All Projects → facebookresearch → Cparser

facebookresearch / Cparser

Licence: mit
A compact C preprocessor and declaration parser written in pure Lua

Programming Languages

lua
6591 projects

CParser

This pure Lua module implements (1) a standard compliant C preprocessor with a couple useful extensions, and (2) a parser that provides a Lua friendly description of all global declarations and definitions in a C header or C program file.

The driver program lcpp invokes the preprocessor and outputs preprocessed code. Although it can be used as a replacement for the normal preprocessor, it is more useful as an extra preprocessing step (see option -Zpass which is on by default.) The same capabilities are offered by functions cparser.cpp and cparser.cppTokenIterator provided by the module cparser.

The driver program lcdecl analyzes a C header file and a C program file and outputs a short descriptions of the declarations and definitions. This program is mostly useful to understand the representations produced by the cparser function cparser.declarationIterator.

This code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree.


Program lcpp

Synopsis

    lcpp [options] inputfile.c [-o outputfile.c]

Preprocess file inputfile.c and write the preprocessed code into file outputfile.c or to the standard output.

Options

The following options are recognized:

  • -Werror
    Cause all warning to be treated as errors. Note that parsing cannot resume after an error. The parser simply throws a Lua error.

  • -w
    Do not print warning messages.

  • -Dsym[=val]
    Define preprocessor symbol sym to value val. The default value of val is 1. Note that it is possible to define function-like symbols with syntax -Dsym(args)=val.

  • -Usym
    Undefine preprocessor symbol sym.

  • -Idir
    Add directory dir to the search path for included files. Note that there is no default search path. When an include file is not found the include directive is simply ignored with a warning (but see also option -Zpass). Therefore all include directives are ignored unless one uses option -I to specify the search path.

  • -I-
    Marks the beginning of the system include path. When an included file is given with angle brackets, (as in #include <stdio.h>), one only searches directories specified by the -I options that follow -I-. Therefore all these include directives are ignored unless one uses option -I- followed by one or more option -I.

  • -dM
    Instead of producing the preprocessed file, dumps all macros defined at the end of the parse.

  • -Zcppdef
    Run the native preprocessor using command cpp -dM < dev/null and copy its predefined symbols. This is useful when using lcpp as a full replacement for the standard preprocessor.

  • -Zpass
    This option is enabled by default (use -Znopass to disable) and indicates that the output of lcpp is going to be reprocessed by a C preprocessor and compiler. This option triggers the following behavior:

    • The preprocessor directives #pragma and #ident are copied verbatim into the output file.
    • When the included file cannot be found in the provided search path, the preprocessor directive #include is copied into the output file.
    • Preprocessor directives prefixed with a double ## are copied verbatim into the output file with a single # prefix. This feature is useful for #if directives that depend on symbols defined by unresolved #include directives.
  • -std=(c|gnu)(89|99|11)
    This option selects a C dialect. In the context of the preprocessor, this impacts the symbols predefined by lcpp and potentially enables GCC extensions of the variadic macro definition syntax.

    • Symbol __CPARSER__ is always defined with value <1>.
    • Symbols __STDC__ and __STDC_VERSION__ are either defined by option -Zcppdef or take values suitable for the target C dialect.
    • Symbols __GNUC__ and __GNUC_MINOR__ are either defined by option -Zcppdef or are defined to values 4 and 2 if the target dialect starts with string gnu.

    This can be further adjusted using the -D or -U options. The default dialect is gnu99.

Preprocessor extensions

The lcpp preprocessor implements several useful nonstandard features. The main feature are multiline macros. The other features are mostly here because they make multiline macros more useful.

String comparison in conditional expressions

The C standard specifies that the expressions following #if directives are constant expressions of integral type. However this processor also handles strings. The only valid operations on strings are the equality and ordering comparisons. This is quite useful to make special cases for certain values of the parameters of a multiline macro, as shown later.

Multiline macros

Preprocessor directives #defmacro and #endmacro can be used to define a function-like macro whose body spans several lines. The #defmacro directive contains the macro name and a mandatory argument list. The body of the macro is composed of all the following lines up to the matching #endmacro. This offers several benefits:

  • The line numbers of the macro-expansion is preserved. This ensures that the compiler produces error messages with meaningful line numbers.

  • The multi-line macro can contain preprocessor directives. Conditional directives are very useful in this context. Note however that preprocessor definitions (with #define, #defmacro, or #undef) nested inside multiline macros are only valid within the macro.

  • The standard stringification # and token concatenation ## operators can be used freely in the body of multiline macros. Note that these operators only work with the parameters of the multiline macros and not with ordinary preprocessor definitions. This is consistent with the standard behavior of these operators in ordinary preprocessor macros.

    Example

      #defmacro DEFINE_VDOT(TNAME, TYPE)
        TYPE TNAME##Vector_dot(TYPE *a, TYPE *b, int n)
        {
          /* try cblas */
        #if #TYPE == "float"
          return cblas_sdot(n, a, 1, b, 1);
        #elif #TYPE == "double"
          return cblas_ddot(n, a, 1, b, 1);
        #else
          int i;
          TYPE s = 0;
          for(i=0;i<n;i++)
            s += a[i] * b[i];
          return s;
        #endif
        }
      #endmacro

      DEFINE_VDOT(Float,float);
      DEFINE_VDOT(Double,double);
      DEFINE_VDOT(Int,int);

Details -- The values of the macro parameters are normally macro-expanded before substituting them into the text of the macro. However this macro-expansion does not happen when the substitution occurs in the context of a stringification or token concatenation operator. All this is consistent with the standard. The novelty is that this macro-expansion does not occur either when the parameter appears in a nested preprocessor directive or multiline macro.

More details -- The stringification operator only works when the next non-space token is a macro parameter. This provides a good way to distinguish a nested directive from a stringification operator appearing in the beginning of a line.

Even more details -- The standard mandates that the tokens generated by a macro-expansion can be combined with the following tokens to compose a new macro invocation. This is not allowed for multiline macros. An error is signaled if the expansion of a multiline macro generates an incomplete macro argument list.

Negative comma in variadic macros

Consider the following variadic macro

     #define macro(msg, ...)  printf(msg, __VA_ARGS__)

The C standard says that it is an error to call this macro with only one argument. Calling this macro with an empty second argument --macro(msg,)-- leaves an annoying comma in the expansion --printf(msg,)-- and causes a compiler syntax error.

This preprocessor accepts invocations of such a macro with a single argument. The value of parameter __VA_ARGS__ is then a so-called negative comma, meaning that the preceding comma is eliminated when this parameter appears in the macro definition between a comma and a closing parenthesis.

Recursive macros

When a new invocation of the macro appears in the expansion of a macro, the standard specifies that the preprocessor must rescan the expansion but should not recursively expand the macro. Although this restriction is both wise and useful, there are rare cases where one would like to use recursive macros. As an experiment, this recursion prevention feature is turned off when one defines a multiline macro with #defrecursivemacro instead of #defmacro. Note that this might prevent the preprocessor from terminating unless the macro eventually takes a conditional branch that does not recursively invoke the macro.


Program lcdecl

Synopsis

    ldecl [options] inputfile.c [-o outputfile.txt]

Preprocess and parse file inputfile.c. The output of a parser is a sequence of Lua data structures representing each C definition or declaration encountered in the code. Program ldecl prints each of them in two forms. The first form directly represent the Lua tables composing the data structure. The second form reconstructs a piece of C code representing the definition or declaration of interest.

This program is mostly useful to people working with the Lua functions offered by the cparser module because it provides a quick way to inspect the resulting data structures.

Options

Program lcdecl accepts all the preprocessing options documented for program lcpp. It also accepts an additional option -Ttypename and also adds to the meaning of options -Zpass and -std=dialect.

  • -Ttypename
    Similar to lcpp, program lcdecl only reads the include files that are found along the path specified by the -I options. It is generally not desirable to read all include files because they often contain declarations that are not directly useful. This also means that the C parser may not be aware of type definitions found in ignored include files. Although the C syntax is sufficiently unambiguous to allow the parser to guess that an identifier is a type name rather than a variable name, this can lead to confusing error messages. Option -Ttypename can then be used to inform the parser than symbol typename represents a type and not a constant, a variable, or a function.

  • -Zpass Unlike lcpp, program lcdecl processes the input file with option -Zpass off by default. Turning it on will just eliminate potentially useful warning messages.

  • -Ztag This option causes lcdecl to treat all structs, unions, and enums as tagged types, possibly using synthetic tags of the form __anon_XXXXX. It is assumed that such names are not used anywhere in the parsed program. This is useful for certain code transformation applications.

  • -std=(c|gnu)(89|99|11)
    The dialect selection options also control whether the parser recognizes keywords introduced by later version of the C standard (e.g., restrict, _Bool, _Complex, _Atomic, _Pragma, inline) or by the GCC compiler (e.g., asm). Many of these keywords have a double-underline-delimited variant that is recognized in all cases (e.g, __restrict__).

Example.

Running ldecl on the following program

const int size = (3+2)*2;
float arr[size];
typedef struct symtable_s { const char *name; SymVal value; } symtable_t;
void printSymbols(symtable_t *p, int n) { do_something(p,n); }

produces the following output (with very long lines).

+--------------------------
| Definition{where="test.c:2",intval=10,type=Qualified{t=Type{n="int"},const=true},name="size",init={..}}
| const int size = 10
+--------------------------
| Definition{where="test.c:3",type=Array{t=Type{n="float"},size=10},name="arr"}
| float arr[10]
+--------------------------
| TypeDef{sclass="[typetag]",where="test.c:4",type=Struct{Pair{Pointer{t=Qualified{t=Type{n="char"},const=true}},"name"},Pair{Type{n="SymVal"},"value"},n="symtable_s"},name="struct symtable_s"}
| [typetag] struct symtable_s{const char*name;SymVal value;}
+--------------------------
| TypeDef{sclass="typedef",where="test.c:4",type=Type{_def={..},n="struct symtable_s"},name="symtable_t"}
| typedef struct symtable_s symtable_t
+--------------------------
| Definition{where="test.c:5",type=Function{Pair{Pointer{t=Type{_def={..},n="symtable_t"}},"p"},Pair{Type{n="int"},"n"},t=Type{n="void"}},name="printSymbols",init={..}}
| void printSymbols(symtable_t*p,int n){..}
+--------------------------

Module cparser

Module cparser exports the following functions:

Preprocessing function

cparser.cpp(filename, outputfile, options)

Program lcpp is implemented by function cparser.cpp.

Calling this function preprocesses file filename and writes the preprocessed code to the specified output. The optional argument outputfile can be a file name or a Lua file descriptor. When this argument is nil, the preprocessed code is written to the standard output. The optional argument options is an array of option strings. All the options documented with program lcpp are supported.

cparser.cppTokenIterator(options, lines, prefix)

Calling this function produces two results:

  • A token iterator function.
  • A macro definition table.

Argument options is an array of options. All the options documented for program lcpp are supported. Argument lines is an iterator that returns input lines. Lua provides many such iterators, including io.lines(filename) to return the lines of the file named filename and filedesc:lines() to return lines from the file descriptor filedesc. You can also use string.gmatch(somestring,'[^\n]+') to return lines from string somestring.

Each successive call of the token iterator function describes a token of the preprocessed code by returning two strings. The first string represent the token text. The second string follows format "filename:lineno" and indicates on which line the token was found. The filename either is the argument prefix or is the actual name of an included file. When all the tokens have been produced, the token iterator function returns nil.

Each named entry of the macro definition table contains the definition of the corresponding preprocessor macros. Function cparser.macroToString can be used to reconstruct the macro definition from this information.

Example:

      ti,macros = cparser.cppTokenIterator(nil, io.lines('test/testmacro.c'), 'testmacro.c')
      for token,location in ti do
        print(token, location)
      end
      for symbol,_ in pairs(macros) do
        local s = cparser.macroToString(macros,symbol)
        if s then print(s) end
      end
cparser.macroToString(macros,name)

This function returns a string representing the definition of the preprocessor macro name found in macro definition table macros. It returns nil if no such macro is defined. Note that the macro definition table contains named entries that are not macro definitions but functions implementing magic symbols such as __FILE__ or __LINE__.

Parsing functions

cparser.parse(filename, outputfile, options)

Program lcdecl is implemented by function cparser.parse.

Calling this function preprocesses and parses file filename, writing a trace into the specified file. The optional argument outputfile can be a file name or a Lua file descriptor. When this argument is nil, the preprocessed code is written to the standard output. The optional argument options is an array of option strings. All the options documented with program lcdecl are supported.

cparser.declarationIterator(options, lines, prefix)

Calling this function produces three results:

  • A declaration iterator function.
  • A symbol table.
  • A macro definition table.

Argument options is an array of options. All the options documented for program lcdecl are supported. Argument lines is an iterator that returns input lines. Lua provides many such iterators, including io.lines(filename) to return the lines of the file named filename and filedesc:lines() to return lines from the file descriptor filedesc. You can also use string.gmatch(somestring,'[^\n]+') to return lines from string somestring.

Each successive call of the declaration iterator function returns a Lua data structure that represents a declaration, a definition, or certain preprocessor events. The format of these data structures is described under function cparser.declToString.

The symbol table contains the definition of all the C language identifiers defined or declared by the parsed files. Type names are represented by the Type{} data structure documented under function cparser.typeToString. Constants, variables, and functions are represented by Definition{} or Declaration{} data structures similar to those returned by the declaration iterator.

The macro definition table contains the definition of the preprocessor macros. See the documentation of function macroToString for details.

Example

      di = cparser.declarationIterator(nil, io.lines('tests/testmacro.c'), 'testmacro.c')
      for decl in di do print(decl) print(">>", cparser.declToString(decl)) end
cparser.typeToString(ty,nam)

This function produces a string suitable for declaring a variable nam with type ty in a C program.

Argument ty is a type data structure. Argument nam should be a string representing a legal identifier. However it defaults to %s in order to compute a format string suitable for the standard Lua function string.format.

Module cparser represents each type with a tree whose nodes are Lua tables tagged by their tag field. These tables are equipped with a convenient metatable method that prints them compactly by first displaying the tag then the remaining fields using the standard Lua construct.

For instance, the type const int is printed as

      Qualified{t=Type{n="int"},const=true}

and corresponds to

      {
        tag="Qualified",
        const=true,
        t= {
             tag="Type",
             n = "int"
           }
      }

The following tags are used to represent types.

  • Type{n=name} is used to represent a named type name. There is only one instance of each named type. Names can be made of multiple keywords, such as int or unsigned long int, they can also be typedef identifiers, such as size_t, or composed names, such as struct foo or enum bar. This construct can also contain a field _def that points to the definition of the named type if such a definition is known.

  • Qualified{t=basetype,...} is used to represent a qualified variant of basetype. Fields named const, volatile, or restrict are set to true to represent the applicable type qualifiers. When the type appears in function parameters and the base type is a pointer, a field named static may contain the guaranteed array size.

  • Pointer{t=basetype} is used to represent a pointer to an object of type basetype. This construct may also contains a field block=true to indicate that the pointer refers to a code block (a C extension found in Apple compilers) or a field ref=true to indicate a reference type (a C extension inspired by C++.)

  • Array{t=basetype,size=s} is used to represent an array of object of type basetype. The optional field size contains the array size when an array size is specified. The size is usually an integer. However there are situations in which the parser is unable to evaluate the size, for instance because it relies on the C keyword sizeof(x). In such cases, field size is a string containing a C expression for the size.

  • Struct{} and Union{} are used to represent the corresponding C types. The optional field n contains the structure tag. Each entry is represented by a Pair{type,name} construct located at successive integer indices. This means that the type of the third entry of structure type ty can be accessed as ty[3][1] and the corresponding name is ty[3][2]. In the case of Struct{} tables, the pairs optionally contain a field bitfield to indicate a bitfield size for the structure entry. Field bitfield usually contains a small integer but can also contain a string representing a C expression (just like field size in the Array{} construct.)

  • Enum{} is used to represent an enumerated type. The optional field n may contain the enumeration tag name. The enumeration constants are reprsented as Pair{name,value} located at successive integer indices. The second pair element is only given when the C code contains an explicit value. It can be an integer or an expression strint (just like field size in Array{}).

  • Function{t=returntype} is used to represent functions returning an object of type returntype. Field withoutProto is set to true when the function does not provide a prototype. Otherwise the arguments are described by Pair{type,name} located as integer indices. The prototype of variadic functions end with a Pair{ellipsis=true} to represent the ... argument.

The Qualified{}, Function{}, Struct{}, Union{}, and Enum{} tables may additionally have a field attr whose contents represents attribute information, such as C11 attributes [[...]], MSVC-style attributes __declspec(...) or GNU attributes __attribute__(...). This is representing by an array containing all the attribute tokens (on odd indices) and their locations (on even indices).

cparser.stringToType(s)

Parses string s as an abstract type name or a variable declaration and returns the type object and possibly the variable name. This function returns nil when the string cannot be interpreted as a type or a declaration, or when the declaration specifies a storage class.

Example

      > return cparser.stringToType("int(*)(const char*)")
      Pointer{t=Function{Pair{Pointer{t=Qualified{const=true,t=Type{n="char"}}}},t=Type{n="int"}}}	nil
cparser.declToString(decl)

This function produces a string that describes the data structures returned by the declaration iterator. There are in fact three kinds of data structures. All these structures have very similar fields. In particular, field where always contains the location of the definition or declaration.

  • TypeDef{name=n,sclass=s,type=ty} represents a type definition. This construct is produced in two different situations. When the C program contains a typedef keyword, field sclass contains the string "typedef", field name contains the new type name, and field type contains the type description. When the C program defines a tagged struct, union, or enum type, field sclass contains the string "[typetag]", field name contains the tagged type name (e.g, "struct foo"), and field type contains the type definition (e.g., Struct{...}).

  • Declaration{name=n,sclass=s,type=ty,...} represents the declaration of a variable or function that is defined elsewhere. Field name gives the variable or function name. Field type gives its type. Field sclass can be empty, "extern", or "static".

  • Definition{name=n,sclass=s,type=ty...} represents the definition of a constant, a variable, or a function. Field name again gives the name, field type gives its type, field sclass gives its storage class, and field init may contain an array of tokens and token locations representing a variable initializer or a function body. Constant definitions may also have a field intval contaning the value of an integer constant. This field works like the size of an array: it often contains a small integer but can also contains a string representing the C expression that the parser was unable to evaluate for one reason or another. Enumeration constants are reported with storage class "[enum]" and with a constant integer type containing an additional field _enum that points to the corresponding enumerated type.

  • CppEvent{directive=dir,...} describes certain preprocessor events that are potentially relevant to a C API. In particular, the definition of an object-like macro s with an integer value v is reported as CppEvent{directive="define",name="s",intval=v} and its deletion as CppEvent{directive="undef",name="s"}. Finally, CppEvent{directive="include",name="fspec"} indicates that an include directive was not resolved.

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].