R 2.7.0. There the parser can return a UTF-8-encoded string if it encounters a ‘\uxxx’ escape for a Unicode point that cannot be represented in the current charset. (This needs MBCS support, and was only enabled⁶ on Windows.) From R 2.10.0 it is enabled for all platforms with MBCS support, and a ‘\uxxx’ or ‘\Uxxxxxxxx’ escape ensures that the parsed string will be marked as UTF-8.

Most of the character manipulation functions now preserve UTF-8 encodings: there are some notes as to which at the top of file src/main/character.c and in file src/library/base/man/Encoding.Rd.

Graphics devices are offered the possibility of handing UTF-8-encoded strings without re-encoding to the native character set, by setting hasTextUTF8 to be ‘TRUE’ and supplying functions textUTF8 and strWidthUTF8 that expect UTF-8-encoded inputs. Normally the symbol font is encoded in Adobe Symbol encoding, but that can be re-encoded to UTF-8 by setting wantSymbolUTF8 to ‘TRUE’. The Windows’ port of cairographics has a rather peculiar assumption: it wants the symbol font to be encoded in UTF-8 as if it were encoded in Latin-1 rather than Adobe Symbol: this is selected by wantSymbolUTF8 = NA_LOGICAL (as from R 2.13.1).

Windows has no UTF-8 locales, but rather expects to work with UCS-2⁷ strings. R (being written in standard C) would not work internally with UCS-2 without extensive changes. The Rgui console⁸ uses UCS-2 internally, but communicates with the R engine in the native encoding. To allow UTF-8 strings to be printed in UTF-8 in Rgui.exe, an escape convention is used (see header file rgui_UTF8.h) which is used by cat, print and autoprinting.

‘Unicode’ (UCS-2LE) files are common in the Windows world, and readLines and scan will read them into UTF-8 strings on Windows if the encoding is declared explicitly on an unopened connection passed to those functions.

1.15 The CHARSXP cache

A global cache for CHARSXPs created by mkChar was introduced in R 2.6.0 – the cache ensures that most CHARSXPs with the same contents share storage (‘contents’ including any declared encoding). Not all CHARSXPs are part of the cache – notably ‘NA_STRING’ is not. CHARSXPs reloaded from the save formats of R prior to 0.99.0 are not cached (since the code used is frozen and few examples still exist).

The cache records the encoding of the string as well as the bytes: all requests to create a CHARSXP should be via a call to mkCharLenCE. Any encoding given in mkCharLenCE call will be ignored if the string’s bytes are all ASCII characters.

1.16 Warnings and errors

Each of warning and stop have two C-level equivalents, warning, warningcall, error and errorcall. The relationship between the pairs is similar: warning tries to fathom out a suitable call, and then calls warningcall with that call as the first argument if it succeeds, and with call = R_NilValue it is does not. When warningcall is called, it includes the deparsed call in its printout unless call = R_NilValue.

warning and error look at the context stack. If the topmost context is not of type CTXT_BUILTIN, it is used to provide the call, otherwise the next context provides the call. This means that when these functions are called from a primitive or .Internal, the imputed call will not be to primitive/.Internal but to the function calling the primitive/.Internal . This is exactly what one wants for a .Internal, as this will give the call to the closure wrapper. (Further, for a .Internal, the call is the argument to .Internal, and so may not correspond to any R function.) However, it is unlikely to be what is needed for a primitive.

The upshot is that that warningcall and errorcall should normally be used for code called from a primitive, and warning and error should be used for code called from a .Internal (and necessarily from .Call, .C and so on, where the call is not passed down). However, there are two complications. One is that code might be called from either a primitive or a .Internal, in which case probably warningcall is more appropriate. The other involves replacement functions, where the call will be of the form (from R < 2.6.0)

> length(x) <- y ~ x
Error in "length<-"(`*tmp*`, value = y ~ x) : invalid value

which is unpalatable to the end user. For replacement functions there will be a suitable context at the top of the stack, so warning should be used. (The results for .Internal replacement functions such as substr<- are not ideal.)

1.17 S4 objects

[This section is currently a preliminary draft and should not be taken as definitive. The description assumes that R_NO_METHODS_TABLES has not been set.]

1.17.1 Representation of S4 objects

S4 objects can be of any SEXPTYPE. They are either an object of a simple type (such as an atomic vector or function) with S4 class information or of type S4SXP. In all cases, the ‘S4 bit’ (bit 4 of the ‘general purpose’ field) is set, and can be tested by the macro/function IS_S4_OBJECT. Note that some code tests IS_S4_OBJECT without first checking that it is an object of any sort, so the ‘S4 bit’ is reserved even for things like primitive functions.

S4 objects are created via new()⁹ and thence via the C function R_do_new_object. This duplicates the prototype of the class, adds a class attribute and sets the S4 bit. All S4 class attributes should be character vectors of length one with an attribute giving (as a character string) the name of the package (or .GlobalEnv) containing the class definition. Since S4 objects have a class attribute, the OBJECT bit is set.

It is currently unclear what should happen if the class attribute is removed from an S4 object, or if this should be allowed.

1.17.2 S4 classes

S4 classes are stored as R objects in the environment in which they are created, with names .__C__classname: as such they are not listed by default by ls.

The objects are S4 objects of class "classRepresentation" which is defined in the methods package.

Since these are just objects, they are subject to the normal scoping rules and can be imported and exported from namespaces like other objects. The directives importClassesFrom and exportClasses are merely convenient ways to refer to class objects without needing to know their internal ‘metaname’ (although exportClasses does a little sanity checking via isClass).

1.17.3 S4 methods

Details of methods are stored in S4 objects of class "MethodsList". They have a non-syntactic name of the form .__M__generic:package for all methods defined in the current environment for the named generic derived from a specific package (which might be .GlobalEnv).

There is also environment .__T__generic:package which has names the signatures of the methods defined, and values the corresponding method functions. This is often referred to as a ‘methods table’.

When a package without a namespace is attached these objects become visible on the search path. library calls methods:::cacheMetaData to update the internal tables.

During an R session there is an environment associated with each non-primitive generic containing objects .AllMTable, .Generic, .Methods, .MTable, .SigArgs and .SigLength. .MTable and AllMTable are merged methods tables containing all the methods defined directly and via inheritance respectively. .Methods is a merged methods list.

Exporting methods from a namespace is more complicated than exporting a class. Note first that you do not export a method, but rather the directive exportMethods will export all the methods defined in the namespace for a specified generic: the code also adds to the list of generics any that are exported directly. For generics which are listed via exportMethods or exported themselves, the corresponding "MethodsList" and environment are exported and so will appear (as hidden objects) in the package environment.

Methods for primitives which are internally S4 generic (see below) are always exported, whether mentioned in the NAMESPACE file or not.

Methods can be imported either via the directive importMethodsFrom or via importing a namespace by import. Also, if a generic is imported via importFrom, its methods are also imported. In all cases the generic will be imported if it is in the namespace, so importMethodsFrom is most appropriate for methods defined on generics in other packages. Since methods for a generic could be imported from several different packages, the methods tables are merged.

When a package with a namespace is attached methods:::cacheMetaData is called to update the internal tables: only the visible methods will be cached.

1.17.4 Mechanics of S4 dispatch

This subsection does not discuss how S4 methods are chosen: see http://developer.r-project.org/howMethodsWork.pdf.

For all but primitive functions, setting a method on an existing function that is not itself S4 generic creates a new object in the current environment which is a call to standardGeneric with the old definition as the default method. Such S4 generics can also be created via a call to setGeneric¹⁰ and are standard closures in the R language, with environment the environment within which they are created. With the advent of namespaces this is somewhat problematic: if myfn was previously in a package with a name space there will be two functions called myfn on the search paths, and which will be called depends on which search path is in use. This is starkest for functions in the base namespace, where the original will be found ahead of the newly created function from any other package with a namespace.

Primitive functions are treated quite differently, for efficiency reasons: this results in different semantics. setGeneric is disallowed for primitive functions. The methods namespace contains a list .BasicFunsList named by primitive functions: the entries are either FALSE or a standard S4 generic showing the effective definition. When setMethod (or setReplaceMethod) is called, it either fails (if the list entry is FALSE) or a method is set on the effective generic given in the list.

Actual dispatch of S4 methods for almost all primitives piggy-backs on the S3 dispatch mechanism, so S4 methods can only be dispatched for primitives which are internally S3 generic. When a primitive that is internally S3 generic is called with a first argument which is an S4 object and S4 dispatch is on (that is, the methods namespace is loaded), DispatchOrEval calls R_possible_dispatch (defined in file src/main/objects.c). (Members of the S3 group generics, which includes all the generic operators, are treated slightly differently: the first two arguments are checked and DispatchGroup is called.) R_possible_dispatch first checks an internal table to see if any S4 methods are set for that generic (and S4 dispatch is currently enabled for that generic), and if so proceeds to S4 dispatch using methods stored in another internal table. All primitives are in the base namespace, and this mechanism means that S4 methods can be set for (some) primitives and will always be used, in contrast to setting methods on non-primitives.

The exception is %*%, which is S4 generic but not S3 generic as its C code contains a direct call to R_possible_dispatch.

The primitive as.double is special, as as.numeric and as.real are copies of it. The methods package code partly refers to generics by name and partly by function, and maps as.double and as.real to as.numeric (since that is the name used by packages exporting methods for it).

Some elements of the language are implemented as primitives, for example }. This includes the subset and subassignment ‘functions’ and they are S4 generic, again piggybacking on S3 dispatch.

.BasicFunsList is generated when methods is installed, by computing all primitives, initially disallowing methods on all and then setting generics for members of .GenericArgsEnv, the S4 group generics and a short exceptions list in file BasicFunsList.R: this currently contains the subsetting and subassignment operators and an override for c.

1.18 Memory allocators

R’s memory allocation is almost all done via routines in file src/main/memory.c. It is important to keep track of where memory is allocated, as the Windows port (by default) makes use of a memory allocator that differs from malloc etc as provided by MinGW. Specifically, there are entry points dlmalloc, dlfree, dlcalloc and dlfree provided by file src/extra/dlmalloc/malloc.c. This was done for two reasons. The primary motivation was performance: the allocator provided by MSVCRT via MinGW was far too slow at handling the many small allocations that the allocation system for SEXPRECs uses. As a side benefit, we can set a limit on the amount of allocated memory: this is useful as whereas Windows does provide virtual memory it is relatively far slower than many other R platforms and so limiting R’s use of swapping is highly advantageous. The high-performance allocator is only called from src/main/memory.c and src/extra/xdr: note that this means that it is not used in packages.

The rest of R should where possible make use of the allocators made available by file src/main/memory.c, which are also the methods recommended in Memory allocation in Writing R Extensions for use in R packages, namely the use of R_alloc, Calloc, Realloc and Free. Memory allocated by R_alloc is freed by the garbage collector once the ‘watermark’ has been reset by calling vmaxset. This is done automatically by the wrapper code calling primitives and .Internal functions (and also by the wrapper code to .Call and .External), but vmaxget and vmaxset can be used to reset the watermark from within internal code if the memory is only required for a short time.

All of the methods of memory allocation mentioned so far are relatively expensive. All R platforms support alloca, and in almost all cases¹¹ this is managed by the compiler, allocates memory on the C stack and is very efficient.

There are two disadvantages in using alloca. First, it is fragile and care is needed to avoid writing (or even reading) outside the bounds of the allocation block returned. Second, it increases the danger of overflowing the C stack. It is suggested that it is only used for smallish allocations (up to tens of thousands of bytes), and that

    R_CheckStack();

is called immediately after the allocation (as R’s stack checking mechanism will warn far enough from the stack limit to allow for modest use of alloca). (do_makeunique in file src/main/unique.c provides an example of both points.)

An alternative strategy has been used for various functions which require intermediate blocks of storage of varying but usually small size, and this has been consolidated into the routines in the header file src/main/RBufferUtils.h. This uses a structure which contains a buffer, the current size and the default size. A call to

    R_AllocStringBuffer(size_t blen, R_StringBuffer *buf);

sets buf->data to a memory area of at least blen+1 bytes. At least the default size is used, which means that for small allocations the same buffer can be reused. A call to R_FreeStringBufferL releases memory if more than the default has been allocated whereas a call to R_FreeStringBuffer frees any memory allocated.

The R_StringBuffer structure needs to be initialized, for example by

static R_StringBuffer ex_buff = {NULL, 0, MAXELTSIZE};

which uses a default size of MAXELTSIZE = 8192 bytes. Most current uses have a static R_StringBuffer structure, which allows the (default-sized) buffer to be shared between calls to e.g. grep and even between functions: this will need to be changed if R ever allows concurrent evaluation threads. So the idiom is

static R_StringBuffer ex_buff = {NULL, 0, MAXELTSIZE};
...
    char *buf;
    for(i = 0; i < n; i++) {
        compute len
        buf = R_AllocStringBuffer(len, &ex_buff);
        use buf
    }
    /*  free allocation if larger than the default, but leave
        default allocated for future use */
   R_FreeStringBufferL(&ex_buff);

1.18.1 Internals of R_alloc

The memory used by R_alloc is allocated as part of the same heap as R objects, and recovered by the same garbage collector, but it is never used as an R object, and may not have all (or any) of the header fields that an R object would have. For historical reasons, at least one more byte is allocated than was asked for, but this should not be relied on. The actual amount allocated will be rounded up to a whole number of 16-byte chunks, perhaps after adding some header overhead.

The vectors allocated are protected via the setting of R_VStack, as the garbage collector marks everything that can be reached from that location. When a vector is R_allocated, its ATTRIB pointer is set to the current R_VStack, and R_VStack is set to the latest allocation. Thus R_VStack is a single-linked chain of vectors currently allocated via R_alloc. Function vmaxset resets the location R_VStack, and should be to a value that has previously be obtained via vmaxget: allocations after the value was obtained will no longer be protected and hence available for garbage collection.

1.19 Internal use of global and base environments

This section notes known use by the system of these environments: the intention is to minimize or eliminate such uses.

1.19.1 Base environment

The graphics devices system maintains two variables .Device and .Devices in the base environment: both are always set. The variable .Devices gives a list of character vectors of the names of open devices, and .Device is the element corresponding to the currently active device. The null device will always be open.

There appears to be a variable .Options, a pairlist giving the current options settings. But in fact this is just a symbol with a value assigned, and so shows up as a base variable.

Similarly, the evaluator creates a symbol .Last.value which appears as a variable in the base environment.

Errors can give rise to objects .Traceback and last.warning in the base environment.

1.19.2 Global environment

The seed for the random number generator is stored in object .Random.seed in the global environment.

Some error handlers may give rise to objects in the global environment: for example dump.frames by default produces last.dump.

The windows() device makes use of a variable .SavedPlots to store display lists of saved plots for later display. This is regarded as a variable created by the user.

1.20 Modules

R makes use of a number of shared objects/DLLs stored in the modules directory. These are parts of the code which have been chosen to be loaded ‘on demand’ rather than linked as dynamic libraries or incorporated into the main executable/dynamic library.

For a few of these (e.g. vfonts) the issue is size: the database for the Hershey fonts is included in the C code of the module and was at one time an appreciable part of the codebase for a rarely used feature. However, for most of the modules the motivation has been the amount of (often optional) code they will bring in via libraries to which they are linked.

internet

The internal HTTP and FTP clients and socket support, which link to system-specific support libraries.

lapack

The code which makes use of the LAPACK library, and is linked to libRlapack or an external LAPACK library.

vfonts

The Hershey font databases and the code to draw with them.

X11

(Unix-alikes only.) The X11(), jpeg(), png() and tiff() devices. These are optional, and links to some or all of the X11, pango, cairo, jpeg, libpng and libtiff libraries.

internet2.dll

(Windows only.) An alternative version of the internet access routines, compiled against Internet Explorer internals (and so loads wininet.dll and wsock32.dll).

1.21 Visibility

1.21.1 Hiding C entry points

We make use of the visibility mechanisms discussed in Controlling visibility in Writing R Extensions, C entry points not needed outside the main R executable/dynamic library (and in particular in no package nor module) should be prefixed by attribute_hidden. Minimizing the visibility of symbols in the R dynamic library will speed up linking to it (which packages will do) and reduce the possibility of linking to the wrong entry points of the same name. In addition, on some platforms reducing the number of entry points allows more efficient versions of PIC to be used: somewhat over half the entry points are hidden. A convenient way to hide variables (as distinct from functions) is to declare them extern0 in header file Defn.h.

The visibility mechanism used is only available with some compilers and platforms, and in particular not on Windows, where an alternative mechanism is used. Entry points will not be made available in R.dll if they are listed in the file src/gnuwin32/Rdll.hide. Entries in that file start with a space and must be strictly in alphabetic order in the C locale (use sort on the file to ensure this if you change it). It is possible to hide Fortran as well as C entry points via this file: the former are lower-cased and have an underline as suffix, and the suffixed name should be included in the file. Some entry points exist only on Windows or need to be visible only on Windows, and some notes on these are provided in file src/gnuwin32/Maintainters.notes.

Because of the advantages of reducing the number of visible entry points, they should be declared attribute_hidden where possible. Note that this only has an effect on a shared-R-library build, and so care is needed not to hide entry points that are legitimately used by packages. So it is best if the decision on visibility is made when a new entry point is created, including the decision if it should be included in header file Rinternals.h. A list of the visible entry points on shared-R-library build on a reasonably standard Unix-alike can be made by something like

nm -g libR.so | grep ' [BCDT] ' | cut -b20-

1.21.2 Variables in Windows DLLs

Windows is unique in that it conventionally treats importing variables differently from functions: variables that are imported from a DLL need to be specified by a prefix (often ‘_imp_’) when being linked to (‘imported’) but not when being linked from (‘exported’). The details depend on the compiler system, and have changed for MinGW during the lifetime of that port. They are in the main hidden behind some macros defined in header file R_ext/libextern.h.

A (non-function) variable in the main R sources that needs to be referred to outside R.dll (in a package, module or another DLL such as Rgraphapp.dll) should be declared with prefix LibExtern. The main use is in Rinternals.h, but it needs to be considered for any public header and also Defn.h.

It would nowadays be possible to make use of the ‘auto-import’ feature of the MinGW port of ld to fix up imports from DLLs (and if R is built for the Cygwin platform this is what happens). However, this was not possible when the MinGW build of R was first constructed in ca 1998, allows less control of visibility and would not work for other Windows compiler suites.

It is only possible to check if this has been handled correctly by compiling the R sources on Windows.

1.22 Lazy loading

Lazy loading was introduced in R 2.0.0, for code in packages and for datasets in packages (the latter is still optional, but code is always lazy-loaded as from R 2.14.0). When a package/namespace which uses it is loaded, the package/namespace environment is populated with promises for all the named objects: when these promises are evaluated they load the actual code from a database.

There are separate databases for code and data, stored in the R and data subdirectories. The database consists of two files, name.rdb and name.rdx. The .rdb file is a concatenation of serialized objects, and the .rdx file contains an index. The objects are stored in (usually) a gzip-compressed format with a 4-byte header giving the uncompressed serialized length (in XDR, that is big-endian, byte order) and read by a call to the primitive lazyLoadDBfetch. (Note that this makes lazy-loading unsuitable for really large objects: the serialized length of an R object can exceed 4GB.)

The index or ‘map’ file name.rdx is a compressed serialized R object to be read by readRDS. It is a list with three elements variables, references and compressed. The first two are named lists of integer vectors of length 2 giving the offset and length of the serialized object in the name.rdb file. Element variables has an entry for each named object: references serializes a temporary environment used when named environments are added to the database. compressed is a logical indicating if the serialized objects were compressed: compression is always used nowadays. R 2.10.0 added the values compressed = 2 and 3 for bzip2 and xz compression (with the possibility of future expansion to other methods): these formats add a fifth byte to the header for the type of compression, and stores serialized objects uncompressed if compression expands them.

The loader for a lazy-load database of code or data is function lazyLoad in the base package, but note that there is a separate copy to load base itself in file R_HOME/base/R/base.

Lazy-load databases are created by the code in src/library/tools/R/makeLazyLoad.R: the main tool is the unexported function makeLazyLoadDB and the insertion of database entries is done by calls to .Call("R_lazyLoadDBinsertValue", ...).

Lazy-load databases of less than 10MB are cached in memory at first use: this was found necessary when using file systems with high latency (removable devices and network-mounted file systems on Windows).

The same database mechanism is used to store parsed Rd files. One or all of the parsed objects is fetched by a call to tools:::fetchRdDB.

1.23 Helper threads and task merging

The pqR facility for doing computations in helper threads and/or deferred evaluation for task merging is supported by the “helpers” library in src/extra/helpers, where documentation on it may be found. The helpers-app.h and helpers-app.c files in that directory define how the helpers facility interfaces to the pqR interpreter.

At present, computation in helper threads (or in other deferred tasks) is confined to vectors, which must be numeric vectors unless the task is scheduled with the "master-only" flag (which makes it doable only by the master thread).

Two bits in the headers of such objects (actually all objects) allow checks on how computations that may be done by helper threads (or that otherwise are deferred) are progressing. These bits are accessed by the helpers_is_being_computed and helpers_is_in_use macros. An object is being computed by a task if computation of its data part was scheduled to be done by a task that has not been completed (or perhaps not even started). An object is in use by a task if it is an input to a scheduled task for which it is not also the output. (However, a object with maximum NAMEDCNT may not be flagged as in use even if it is, since such an object will never be modified anyway.) Note that these bits are set only by the master thread (whenever the master thread looks to see what tasks performed by helpers have completed), so access to them does not raise any synchronization issues.

The WAIT_UNTIL_COMPUTED and WAIT_UNTIL_COMPUTED_2 macros wait until an object is not being computed, or until two objects are both not being computed.

Objects that may not have been computed yet should be visible only in code that has explicitly asked to see such “pending” objects. For example, the result of calling eval will never be an object that is still being computed. Code that wishes to receive a pending object as the result of evaluating an expression must instead call evalv, passing a variant argument with the VARIANT_PENDING_OK bit set. Similarly, PRVALUE will never return a pending object, but PRVALUE_PENDING_OK may.

Objects that are being used by a task that has not completed may be seen anywhere within the code for the interpeter, or in user-written C code. Data in such objects may be accessed without regard to their possibly being used by a helper thread, but this data must not be changed. This is ensured by making NAMEDCNT (and hence also NAMED) check whether the object is in use, and if so wait until it is not in use before returning the appropriate value (unless it is MAX_NAMEDCNT) — in effect, NAMEDCNT is temporarily increased while the object is in use by a task. Correct code that only alters objects after verifying that NAMEDCNT is zero will therefore work properly with objects that must not be altered until a task (or tasks) has completed. However, the macro NAMEDCNT_GT_0 is written to return TRUE if NAMEDCNT is greater than zero, without waiting until the object is not in use; similarly, NAMEDCNT_EQ_0 and NAMEDCNT_GT_1 return immediately when the comparison does not depend on whether the object is in use by a task.

To avoid problems with the use of multiple threads by the helpers facility, the Unix/Linux fork function, used by package parallel and others, is redefined to Rf_fork in Rinternals.h. The Rf_fork function waits for all helper threads to become idle before forking, and then disables use of helper threads in the child process.

2 .Internal vs .Primitive

C code compiled into R at build time can be called directly in what are termed primitives or via the .Internal interface, which is very similar to the .External interface except in syntax. More precisely, R maintains a table of R function names and corresponding C functions to call, which by convention all start with ‘do_’ and return a SEXP. This table (R_FunTab in file src/main/names.c) also specifies how many arguments to a function are required or allowed, whether or not the arguments are to be evaluated before calling, and whether the function is ‘internal’ in the sense that it must be accessed via the .Internal interface, or directly accessible in which case it is printed in R as .Primitive.

Functions using .Internal() wrapped in a closure are in general preferred as this ensures standard handling of named and default arguments. For example, axis is defined as

axis <- function(side, at = NULL, labels = NULL, ...)
    .Internal(axis(side, at, labels, ...))

However, for reasons of convenience and also efficiency (as there is some overhead in using the .Internal interface wrapped in a function closure), the primitive functions are exceptions that can be accessed directly. And of course, primitive functions are needed for basic operations—for example .Internal is itself a primitive. Note that primitive functions make no use of R code, and hence are very different from the usual interpreted functions. In particular, formals and body return R_NilValue for such objects, and argument matching can be handled differently. For some primitives (including call, switch, .C and .subset) positional matching is important to avoid partial matching of the first argument.

The list of primitive functions is subject to change; currently, it includes the following.

1. “Special functions” which really are language elements, but implemented as primitive functions:

   {       (         if     for      while  repeat  break  next
   return  function  quote  switch
   

2. Language elements and basic operators (i.e., functions usually not called as foo(a, b, ...)) for subsetting, assignment, arithmetic and logic (but @<- is currently not primitive):

   [    [[   $    @     [<-  [[<-  $<-
   <-   <<-  =    ->    ->>
   
   +    -    *    /     ^    %%   %*%  %/%
   <    <=   ==   !=    >=   >
   |    ||   &    &&    !
   

3. “Low level” 0– and 1–argument functions which belong to one of the following groups of functions:
   1. Basic mathematical functions with a single argument, i.e.,

      abs     sign    sqrt
      floor   ceiling
      

      exp     expm1
      log2    log10   log1p
      cos     sin     tan
      acos    asin    atan
      cosh    sinh    tanh
      acosh   asinh   atanh
      

      gamma   lgamma  digamma trigamma
      

      cumsum  cumprod cummax  cummin
      

      Im  Re  Arg  Conj  Mod
      

      log is a primitive function of one or two arguments with named argument matching.

      trunc is a difficult case: it is a primitive that can have one or more arguments: the default method handled in the primitive has only one.

   2. Functions rarely used outside of “programming” (i.e., mostly used inside other functions), such as

      nargs          missing        on.exit        interactive
      as.call        as.character   as.complex     as.double
      as.environment as.integer     as.logical     as.raw
      is.array       is.atomic      is.call        is.character  
      is.complex     is.double      is.environment is.expression 
      is.finite      is.function    is.infinite    is.integer    
      is.language    is.list        is.logical     is.matrix     
      is.na          is.name        is.nan         is.null       
      is.numeric     is.object      is.pairlist    is.raw        
      is.real        is.recursive   is.single      is.symbol     
      baseenv        emptyenv       globalenv      pos.to.env
      unclass        invisible      seq_along      seq_len
      

   3. The programming and session management utilities

      browser  proc.time  gc.time  pnamedcnt
      

4. The following basic replacement and extractor functions

   length      length<-
   class       class<-
   oldClass    oldCLass<-
   attr        attr<-
   attributes  attributes<-
   names       names<-
   dim         dim<-
   dimnames    dimnames<-
               environment<-
               levels<-
               storage.mode<-
   

   Note that optimizing NAMED = 1 is only effective within a primitive (as the closure wrapper of a .Internal will set NAMED = 2 when the promise to the argument is evaluated) and hence replacement functions should where possible be primitive to avoid copying (at least in their default methods).

5. The following functions are primitive for efficiency reasons:

   :          ~          c           list
   call       expression substitute
   UseMethod  standardGeneric
   .C         .Fortran   .Call       .External
   round      signif      rep        seq.int
   

   as well as the following internal-use-only functions

   .Primitive     .Internal
   .Call.graphics .External.graphics
   .subset        .subset2
   .primTrace     .primUntrace
   lazyLoadDBfetch
   

The multi-argument primitives

call       switch
.C         .Fortran   .Call       .External

intentionally use positional matching, and need to do so to avoid partial matching to their first argument. They do check that the first argument (partially) matched the formal argument name. On the other hand,

attr       attr<-     browser     rememtrace substitute  UseMethod
log        round      signif      rep        seq.int

manage their own argument matching and do work in the standard way.

All the one-argument primitives check that if they are called with a named argument that this (partially) matches the name given in the documentation: this is also done for replacement functions with one argument plus value.

The net effect is that from R 2.11.0 argument matching for primitives intended for end-user use is done in the same way as for interpreted functions except for the six exceptions where positional matching is required.

2.1 Special primitives

A small number of primitives are specials rather than builtins, that is they are entered with unevaluated arguments. This is clearly necessary for the language constructs and the assignment operators, as well as for && and || which conditionally evaluate their second argument, and ~, .Internal, call, expression, missing, on.exit, quote and substitute which do not evaluate some of their arguments.

rep is special as it evaluates some of its arguments conditional on which are non-missing.

log, round and signif are special to allow default values to be given to missing arguments.

The subsetting, subassignment and @ operators are all special. (For both extraction and replacement forms, $ and @ take a symbol argument, and [ and [[ allow missing arguments.)

UseMethod is special to avoid the additional contexts added to calls to builtins.

2.2 Special internals

There are also special .Internal functions: NextMethod, Recall, withVisible, cbind, rbind (to allow for the deparse.level argument), eapply, lapply and vapply.

2.3 Prototypes for primitives

Prototypes are available for the primitive functions and operators, and these are used for printing, args and package checking (e.g. by tools::checkS3methods and by package codetools). There are two environments in the base package (and namespace), ‘.GenericArgsEnv’ for those primitives which are internal S3 generics, and ‘.ArgsEnv’ for the rest. Those environments contain closures with the same names as the primitives, formal arguments derived (manually) from the help pages, a body which is a suitable call to UseMethod or R_NilValue and environment the base namespace.

The C code for print.default and args uses the closures in these environments in preference to the definitions in base (as primitives).

The QC function undoc checks that all the functions prototyped in these environments are currently primitive, and that the primitives not included are better thought of as language elements (at the time of writing

$  $<-  &&  (  :  @  [  [[  [[<-  [<-  {  ||  ~  <-  <<-  =  ->   ->>
break  for function  if  next  repeat  return  while

). One could argue about ~, but it is known to the parser and has semantics quite unlike a normal function. And : is documented with different argument names in its two meanings.)

The QC functions codoc and checkS3methods also make use of these environments (effectively placing them in front of base in the search path), and hence the formals of the functions they contain are checked against the help pages by codoc. However, there are two problems with the generic primitives. The first is that many of the operators are part of the S3 group generic Ops and that defines their arguments to be e1 and e2: although it would be very unusual, an operator could be called as e.g. "+"(e1=a, e2=b) and if method dispatch occurred to a closure, there would be an argument name mismatch. So the definitions in environment .GenericArgsEnv have to use argument names e1 and e2 even though the traditional documentation is in terms of x and y: codoc makes the appropriate adjustment via tools:::.make_S3_primitive_generic_env. The second discrepancy is with the Math group generics, where the group generic is defined with argument list (x, ...), but most of the members only allow one argument when used as the default method (and round and signif allow two as default methods): again fix-ups are used.

Those primitives which are in .GenericArgsEnv are checked (via tests/primitives.R) to be generic via defining methods for them, and a check is made that the remaining primitives are probably not generic, by setting a method and checking it is not dispatched to (but this can fail for other reasons). However, there is no certain way to know that if other .Internal or primitive functions are not internally generic except by reading the source code.

3 Internationalization in the R sources

The process of marking messages (errors, warnings etc) for translation in an R package is described in Internationalization in Writing R Extensions, and the standard packages included with R have (with an exception in grDevices for the menus of the windows() device) been internationalized in the same way as other packages.

3.1 R code

Internationalization for R code is done in exactly the same way as for extension packages. As all standard packages which have R code also have a namespace, it is never necessary to specify domain, but for efficiency calls to message, warning and stop should include domain = NA when the message is constructed via gettextf, gettext or ngettext.

For each package, the extracted messages and translation sources are stored under package directory po in the source package, and compiled translations under inst/po for installation to package directory po in the installed package. This also applies to C code in packages.

3.2 Main C code

The main C code (e.g. that in files src/*/*.c and in the modules) is where R is closest to the sort of application for which ‘gettext’ was written. Messages in the main C code are in domain R and stored in the top-level directory po with compiled translations under share/locale.

The list of files covered by the R domain is specified in file po/POTFILES.in.

The normal way to mark messages for translation is via _("msg") just as for packages. However, sometimes one needs to mark passages for translation without wanting them translated at the time, for example when declaring string constants. This is the purpose of the N_ macro, for example

{ ERROR_ARGTYPE,           N_("invalid argument type")},

from file src/main/errors.c.

The P_ macro

#ifdef ENABLE_NLS
#define P_(StringS, StringP, N) ngettext (StringS, StringP, N)
#else
#define P_(StringS, StringP, N) (N > 1 ? StringP: StringS)
#endif

may be used as a wrapper for ngettext: however in some cases the preferred approach has been to conditionalize (on ENABLE_NLS) code using ngettext.

The macro _("msg") can safely be used in directory src/appl; the header for standalone ‘nmath’ skips possible translation. (This does not apply to N_ or P_).

3.3 Windows-GUI-specific code

Messages for the Windows GUI are in a separate domain ‘RGui’. This was done for two reasons:

• The translators for the Windows version of R might be separate from those for the rest of R (familiarity with the GUI helps), and
• Messages for Windows are most naturally handled in the native charset for the language, and in the case of CJK languages the charset is Windows-specific. (It transpires that as the iconv we ported works well under Windows, this is less important than anticipated.)

Messages for the ‘RGui’ domain are marked by G_("msg"), a macro that is defined in header file src/gnuwin32/win-nls.h. The list of files that are considered is hardcoded in the RGui.pot-update target of file po/Makefile.in.in: note that this includes devWindows.c as the menus on the windows device are considered to be part of the GUI. (There is also GN_("msg"), the analogue of N_("msg").)

The template and message catalogs for the ‘RGui’ domain are in the top-level po directory.

3.4 Mac OS X GUI

This is handled separately: see http://developer.r-project.org/Translations.html.

3.5 Updating

See file po/README for how to update the message templates and catalogs.

4 Structure of an Installed Package

The structure of a source packages is described in Creating R packages in Writing R Extensions: this chapter is concerned with the structure of installed packages.

An installed package has a top-level file DESCRIPTION, a copy of the file of that name in the package sources with a ‘Built’ field appended, and file INDEX, usually describing the objects on which help is available, a file NAMESPACE if the package has a name space, optional files such as CITATION, LICENCE and NEWS, and any other files copied in from inst. It will have directories Meta, help and html (even if the package has no help pages), almost always has a directory R and often has a directory libs to contain compiled code. Other directories with known meaning to R are data, demo, doc and po.

Function library looks for a namespace and if one is found passes control to loadNamespace. Then library or loadNamespace looks for file R/pkgname, warns if it is not found and otherwise sources the code (using sys.source) into the package’s environment, then lazy-loads a database R/sysdata if present. So how R code gets loaded depends on the contents of R/pkgname: a standard template to load lazy-load databases are provided in share/R/nspackloader.R.

How (and if) compiled code is loaded is down to the package’s startup code such as .First.lib or .onLoad, although a useDynlib directive in a namespace provides an alternative. Conventionally compiled code is loaded by a call to library.dynam and this looks in directory libs (and in an appropriate sub-directory if sub-architectures are in use) for a shared object (Unix-alike) or DLL (Windows).

Subdirectory data serves two purposes. In a package using lazy-loading of data, it contains a lazy-load database Rdata, plus a file Rdata.rds which contain a named character vector used by data() in the (unusual) event that it is used for such a package. Otherwise it is a copy of the data directory in the sources, with saved images re-compressed if R CMD INSTALL --resave-data was used. Prior to R 2.12.0 the contents of the directory could be moved to a zip file Rdata.zip and a listing written in file filelist: this was principally done on Windows.

Subdirectory demo supports the demo function, and is copied from the sources.

Subdirectory po contains (in subdirectories) compiled message catalogs.

4.1 Metadata

Directory Meta contains several files in .rds format, that is serialized R objects written by saveRDS. All packages have files Rd.rds, hsearch.rds, links.rds and package.rds. Packages with namespaces have a file nsInfo.rds, and those with data, demos or vignettes have data.rds, demo.rds or vignette.rds files.

The structure of these files (and their existence and names) is private to R, so the description here is for those trying to follow the R sources: there should be no reference to these files in non-base packages.

File package.rds is a dump of information extracted from the DESCRIPTION file. It is a list of several components. The first, ‘DESCRIPTION’, is a character vector, the DESCRIPTION file as read by read.dcf. Further elements ‘Depends’, ‘Suggests’, ‘Imports’, ‘Rdepends’ and ‘Rdepends2’ record the ‘Depends’, ‘Suggests’ and ‘Imports’ fields. These are all lists, and can be empty. The first three have an entry for each package named, each entry being a list of length 1 or 3, which element ‘name’ (the package name) and optional elements ‘op’ (a character string) and ‘version’ (an object of class ‘"package_version"’). Element ‘Rdepends’ is used for the first version dependency on R, and ‘Rdepends2’ is a list of zero or more R version dependencies—each is a three-element list of the form described for packages. Element ‘Rdepends’ is no longer used, but it is still potentially needed so R < 2.7.0 can detect that the package was not installed for it.

File nsInfo.rds records a list, a parsed version of the NAMESPACE file.

File Rd.rds records a data frame with one row for each help file. The (character) columns are ‘File’ (the file name with extension), ‘Name’ (the ‘\name’ section), ‘Type’ (from the optional ‘\docType’ section), ‘Title’, ‘Encoding’, ‘Aliases’, ‘Concepts’ and ‘Keywords’. The last three are character strings with zero or more entries separated by ‘, ’.

File hsearch.rds records the information to be used by ‘help.search’. This is a list of four unnamed elements which are character vectors for help files, aliases, keywords and concepts. All the matrices have columns ‘ID’ and ‘Package’ which are used to tie the aliases, keywords and concepts (the remaining column of the last three elements) to a particular help file. The first element has further columns ‘LibPath’ (empty since R 2.3.0), ‘name’, ‘title’, ‘topic’ (the first alias, used when presenting the results as ‘pkgname::topic’) and ‘Encoding’.

File links.rds records a named character vector, the names being aliases and the values character strings of the form

"../../pkgname/html/filename.html"

File data.rds records a two-column character matrix with columns of dataset names and titles from the corresponding help file. File demo.rds has the same structure for package demos.

File vignette.rds records a dataframe with one row for each ‘vignette’ (.[RS]nw file in inst/doc) and with columns ‘File’ (the full file path in the sources), ‘Title’, ‘PDF’ (the pathless file name of the installed PDF version, if present), ‘Depends’, ‘Keywords’ and ‘R’ (the pathless file name of the installed R code, if present).

4.2 Help

All installed packages, whether they had any .Rd files or not, have help and html directories. The latter normally only contains the single file 00Index.html, the package index which has hyperlinks to the help topics (if any).

Directory help contains files AnIndex, paths.rds and pkgname.rd[bx]. The latter two files are a lazy-load database of parsed .Rd files, accessed by tools:::fetchRdDB. File paths.rds is a saved character vector of the original path names of the .Rd files, used when updating the database.

File AnIndex is a two-column tab-delimited file: the first column contains the aliases defined in the help files and the second the basename (without the .Rd or .rd extension) of the file containing that alias. It is read by utils:::index.search to search for files matching a topic (alias), and read by scan in utils:::matchAvailableTopics, part of the completion system.

File aliases.rds is the same information as AnIndex as a named character vector (names the topics, values the file basename), for faster access.

5 Files

R provides many functions to work with files and directories: many of these have been added relatively recently to facilitate scripting in R and in particular the replacement of Perl scripts by R scripts in the management of R itself.

These functions are implemented by standard C/POSIX library calls, except on Windows. That means that filenames must be encoded in the current locale as the OS provides no other means to access the file system: increasingly filenames are stored in UTF-8 and the OS will translate filenames to UTF-8 in other locales. So using a UTF-8 locale gives transparent access to the whole file system.

Windows is another story. There the internal view of filenames is in UTF-16LE (so-called ‘Unicode’), and standard C library calls can only access files whose names can be expressed in the current codepage. To circumvent that restriction, there is a parallel set of Windows-specific calls which take wide-character arguments for filepaths. Much of the file-handling in R has been moved over to using these functions, so filenames can be manipulated in R as UTF-8 encoded character strings, converted to wide characters (which on Windows are UTF-16LE) and passed to the OS. The utilities RC_fopen and filenameToWchar help this process. Currently file.copy to a directory, list.files, list.dirs and path.expand work only with filepaths encoded in the current codepage.

All these functions do tilde expansion, in the same way as path.expand, with the deliberate exception of Sys.glob.

File names may be case sensitive or not: the latter is the norm on Windows and Mac OS X, the former on other Unix-alikes. Note that this is a property of both the OS and the file system: it is often possible to map names to upper or lower case when mounting the file system. This can affect the matching of patterns in list.files and Sys.glob.

File names commonly contain spaces on Windows and Mac OS X but not elsewhere. As file names are handled as character strings by R, spaces are not usually a concern unless file names are passed to other process, e.g. by a system call.

Windows has another couple of peculiarities. Whereas a POSIX file system has a single root directory (and other physical file systems are mounted onto logical directories under that root), Windows has separate roots for each physical or logical file system (‘volume’), organized under drives (with file paths starting D: for an ASCII letter, case-insensitively) and network shares (with paths like \netname\topdir\myfiles\a file. There is a current drive, and path names without a drive part are relative to the current drive. Further, each drive has a current directory, and relative paths are relative to that current directory, on a particular drive if one is specified. So D:dir\file and D: are valid path specifications (the last being the current directory on drive D:).

6 Graphics

R’s graphics internals were revised for version 1.4.0 (and tidied up for 2.7.0). This was to enable multiple graphics systems to be installed on top on the graphics ‘engine’ – currently there are two such systems, one supporting ‘base’ graphics (based on that in S and whose R code¹² is in package graphics) and one implemented in package grid.

Some notes on the changes for 1.4.0 can be found at http://www.stat.auckland.ac.nz/~paul/R/basegraph.html and http://www.stat.auckland.ac.nz/~paul/R/graphicsChanges.html.

At the lowest level is a graphics device, which manages a plotting surface (a screen window or a representation to be written to a file). This implements a set of graphics primitives, to ‘draw’

• a circle, optionally filled
• a rectangle, optionally filled
• a line
• a set of connected lines
• a polygon, optionally filled
• a paths, optionally filled using a winding rule
• text
• a raster image (optional)
• and to set a clipping rectangle

as well as requests for information such as

• the width of a string if plotted
• the metrics (width, ascent, descent) of a single character
• the current size of the plotting surface

and requests/opportunities to take action such as

• start a new ‘page’, possibly after responding to a request to ask the user for confirmation.
• return the position of the device pointer (if any).
• when a device become the current device or stops being the current device (this is usually used to change the window title on a screen device).
• when drawing starts or finishes (e.g. used to flush graphics to the screen when drawing stops).
• wait for an event, for example a mouse click or keypress.
• an ‘onexit’ action, to clean up if plotting is interrupted (by an error or by the user).
• capture the current contents of the device as a raster image.
• close the device.

The device also sets a number of variables, mainly Boolean flags indicating its capabilities. Devices work entirely in ‘device units’ which are up to its developer: they can be in pixels, big points (1/72 inch), twips, …, and can differ¹³ in the ‘x’ and ‘y’ directions.

The next layer up is the graphics ‘engine’ that is the main interface to the device (although the graphics subsystems do talk directly to devices). This is responsible for clipping lines, rectangles and polygons, converting the pch values 0...26 to sets of lines/circles, centring (and otherwise adjusting) text, rendering mathematical expressions (‘plotmath’) and mapping colour descriptions such as names to the internal representation.

Another function of the engine is to manage display lists and snapshots. Some but not all instances of graphics devices maintain display lists, a ‘list’ of operations that have been performed on the device to produce the current plot (since the device was opened or the plot was last cleared, e.g. by plot.new). Screen devices generally maintain a display list to handle repaint and resize events whereas file-based formats do not—display lists are also used to implement dev.copy() and friends. The display list is a pairlist of .Internal (base graphics) or .Call.graphics (grid graphics) calls, which means that the C code implementing a graphics operation will be re-called when the display list is replayed: apart from the part which records the operation if successful.

Snapshots of the current graphics state are taken by GEcreateSnapshot and replayed later in the session by GEplaySnapshot. These are used by recordPlot(), replayPlot() and the GUI menus of the windows() device. The ‘state’ includes the display list.

The top layer comprises the graphics subsystems. Although there is provision for 24 subsystems, after 6 years only two exist, ‘base’ and ‘grid’. The base subsystem is registered with the engine when R is initialized, and unregistered (via KillAllDevices) when an R session is shut down. The grid subsystem is registered in its .onLoad function and unregistered in the .onUnload function. The graphics subsystem may also have ‘state’ information saved in a snapshot (currently base does and grid does not).

Package grDevices was originally created to contain the basic graphics devices (although X11 is in a separate load-on-demand module because of the volume of external libraries it brings in). Since then it has been used for other functionality that was thought desirable for use with grid, and hence has been transferred from package graphics to grDevices. This is principally concerned with the handling of colours and recording and replaying plots.

6.1 Graphics Devices

R ships with several graphics devices, and there is support for third-party packages to provide additional devices—several packages now do. This section describes the device internals from the viewpoint of a would-be writer of a graphics device.

6.1.1 Device structures

There are two types used internally which are pointers to structures related to graphics devices.

The DevDesc type is a structure defined in the header file R_ext/GraphicsDevice.h (which is included by R_ext/GraphicsEngine.h). This describes the physical characteristics of a device, the capabilities of the device driver and contains a set of callback functions that will be used by the graphics engine to obtain information about the device and initiate actions (e.g. a new page, plotting a line or some text). Type pDevDesc is a pointer to this type.

Prior to R 2.14.0 all of the callback functions need to be set, to dummy functions if the capability (such as a locator) is not available. In devices which will only be used in R 2.14.0 or later the following callbacks can be omitted (or set to the null pointer, their default value) when appropriate default behaviour will be taken by the graphics engine: activate, cap, deactivate, locator, holdflush (API version 9), mode, newFrameConfirm, path, raster and size.

The relationship of device units to physical dimensions is set by the element ipr of the DevDesc structure: a ‘double’ array of length 2.

The GEDevDesc type is a structure defined in R_ext/GraphicsEngine.h (with comments in the file) as

typedef struct _GEDevDesc GEDevDesc;
struct _GEDevDesc {
    pDevDesc dev;
    Rboolean displayListOn;
    SEXP displayList;
    SEXP DLlastElt;
    SEXP savedSnapshot;
    Rboolean dirty;
    Rboolean recordGraphics;
    GESystemDesc *gesd[MAX_GRAPHICS_SYSTEMS];
    Rboolean ask;
}

So this is essentially a device structure plus information about the device maintained by the graphics engine and normally¹⁴ visible to the engine and not to the device. Type pGEDevDesc is a pointer to this type.

The graphics engine maintains an array of devices, as pointers to GEDevDesc structures. The array is of size 64 but the first element is always occupied by the "null device" and the final element is kept as NULL as a sentinel.¹⁵ This array is reflected in the R variable ‘.Devices’. Once a device is killed its element becomes available for reallocation (and its name will appear as "" in ‘.Devices’). Exactly one of the devices is ‘active’: this is the the null device if no other device has been opened and not killed.

Each instance of a graphics device needs to set up a GEDevDesc structure by code very similar to

    pGEDevDesc gdd;

    R_GE_checkVersionOrDie(R_GE_version);
    R_CheckDeviceAvailable();
    BEGIN_SUSPEND_INTERRUPTS {
        pDevDesc dev;
        /* Allocate and initialize the device driver data */
        if (!(dev = (pDevDesc) calloc(1, sizeof(DevDesc))))
            return 0; /* or error() */
        /* set up device driver or free 'dev' and error() */
        gdd = GEcreateDevDesc(dev);
        GEaddDevice2(gdd, "dev_name");
    } END_SUSPEND_INTERRUPTS;

The DevDesc structure contains a void * pointer ‘deviceSpecific’ which is used to store data specific to the device. Setting up the device driver includes initializing all the non-zero elements of the DevDesc structure.

Note that the device structure is zeroed when allocated: this provides some protection against future expansion of the structure since the graphics engine can add elements that need to be non-NULL/non-zero to be ‘on’ (and the structure ends with 64 reserved bytes which will be zeroed and allow for future expansion).

Rather more protection is provided by the version number of the engine/device API, R_GE_version defined in R_ext/GraphicsEngine.h together with access functions

int R_GE_getVersion(void);
void R_GE_checkVersionOrDie(int version);

If a graphics device calls R_GE_checkVersionOrDie(R_GE_version) it can ensure it will only be used in versions of R which provide the API it was designed for and compiled against.

6.1.2 Device capabilities

The following ‘capabilities’ can be defined for the device’s DevDesc structure.

• canChangeGamma – Rboolean: can the display gamma be adjusted? This is now ignored, as gamma support has been removed.
• canHadj – integer: can the device do horizontal adjustment of text via the text callback, and if so, how precisely? 0 = no adjustment, 1 = {0, 0.5, 1} (left, centre, right justification) or 2 = continuously variable (in [0,1]) between left and right justification.
• canGenMouseDown – Rboolean: can the device handle mouse down events? This flag and the next three are not currently used by R, but are maintained for back compatibility.
• canGenMouseMove – Rboolean: ditto for mouse move events.
• canGenMouseUp – Rboolean: ditto for mouse up events.
• canGenKeybd – Rboolean: ditto for keyboard events.
• hasTextUTF8 – Rboolean: should non-symbol text be sent (in UTF-8) to the textUTF8 and strWidthUTF8 callbacks, and sent as Unicode points (negative values) to the metricInfo callback?
• wantSymbolUTF8 – Rboolean: should symbol text be handled in UTF-8 in the same way as other text? Requires textUTF8 = TRUE.

Several more were added at R 2.14.0 to support the dev.capabilities function: these are all small integers.

• haveTransparency: does the device support semi-transparent colours?
• haveTransparentBg: can the background be fully or semi-transparent?
• haveRaster: is there support for rendering raster images?
• haveCapture: is there support for grid::grid.cap?
• haveLocator: is there an interactive locator?

The last three can often be deduced to be false from the presence of NULL entries instead of the corresponding functions.

6.1.3 Handling text

Handling text is probably the hardest task for a graphics device, and the design allows for the device to optionally indicate that it has additional capabilities. (If the device does not, these will if possible be handled in the graphics engine.)

The three callbacks for handling text that must be in all graphics devices are text, strWidth and metricInfo with declarations

void text(double x, double y, const char *str, double rot, double hadj,
          pGgcontext gc, pDevDesc dd);

double strWidth(const char *str, pGEcontext gc, pDevDesc dd);

void metricInfo(int c, pGEcontext gc,
               double* ascent, double* descent, double* width,
               pDevDesc dd);

The ‘gc’ parameter provides the graphics context, most importantly the current font and fontsize, and ‘dd’ is a pointer to the active device’s structure.

The text callback should plot ‘str’ at ‘(x, y)’¹⁶ with an anti-clockwise rotation of ‘rot’ degrees. (For ‘hadj’ see below.) The interpretation for horizontal text is that the baseline is at y and the start is a x, so any left bearing for the first character will start at x.

The strWidth callback computes the width of the string which it would occupy if plotted horizontally in the current font. (Width here is expected to include both (preferably) or neither of left and right bearings.)

The metricInfo callback computes the size of a single character: ascent is the distance it extends above the baseline and descent how far it extends below the baseline. width is the amount by which the cursor should be advanced when the character is placed. For ascent and descent this is intended to be the bounding box of the ‘ink’ put down by the glyph and not the box which might be used when assembling a line of conventional text (it needs to be for e.g. hat(beta) to work correctly). However, the width is used in plotmath to advance to the next character, and so needs to include left and right bearings.

The interpretation of ‘c’ depends on the locale. In a single-byte locale values 32...255 indicate the corresponding character in the locale (if present). For the symbol font (as used by ‘graphics::par(font=5)’, ‘grid::gpar(fontface=5’) and by ‘plotmath’), values 32...126, 161...239, 241...254 indicate glyphs in the Adobe Symbol encoding. In a multibyte locale, c represents a Unicode point (except in the symbol font). So the function needs to include code like

    Rboolean Unicode = mbcslocale && (gc->fontface != 5);
    if (c < 0) { Unicode = TRUE; c = -c; } 
    if(Unicode) UniCharMetric(c, ...); else CharMetric(c, ...);

In addition, if device capability hasTextUTF8 (see below) is true, Unicode points will be passed as negative values: the code snippet above shows how to handle this. (This applies to the symbol font only if device capability wantSymbolUTF8 is true.)

If possible, the graphics device should handle clipping of text. It indicates this by the structure element canClip which if true will result in calls to the callback clip to set the clipping region. If this is not done, the engine will clip very crudely (by omitting any text that does not appear to be wholly inside the clipping region).

The device structure has an integer element canHadj, which indicates if the device can do horizontal alignment of text. If this is one, argument ‘hadj’ to text will be called as 0 ,0.5, 1 to indicate left-, centre- and right-alignment at the indicated position. If it is two, continuous values in the range [0, 1] are assumed to be supported.

A new capability in R 2.7.0 (graphics API version 4) was hasTextUTF8. If this is true, it has two consequences. First, there are callbacks textUTF8 and strWidthUTF8 that should behave identically to text and strWidth except that ‘str’ is assumed to be in UTF-8 rather than the current locale’s encoding. The graphics engine will call these for all text except in the symbol font. Second, Unicode points will be passed to the metricInfo callback as negative integers. If your device would prefer to have UTF-8-encoded symbols, define wantSymbolUTF8 as well as hasTextUTF8. In that case text in the symbol font is sent to textUTF8 and strWidthUTF8.

Some devices can produce high-quality rotated text, but those based on bitmaps often cannot. Those which can should set useRotatedTextInContour to be true from graphics API version 4.

Several other elements relate to the precise placement of text by the graphics engine:

double xCharOffset;
double yCharOffset;
double yLineBias;
double cra[2];

These are more than a little mysterious. Element cra provides an indication of the character size, par("cra") in base graphics, in device units. The mystery is what is meant by ‘character size’: which character, which font at which size? Some help can be obtained by looking at what this is used for. The first element, ‘width’, is not used by R except to set the graphical parameters. The second, ‘height’, is use to set the line spacing, that is the relationship between par("mai") and par("mai") and so on. It is suggested that a good choice is

dd->cra[0] = 0.9 * fnsize;
dd->cra[1] = 1.2 * fnsize;

where ‘fnsize’ is the ‘size’ of the standard font (cex=1) on the device, in device units. So for a 12-point font (the usual default for graphics devices), ‘fnsize’ should be 12 points in device units.

The remaining elements are yet more mysterious. The postscript() device says

    /* Character Addressing Offsets */
    /* These offsets should center a single */
    /* plotting character over the plotting point. */
    /* Pure guesswork and eyeballing ... */

    dd->xCharOffset =  0.4900;
    dd->yCharOffset =  0.3333;
    dd->yLineBias = 0.2;

It seems that xCharOffset is not currently used, and yCharOffset is used by the base graphics system to set vertical alignment in text() when pos is specified, and in identify(). It is occasionally used by the graphic engine when attempting exact centring of text, such as character string values of pch in points() or grid.points()—however, it is only used when precise character metric information is not available or for multi-line strings.

yLineBias is used in the base graphics system in axis() and mtext() to provide a default for their ‘padj’ argument.

6.1.4 Conventions

The aim is to make the (default) output from graphics devices as similar as possible. Generally people follow the model of the postscript and pdf devices (which share most of their internal code).

The following conventions have become established:

• The default size of a device should be 7 inches square.
• There should be a ‘pointsize’ argument which defaults to 12, and it should give the pointsize in big points (1/72 inch). How exactly this is interpreted is font-specific, but it should use a font which works with lines packed 1/6 inch apart, and looks good with lines 1/5 inch apart (that is with 2pt leading).
• The default font family should be a sans serif font, e.g Helvetica or similar (e.g. Arial on Windows).
• lwd = 1 should correspond to a line width of 1/96 inch. This will be a problem with pixel-based devices, and generally there is a minimum line width of 1 pixel (although this may not be appropriate where anti-aliasing of lines is used, and cairo prefers a minimum of 2 pixels).
• Even very small circles should be visible, e.g. by using a minimum radius of 1 pixel or replacing very small circles by a single filled pixel.
• How RGB colour values will be interpreted should be documented, and preferably be sRGB.
• The help page should describe its policy on these conventions.

These conventions are less clear-cut for bitmap devices, especially where the bitmap format does not have a design resolution.

The interpretation of the line texture (par("lty") is described in the header GraphicsEngine.h and in the help for par: note that the ‘scale’ of the pattern should be proportional to the line width (at least for widths above the default).

6.1.5 ‘Mode’

One of the device callbacks is a function mode, documented in the header as

     * device_Mode is called whenever the graphics engine
     * starts drawing (mode=1) or stops drawing (mode=0)
     * GMode (in graphics.c) also says that 
     * mode = 2 (graphical input on) exists.
     * The device is not required to do anything

Since mode = 2 has only recently been documented at device level, it is not surprising that was it not used by any device: prior to R 2.7.0 it was not set by grid::grid.locator. It could be used to change the graphics cursor, but devices currently do that in the locator callback. (In base graphics the mode is set for the duration of a locator call, but if type != "n" is switched back for each point whilst annotation is being done.)

Many devices do indeed do nothing on this call, but some screen devices ensure that drawing is flushed to the screen when called with mode = 0. It is tempting to use it for some sort of buffering, but note that ‘drawing’ is interpreted at quite a low level and a typical single figure will stop and start drawing many times. The buffering introduced in the X11() device makes use of mode = 0 to indicate activity: it updates the screen after ca 100ms of inactivity.

As from R 2.14.0 this callback need not be supplied if it does nothing.

6.1.6 Graphics events

Graphics devices may be designed to handle user interaction. The current model is similar to the one introduced in R 2.1.0 for the Windows screen device, but the design was changed in R 2.12.0 to be more open ended.

Users may use grDevices::setGraphicsEventEnv to set the eventEnv environment in the device driver to hold event handlers. When the user calls grDevices::getGraphicsEvent, R will take three steps. First, it sets the device driver member gettingEvent to true for each device with a non-NULL eventEnv entry, and calls initEvent(dd, true) if the callback is defined. It then enters an event loop. Each time through the loop R will process events once, then check whether any device has set the result member of eventEnv to a non-NULL value, and will save the first such value found to be returned. C functions doMouseEvent and doKeybd are provided to call the R event handlers onMouseDown, onMouseMove, onMouseUp, and onKeybd and set eventEnv$result during this step. Finally, initEvent is called again with init=false to inform the the devices that the loop is done, and the result is returned to the user.

6.1.7 Specific devices

Specific devices are mostly documented by comments in their sources, although for devices of many years’ standing those comments can be in need of updating. This subsection is a repository of notes on design decisions.

6.1.7.1 X11()

The X11(type="Xlib") device dates back to the mid 1990’s and was written then in Xlib, the most basic X11 toolkit. It has since optionally made use of a few features from other toolkits: libXt is used to read X11 resources, and libXmu is used in the handling of clipboard selections.

Using basic Xlib code makes drawing fast, but is limiting. There is no support of translucent colours (that came in the Xrender toolkit of 2000) nor for rotated text (which R implements by rendering text to a bitmap and rotating the latter).

The hinting for the X11 window asks for backing store to be used, and some windows managers may use it to handle repaints, but it seems that most repainting is done by replaying the display list (and here the fast drawing is very helpful).

There are perennial problems with finding fonts. Many users fail to realize that fonts are a function of the X server and not of the machine that R is running on. After many difficulties, R tries first to find the nearest size match in the sizes provided for Adobe fonts in the standard 75dpi and 100dpi X11 font packages—even that will fail to work when users of near-100dpi screens have only the 75dpi set installed. The 75dpi set allows sizes down to 6 points on a 100dpi screen, but some users do try to use smaller sizes and even 6 and 8 point bitmapped fonts do not look good.

Introduction of UTF-8 locales has caused another wave of difficulties. X11 has very few genuine UTF-8 fonts, and produces composite fontsets for the iso10646-1 encoding. Unfortunately these seem to have low coverage apart from a few monospaced fonts in a few sizes (which are not suitable for graph annotation), and where glyphs are missing what is plotted is often quite unsatisfactory.

The approach being taken as from R 2.7.0 is to make use of more modern toolkits, namely cairo for rendering and Pango for font management—because these are associated with Gtk+2 they are widely available. Cairo supports translucent colours and alpha-blending (via Xrender), and anti-aliasing for the display of lines and text. Pango’s font management is based on fontconfig and somewhat mysterious, but it seems mainly to use Type 1 and TrueType fonts on the machine running R and send grayscale bitmaps to cairo.

6.1.7.2 windows()

The windows() device is a family of devices: it supports plotting to Windows (enhanced) metafiles, BMP, JPEG, PNG and TIFF files as well as to Windows printers.

In most of these cases the primary plotting is to a bitmap: this is used for the (default) buffering of the screen device, which also enables the current plot to be saved to BMP, JPEG, PNG or TIFF (it is the internal bitmap which is copied to the file in the appropriate format).

The device units are pixels (logical ones on a metafile device).

The code was originally written by Guido Masarotto with extensive use of macros, which can make it hard to disentangle.

For a screen device, xd->gawin is the canvas of the screen, and xd->bm is the off-screen bitmap. So macro DRAW arranges to plot to xd->bm, and if buffering is off, also to xd->gawin. For all other device, xd->gawin is the canvas, a bitmap for the jpeg() and png() device, and an internal representation of a Windows metafile for the win.metafile() and win.print device. Since ‘plotting’ is done by Windows GDI calls to the appropriate canvas, its precise nature is hidden by the GDI system.

Buffering on the screen device is achieved by running a timer, which when it fires copies the internal bitmap to the screen. This is set to fire every 500ms (by default) and is reset to 100ms after plotting activity.

Repaint events are handled by copying the internal bitmap to the screen canvas (and then reinitializing the timer), unless there has been a resize. Resizes are handled by replaying the display list: this might not be necessary if a fixed canvas with scrollbars is being used, but that is the least popular of the three forms of resizing.

Text on the device has moved to ‘Unicode’ (UCS-2) in recent years. As from R 2.7.0, UTF-8 is requested (hasTextUTF8 = TRUE) for standard text, and converted to UCS-2 in the plotting functions in file src/extra/graphapp/gdraw.c. However, GDI has no support for Unicode symbol fonts, and symbols are handled in Adobe Symbol encoding.

Support for translucent colours (with alpha channel between 0 and 255) was introduced in R 2.6.0 for the screen device only, and extended to the bitmap devices in R 2.7.0.¹⁷ This is done by drawing on a further internal bitmap, xd->bm2, in the opaque version of the colour then alpha-blending that bitmap to xd->bm. The alpha-blending routine is in a separate DLL, msimg32.dll, which is loaded on first use.¹⁸ As small a rectangular region as reasonably possible is alpha-blended (this is rectangle r in the code), but things like mitre joins make estimation of a tight bounding box too much work for lines and polygonal boundaries. Translucent-coloured lines are not common, and the performance seems acceptable.

The support for a transparent background in png() predates full alpha-channel support in libpng (let alone in PNG viewers), so makes use of the limited transparency support in earlier versions of PNG. Where 24-bit colour is used, this is done by marking a single colour to be rendered as transparent. R chose ‘#fdfefd’, and uses this as the background colour (in GA_NewPage if the specified background colour is transparent (and all non-opaque background colours are treated as transparent). So this works by marking that colour in the PNG file, and viewers without transparency support see a slightly-off-white background, as if there were a near-white canvas. Where a palette is used in the PNG file (if less than 256 colours were used) then this colour is recorded with full transparency and the remaining colours as opaque. If 32-bit colour were available then we could add a full alpha channel, but this is dependent on the graphics hardware and undocumented properties of GDI.

6.2 Colours

Devices receive colours as an unsigned int (in the GPar structure and some of the devices as the typedef rcolor): the comments in file R_ext/GraphicsDevice.h are the primary documentation. The 4 bytes in the unsigned int are R,G,B and alpha from least to most significant. So each of RGB has 256 levels of luminosity from 0 to 255. The alpha byte represents (from R 2.0.0) opacity, so value 255 is fully opaque and 0 fully transparent: many but not all devices handle semi-transparent colours.

Colors can be created in C via the macro R_RGBA, and a set of macros are defined in R_ext/GraphicsDevice.h to extract the various components.

Colours in the base graphics system were originally adopted from S (and before that the GRZ library from Bell Labs), with the concept of a (variable-sized) palette of colours referenced by numbers ‘1...N’ plus ‘0’ (the background colour). R introduced the idea of referring to colours by character strings, either in the forms ‘#RRGGBB’ or ‘#RRGGBBAA’ (representing the bytes in hex) as given by function rgb() or via names: the 657 known names are given in the character vector colors and in a table in file colors.c. Note that semi-transparent colours are not ‘premultiplied’, so 50% transparent white is ‘#ffffff80’.

What is not clear is how the RGB values are to be mapped to display colours in the graphics device. There was a proposal (http://developer.r-project.org/sRGB-RFC.html) to regard the mapping as the colorspace ‘sRGB’, which was adopted in R 2.13.0. The sRGB colorspace is an industry standard: it is used by Web browsers and JPEGs from all but high-end digital cameras. The interpretation is a matter for graphics devices and for code that manipulates colours, but not for the graphics engine or subsystems.

R uses a painting model similar to PostScript and PDF. This means that where shapes (circles, rectangles and polygons) can both be filled and have a stroked border, the fill should be painted first and then the border (or otherwise only half the border will be visible). Where both the fill and the border are semi-transparent there is some room for interpretation of the intention. Most devices first paint the fill and then the border, alpha-blending at each step. However, PDF does some automatic grouping of objects, and when the fill and the border have the same alpha, they are painted onto the same layer and then alpha-blended in one step. (See p. 569 of the PDF Reference Sixth Edition, version 1.7. Unfortunately, although this is what the PDF standard says should happen, it is not correctly implemented by some viewers.)

6.3 Base graphics

The base graphics system is likely to move to package graphics at some stage, but it currently implemented in files in src/main.

For historical reasons it is largely implemented in two layers. Files plot.c, plot3d.c and par.c contain the code for the around 30 .Internal calls that implement the basic graphics operations. This code then calls functions with names starting with G and declared in header Rgraphics.h in file graphics.c, which in turn call the graphics engine (whose functions almost all have names starting with GE).

A large part of the infrastructure of the base graphics subsystem are the graphics parameters (as set/read by par()). These are stored in a GPar structure declared in the private header Graphics.h. This structure has two variables (state and valid) tracking the state of the base subsystem on the device, and many variables recording the graphics parameters and functions of them.

The base system state is contained in baseSystemState structure defined in R_ext/GraphicsBase.h. This contains three GPar structures and a Boolean variable used to record if plot.new() (or persp) has been used successfully on the device.

The three copies of the GPar structure are used to store the current parameters (accessed via gpptr), the ‘device copy’ (accessed via dpptr) and space for a saved copy of the ‘device copy’ parameters. The current parameters are, clearly, those currently in use and are copied from the ‘device copy’ whenever plot.new() is called (whether or not that advances to the next ‘page’). The saved copy keeps the state when the device was last completely cleared (e.g. when plot.new() was called with par(new=TRUE)), and is used to replay the display list.

The separation is not completely clean: the ‘device copy’ is altered if a plot with log scale(s) is set up via plot.window().

There is yet another copy of most of the graphics parameters in static variables in graphics.c which are used to preserve the current parameters across the processing of inline parameters in high-level graphics calls (handled by ProcessInlinePars).

Snapshots of the base subsystem record the ‘saved device copy’ of the GPar structure.

There remain quite a number of anomalies. For example, function GEcontourLines is (despite its name) coded in file plot3d.c and used to support function contourLines in package grDevices.

6.4 Grid graphics

[At least pointers to documentation.]

7 Tools

The behavior of R CMD check can be controlled through a variety of command line arguments and environment variables.

There is an internal --install=value command line argument not shown by R CMD check --help, with possible values

check:file

Assume that installation was already performed with stdout/stderr to file, the contents of which need to be checked (without repeating the installation). This is useful for checks applied by repository maintainers: it reduces the check time by the installation time given that the package has already been installed. In this case, one also needs to specify where the package was installed to using command line option --library.

fake

Fake installation, and turn off the run-time tests.

skip

Skip installation, e.g., when testing recommended packages bundled with R.

no

The same as --no-install : turns off installation and the tests which require the package to be installed.

The following environment variables can be used to customize the operation of check: a convenient place to set these is the file ~/.R/check.Renviron.

_R_CHECK_ALL_NON_ISO_C_

If true, do not ignore compiler (typically GCC) warnings about non ISO C code in system headers. Note that this may also show additional ISO C++ warnings. Default: false.

_R_CHECK_FORCE_SUGGESTS_

If true, give an error if suggested packages are not available. Default: true (but false for CRAN submission checks).

_R_CHECK_RD_CONTENTS_

If true, check Rd files for auto-generated content which needs editing, and missing argument documentation. Default: true.

_R_CHECK_RD_STYLE_

If true, check whether Rd usage entries for S3 methods use the full function name rather than the appropriate \method markup. Default: true.

_R_CHECK_RD_XREFS_

If true, check the cross-references in .Rd files. Default: true.

_R_CHECK_SUBDIRS_NOCASE_

If true, check the case of directories such as R and man. Default: true

_R_CHECK_SUBDIRS_STRICT_

Initial setting for --check-subdirs. Default: ‘default’ (which checks only tarballs, and checks in the src only if there is no configure file).

_R_CHECK_USE_CODETOOLS_

If true, make use of the codetools package, which provides a detailed analysis of visibility of objects (but may give false positives). Default: true.

_R_CHECK_USE_INSTALL_LOG_

If true, record the output from installing a package as part of its check to a log file (00install.out by default), even when running interactively. Default: true.

_R_CHECK_VIGNETTES_NLINES_

Maximum number of lines to show of the bottom of the output when reporting errors in running vignettes. Default: 10.

_R_CHECK_CODOC_S4_METHODS_

Control whether codoc() testing is also performed on S4 methods. Default: true.

_R_CHECK_DOT_INTERNAL_

Control whether the package code is scanned for .Internal calls, which should only be used by base (and occasionally by recommended) packages. Default: true.

_R_CHECK_EXECUTABLES_

Control checking for executable (binary) files. Default: true.

_R_CHECK_EXECUTABLES_EXCLUSIONS_

Control whether checking for executable (binary) files ignores files listed in the package’s BinaryFiles file. Default: true (but false for CRAN submission checks). However, most likely this package-level override mechanism will be removed eventually.

_R_CHECK_PERMISSIONS_

Control whether permissions of files should be checked. Default: true iff .Platform$OS.type == "unix".

_R_CHECK_FF_CALLS_

Allows turning off checkFF() testing. Legacy mostly. Default: true.

_R_CHECK_LICENSE_

Control whether/how license checks are performed. A possible value is ‘maybe’ (warn in case of problems, but not about standardizable non-standard license specs). Default: true.

_R_CHECK_RD_EXAMPLES_T_AND_F_

Control whether check_T_and_F() also looks for “bad” (global) ‘T’/‘F’ uses in examples. Off by default because this can result in false positives.

_R_CHECK_RD_CHECKRD_MINLEVEL_

Controls the minimum level for reporting warnings from checkRd. Default: -1.

_R_CHECK_XREFS_REPOSITORIES_

If set to a non-empty value, a space-separated list of repositories to use to determine known packages. Default: empty, when the CRAN, Omegahat and Bioconductor repositories known to R is used.

_R_CHECK_SRC_MINUS_W_IMPLICIT_

Control whether installation output is checked for compilation warnings about implicit function declarations (as spotted by GCC with command line option -Wimplicit-function-declaration, which is implied by -Wall). Default: false.

_R_CHECK_SRC_MINUS_W_UNUSED_

Control whether installation output is checked for compilation warnings about unused code constituents (as spotted by GCC with command line option -Wunused, which is implied by -Wall). Default: true.

_R_CHECK_WALL_FORTRAN_

Control whether gfortran 4.0 or later -Wall warnings are used in the analysis of installation output. Default: false, even though the warnings are justifiable.

_R_CHECK_ASCII_CODE_

If true, check R code for non-ascii characters. Default: true.

_R_CHECK_ASCII_DATA_

If true, check data for non-ascii characters. Default: true.

_R_CHECK_COMPACT_DATA_

If true, check data for ascii and uncompressed saves, and also check if using bzip2 or xz compression would be significantly better. Default: true.

_R_CHECK_SKIP_ARCH_

Comma-separated list of architectures that will be omitted from checking in a multi-arch setup. Default: none.

_R_CHECK_SKIP_TESTS_ARCH_

Comma-separated list of architectures that will be omitted from running tests in a multi-arch setup. Default: none.

_R_CHECK_SKIP_EXAMPLES_ARCH_

Comma-separated list of architectures that will be omitted from running examples in a multi-arch setup. Default: none.

_R_CHECK_VC_DIRS_

Should the unpacked package directory be checked for version-control directories (CVS, .svn …)? Default: true for tarballs.

_R_CHECK_PKG_SIZES_

Should du be used to find the installed sizes of packages? R CMD check does check for the availability of du. but this option allows the check to be overruled if an unsuitable command is found (including one that does not respect the -k flag to report in units of 1Kb, or reports in a different format – the GNU, Mac OS X and Solaris du commands have been tested). Default: true if du is found.

_R_CHECK_DOC_SIZES_

Should qpdf be used to check the installed sizes of PDFs? Default: true if qpdf is found.

_R_CHECK_DOC_SIZES2_

Should gs be used to check the installed sizes of PDFs? This is slower than (and in addition to) the previous check, but does detect figures with excessive detail (often hidden by over-plotting) or bitmap figures with too high a resolution. Requires that R_GSCMD is set to a valid program, or gs (or on Windows, gswin32.exe or gswin64c.exe) is on the path. Default: false (but true for CRAN submission checks).

_R_CHECK_ALWAYS_LOG_VIGNETTE_OUTPUT_

By default the output from running the R code in the vignettes is kept only if there is an error. Default: false.

_R_CHECK_CLEAN_VIGN_TEST_

Should the vign_test directory be removed if the test is successful? Default: true.

_R_CHECK_REPLACING_IMPORTS_

Should warnings about replacing imports be reported? These mainly come from auto-generated NAMESPACE files in other packages. Default: false.

_R_CHECK_UNSAFE_CALLS_

Check for calls that appear to tamper with (or allow tampering with) already loaded code not from the current package: such calls may well contravene CRAN policies. Default: true.

_R_CHECK_TIMINGS_

Optionally report timings for installation, examples, tests and running/re-building vignettes as part of the check log. The format is ‘[as/bs]’ for the total CPU time (including child processes) ‘a’ and elapsed time ‘b’, except on Windows, when it is ‘[bs]’. In most cases timings are only given for ‘OK’ checks. Times with an elapsed component over 10 mins are reported in minutes (with abbreviation ‘m’). The value is the smallest numerical value in elapsed seconds that should be reported: non-numerical values indicate that no report is required, a value of ‘0’ that a report is always required. Default: "". (10 for CRAN checks.)

_R_CHECK_INSTALL_DEPENDS_

If set to a true value and a test installation is to be done, this is done with .libPaths() containing just a temporary library directory and .Library. The temporary library is populated by symbolic links¹⁹ to the installed copies of all the Depends/Imports/LinkingTo packages which are not in .Library. Default: false (but true for CRAN submission checks).

Note that this is actually implemented in R CMD INSTALL, so it is available to those who first install recording to a log, then call R CMD check.

_R_CHECK_DEPENDS_ONLY_

_R_CHECK_SUGGESTS_ONLY_

If set to a true value, running examples, tests and vignettes is done with .libPaths() containing just a temporary library directory and .Library. The temporary library is populated by symbolic links²⁰ to the installed copies of all the Depends/Imports/LinkingTo and (for the second only) Suggests packages which are not in .Library. Default: false (but true for CRAN checks).

_R_CHECK_NO_RECOMMENDED_

If set to a true value, augment the previous checks to make recommended packages unavailable unless declared. Default: false (but true for CRAN submission checks).

This may give false positives on code which uses grDevices::densCols and stats:::asSparse as these invoke KernSmooth and Matrix respectively.

_R_CHECK_CODETOOLS_PROFILE_

A string with comma-separated name=value pairs (with value a logical constant) giving additional arguments for the codetools functions used for analyzing package code. E.g., use _R_CHECK_CODETOOLS_PROFILE_="suppressLocalUnused=FALSE" to turn off suppressing warnings about unused local variables. Default: no additional arguments, corresponding to using skipWith = TRUE, suppressPartialMatchArgs = FALSE and suppressLocalUnused = TRUE.

_R_CHECK_CRAN_INCOMING_

Check whether package is suitable for publication on CRAN. Default: false, except for CRAN submission checks.

_R_CHECK_XREFS_USE_ALIASES_FROM_CRAN_

When checking anchored Rd xrefs, use Rd aliases from the CRAN package web areas in addition to those in the packages installed locally. Default: false.

_R_SHLIB_BUILD_OBJECTS_SYMBOL_TABLES_

Make the checks of compiled code more accurate by recording the symbol tables for objects (.o files) at installation in a file symbols.rds. (Only currently supported on Linux, Solaris, OS X, Windows and FreeBSD.) Default: true.

CRAN’s submission checks use something like

_R_CHECK_CRAN_INCOMING_=TRUE
_R_CHECK_VC_DIRS_=TRUE
_R_CHECK_TIMINGS_=10
_R_CHECK_INSTALL_DEPENDS_=TRUE
_R_CHECK_SUGGESTS_ONLY_=TRUE
_R_CHECK_NO_RECOMMENDED_=TRUE
_R_CHECK_EXECUTABLES_EXCLUSIONS_=FALSE
_R_CHECK_DOC_SIZES2_=TRUE

These are turned on by R CMD check --as-cran: the incoming checks also use

_R_CHECK_FORCE_SUGGESTS_=FALSE

since some packages do suggest other packages not available on CRAN or other commonly-used repositories.

8 R coding standards

R is meant to run on a wide variety of platforms, including Linux and most variants of Unix as well as Windows and Mac OS X. Therefore, when extending R by either adding to the R base distribution or by providing an add-on package, one should not rely on features specific to only a few supported platforms, if this can be avoided. In particular, although most R developers use GNU tools, they should not employ the GNU extensions to standard tools. Whereas some other software packages explicitly rely on e.g. GNU make or the GNU C++ compiler, R does not. Nevertheless, R is a GNU project, and the spirit of the GNU Coding Standards should be followed if possible.

The following tools can “safely be assumed” for R extensions.

• An ISO C99 C compiler. Note that extensions such as POSIX 1003.1 must be tested for, typically using Autoconf unless you are sure they are supported on all mainstream R platforms (including Windows and Mac OS X). Packages will be more portable to R < 2.12.0 if written assuming only C89, but this should not be done where using C99 features will make for cleaner or more robust code.
• A FORTRAN 77 compiler (but not Fortran 9x).
• A simple make, considering the features of make in 4.2 BSD systems as a baseline.

  GNU or other extensions, including pattern rules using ‘%’, the automatic variable ‘$^’, the ‘+=’ syntax to append to the value of a variable, the (“safe”) inclusion of makefiles with no error, conditional execution, and many more, must not be used (see Chapter “Features” in the GNU Make Manual for more information). On the other hand, building R in a separate directory (not containing the sources) should work provided that make supports the VPATH mechanism.

  Windows-specific makefiles can assume GNU make 3.79 or later, as no other make is viable on that platform.

• A Bourne shell and the “traditional” Unix programming tools, including grep, sed, and awk.

  There are POSIX standards for these tools, but these may not be fully supported. Baseline features could be determined from a book such as The UNIX Programming Environment by Brian W. Kernighan & Rob Pike. Note in particular that ‘|’ in a regexp is an extended regexp, and is not supported by all versions of grep or sed. The Open Group Base Specifications, Issue 7, which are technically identical to IEEE Std 1003.1 (POSIX), 2008, are available at http://pubs.opengroup.org/onlinepubs/9699919799/mindex.html.

Under Windows, most users will not have these tools installed, and you should not require their presence for the operation of your package. However, users who install your package from source will have them, as they can be assumed to have followed the instructions in “the Windows toolset” appendix of the “R Installation and Administration” manual to obtain them. Redirection cannot be assumed to be available via system as this does not use a standard shell (let alone a Bourne shell).

In addition, the following tools are needed for certain tasks.

• Perl version 5 is only needed for a few uncommonly-used tools: make install-info needs Perl installed if there is no command install-info on the system, and for the maintainer-only script tools/help2man.pl.
• Makeinfo version 4.7 or later is needed to build the Info files for the R manuals written in the GNU Texinfo system.

It is also important that code is written in a way that allows others to understand it. This is particularly helpful for fixing problems, and includes using self-descriptive variable names, commenting the code, and also formatting it properly. The R Core Team recommends to use a basic indentation of 4 for R and C (and most likely also Perl) code, and 2 for documentation in Rd format. Emacs (21 or later) users can implement this indentation style by putting the following in one of their startup files, and using customization to set the c-default-style to "bsd" and c-basic-offset to 4.)

;;; ESS
(add-hook 'ess-mode-hook
          (lambda ()
            (ess-set-style 'C++ 'quiet)
            ;; Because
            ;;                                 DEF GNU BSD K&R C++
            ;; ess-indent-level                  2   2   8   5   4
            ;; ess-continued-statement-offset    2   2   8   5   4
            ;; ess-brace-offset                  0   0  -8  -5  -4
            ;; ess-arg-function-offset           2   4   0   0   0
            ;; ess-expression-offset             4   2   8   5   4
            ;; ess-else-offset                   0   0   0   0   0
            ;; ess-close-brace-offset            0   0   0   0   0
            (add-hook 'local-write-file-hooks
                      (lambda ()
                        (ess-nuke-trailing-whitespace)))))
(setq ess-nuke-trailing-whitespace-p 'ask)
;; or even
;; (setq ess-nuke-trailing-whitespace-p t)

;;; Perl
(add-hook 'perl-mode-hook
          (lambda () (setq perl-indent-level 4)))

(The ‘GNU’ styles for Emacs’ C and R modes use a basic indentation of 2, which has been determined not to display the structure clearly enough when using narrow fonts.)

9 Testing R code

When you (as R developer) add new functions to the R base (all the packages distributed with R), be careful to check if make test-Specific or particularly, cd tests; make no-segfault.Rout still works (without interactive user intervention, and on a standalone computer). If the new function, for example, accesses the Internet, or requires GUI interaction, please add its name to the “stop list” in tests/no-segfault.Rin.

[To be revised: use make check-devel, check the write barrier if you change internal structures.]

10 Use of TeX dialects

Various dialects of TeX and used for different purposes in R. The policy is that manuals be written in ‘texinfo’, and for convenience the main and Windows FAQs are also. This has the advantage that is is easy to produce HTML and plain text versions as well as typeset manuals.

LaTeX is not used directly, but rather as an intermediate format for typeset help documents and for vignettes.

Care needs to be taken about the assumptions made about the R user’s system: it may not have either ‘texinfo’ or a TeX system installed. We have attempted to abstract out the cross-platform differences, and almost all the setting of typeset documents is done by tools::texi2dvi. This is used for offline printing of help documents, preparing vignettes and for package manuals via R CMD Rd2pdf. It is not currently used for the R manuals created in directory doc/manual.

tools::texi2dvi makes use of a system command texi2dvi where available. On a Unix-alike this is usually part of ‘texinfo’, whereas on Windows if it exists at all it would be an executable, part of MiKTeX. If none is available, the R code runs a sequence of (pdf)latex, bibtex and makeindex commands.

This process has been rather vulnerable to the versions of the external software used: particular issues have been texi2dvi and texinfo.tex updates, mismatches between the two²¹, versions of the LaTeX package ‘hyperref’ and quirks in index production. The licenses used for LaTeX and latterly ‘texinfo’ prohibit us from including ‘known good’ versions in the R distribution.

On a Unix-alike configure looks for the executables for TeX and friends and if found records the absolute paths in the system Renviron file. This used to record ‘false’ if no command was found, but it nowadays records the name for looking up on the path at run time. The latter can be important for binary distributions: one does not want to be tied to, for example, TeXLive 2007.

Function and variable index

Concept index