XS::Parse::Keyword(3pm) User Contributed Perl Documentation XS::Parse::Keyword(3pm)

"XS::Parse::Keyword" - XS functions to assist in parsing keyword syntax

This module provides some XS functions to assist in writing syntax modules that provide new perl-visible syntax, primarily for authors of keyword plugins using the "PL_keyword_plugin" hook mechanism. It is unlikely to be of much use to anyone else; and highly unlikely to be any use when writing perl code using these. Unless you are writing a keyword plugin using XS, this module is not for you.

This module is also currently experimental, and the design is still evolving and subject to change. Later versions may break ABI compatibility, requiring changes or at least a rebuild of any module that depends on it.

void boot_xs_parse_keyword(double ver);

Call this function from your "BOOT" section in order to initialise the module and parsing hooks.

ver should either be 0 or a decimal number for the module version requirement; e.g.

boot_xs_parse_keyword(0.14);

void register_xs_parse_keyword(const char *keyword,
  const struct XSParseKeywordHooks *hooks, void *hookdata);

This function installs a set of parsing hooks to be associated with the given keyword. Such a keyword will then be handled automatically by a keyword parser installed by "XS::Parse::Keyword" itself.

The "XSParseKeywordHooks" structure provides the following hook stages, which are invoked in the given order.

The following flags are defined:

"XPK_FLAG_EXPR"
The parse or build function is expected to return "KEYWORD_PLUGIN_EXPR".
"XPK_FLAG_STMT"
The parse or build function is expected to return "KEYWORD_PLUGIN_STMT".

These two flags are largely for the benefit of giving static information at registration time to assist static parsing or other related tasks to know what kind of grammatical element this keyword will produce.

"XPK_FLAG_AUTOSEMI"
The syntax forms a complete statement, which should be followed by a statement separator semicolon (";"). This semicolon is optional at the end of a block.

The semicolon, if present, will be consumed automatically.

"XPK_FLAG_BLOCKSCOPE"
The entire parse and build process will be wrapped in a pair of block_start() and block_end() calls. This ensures that, for example, any newly-introduced lexical variables do not escape from the scope of the syntax created by the keyword.

const char *permit_hintkey;
bool (*permit) (pTHX_ void *hookdata);

Called by the installed keyword parser hook which is used to handle keywords registered by "register_xs_parse_keyword".

As a shortcut for the common case, the "permit_hintkey" may point to a string to look up from the hints hash. If the given key name is not found in the hints hash then the keyword is not permitted. If the key is present then the "permit" function is invoked as normal.

If not rejected by a hint key that was not found in the hints hash, the function part of the stage is called next and should inspect whether the keyword is permitted at this time perhaps by inspecting other lexical clues, and return true only if the keyword is permitted.

Both the string and the function are optional. Either or both may be present. If neither is present then the keyword is always permitted - which is likely not what you wanted to do.

void (*check)(pTHX_ void *hookdata);

Invoked once the keyword has been permitted. If present, this hook function can check the surrounding lexical context, state, or other information and throw an exception if it is unhappy that the keyword should apply in this position.

This stage is invoked once the keyword has been checked, and actually parses the incoming text into an optree. It is implemented by calling the first of the following function pointers which is not NULL. The invoked function may optionally build an optree to represent the parsed syntax, and place it into the variable addressed by "out". If it does not, then a simple "OP_NULL" will be constructed in its place.

lex_read_space() is called both before and after this stage is invoked, so in many simple cases the hook function itself does not need to bother with it.

int (*parse)(pTHX_ OP **out, void *hookdata);

If present, this should consume text from the parser buffer by invoking "lex_*" or "parse_*" functions and eventually return a "KEYWORD_PLUGIN_*" result value.

This is the most generic and powerful of the options, but requires the most amount of implementation work.

int (*build)(pTHX_ OP **out, XSParseKeywordPiece *args[], size_t nargs, void *hookdata);

If "parse" is not present, this is called instead after parsing a sequence of arguments, of types given by the pieces field; which should be a zero- terminated array of piece types.

This alternative is somewhat less generic and powerful than providing "parse" yourself, but involves much less parsing work and is shorter and easier to implement.

int (*build1)(pTHX_ OP **out, XSParseKeywordPiece *arg0, void *hookdata);

If neither "parse" nor "build" are present, this is called as a simpler variant of "build" when only a single argument is required. It takes its type from the "piece1" field instead.

When using the "build" or "build1" alternatives for the "parse" phase, the actual syntax is parsed automatically by this module, according to the specification given by the pieces or piece1 field. The result of that parsing step is placed into the args or arg0 parameter to the invoked function, using a "struct" type consisting of the following fields:

typedef struct
   union {
      OP *op;
      CV *cv;
      SV *sv;
      int i;
      struct {
         SV *name;
         SV *value;
      } attr;
      PADOFFSET padix;
      struct XSParseInfixInfo *infix;
   };
   int line;
} XSParseKeywordPiece;

Which field of the anonymous union is set depends on the type of the piece. The line field contains the line number of the source file where parsing of that piece began.

Some piece types are "atomic", whose definition is self-contained. Others are structural, defined in terms of inner pieces. Together these form an entire tree-shaped definition of the syntax that the keyword expects to find.

Atomic types generally provide exactly one argument into the list of args (with the exception of literal matches, which do not provide anything). Structural types may provide an initial argument themselves, followed by a list of the values of each sub-piece they contained inside them. Thus, while the data structure defining the syntax shape is a tree, the argument values it parses into is passed as a flat array to the "build" function.

Some structural types need to be able to determine whether or not syntax relating some optional part of them is present in the incoming source text. In this case, the pieces relating to those optional parts must support "probing". This ability is also noted below.

Many of the atomic piece types have a variant which is optional; if the given input does not look like the expected syntax for the piece type then an "_OPT"-suffixed version of the type will instead yield "NULL" in its result pointer.

The type of each piece should be one of the following macro values.

atomic, can probe, emits op.

XPK_BLOCK

A brace-delimited block of code is expected, passed as an optree in the op field. This will be parsed as a block within the current function scope.

This can be probed by checking for the presence of an open-brace ("{") character.

Be careful defining grammars with this because an open-brace is also a valid character to start a term expression, for example. Given a choice between "XPK_BLOCK" and "XPK_TERMEXPR", either of them could try to consume such code as

{ 123, 456 }

Variants of "XPK_BLOCK" which wrap a void, scalar or list-context scope around the block.

structural, emits op.

XPK_PREFIXED_BLOCK(pieces ...)

Some pieces are expected, followed by a brace-delimited block of code, which is passed as an optree in the op field. The prefix pieces are parsed first, and their results are passed before the block itself.

The entire sequence, including the prefix items, is contained within a pair of block_start() / block_end() calls. This permits the prefix pieces to introduce new items into the lexical scope of the block - for example by the use of "XPK_LEXVAR_MY".

A call to intro_my() is automatically made at the end of the prefix pieces, before the block itself is parsed, ensuring any new lexical variables are now visible.

In addition, the following extra piece types are recognised here:

void setup(pTHX_ void *hookdata);
XPK_SETUP(&setup)

atomic, emits nothing.

This piece type runs a function given by pointer. Typically this function may be used to introduce new lexical state into the parser, or in some other way have some side-effect on the parsing context of the block to be parsed.

A variant of "XPK_PREFIXED_BLOCK" which additionally wraps the entire parsing operation, including the block_start(), block_end() and any calls to "XPK_SETUP" functions, within a "ENTER"/"LEAVE" pair.

This should not make a difference to the standard parser pieces provided here, but may be useful behaviour for the code in the setup function, especially if it wishes to modify parser state and use the savestack to ensure it is restored again when parsing has finished.

atomic, emits cv.

A brace-delimited block of code is expected, and assembled into the body of a new anonymous subroutine. This will be passed as a protosub CV in the cv field.

XPK_STAGED_ANONSUB(stages ...)

structural, emits cv.

A variant of "XPK_ANONSUB" which accepts additional function pointers to be invoked at various points during parsing and compilation. These can be used to interrupt the normal parsing in a manner similar to XS::Parse::Sublike, though currently somewhat less flexibly.

The stages list may contain elements of the following types. Not every stage must be present, but any that are present must be in the following order. Multiple copies of each stage are permitted; they are invoked in the written order, with parser code happening inbetween.

XPK_ANONSUB_PREPARE(&callback)

atomic, emits nothing.

Invokes the callback before start_subparse().

XPK_ANONSUB_START(&callback)

atomic, emits nothing.

Invokes the callback after block_start() but before parsing the actual block contents.

OP *op_wrapper_callback(pTHX_ OP *o, void *hookdata);
XPK_ANONSUB_END(&op_wrapper_callback)

atomic, emits nothing.

Invokes the callback after parsing the block contents but before calling block_end(). The callback may modify the optree if required and return a new one.

XPK_ANONSUB_WRAP(&op_wrapper_callback)

atomic, emits nothing.

Invokes the callback after block_end() but before passing the optree to newATTRSUB(). The callback may modify the optree if required and return a new one.

atomic, emits op.

XPK_ARITHEXPR

An arithmetic expression is expected, parsed using parse_arithexpr(), and passed as an optree in the op field.

Variants of "XPK_ARITHEXPR" which puts the expression in void or scalar context.

atomic, emits op.

XPK_TERMEXPR

A term expression is expected, parsed using parse_termexpr(), and passed as an optree in the op field.

Variants of "XPK_TERMEXPR" which puts the expression in void or scalar context.

atomic, emits op.

XPK_LISTEXPR

A list expression is expected, parsed using parse_listexpr(), and passed as an optree in the op field.

Variant of "XPK_LISTEXPR" which puts the expression in list context.

XPK_PREFIXED_TERMEXPR_ENTERLEAVE(pieces ...)
XPK_PREFIXED_LISTEXPR_ENTERLEAVE(pieces ...)

Variants of "XPK_TERMEXPR" or "XPK_LISTEXPR" which expect a sequence pieces first before it parses a term expression, similar to how "XPK_PREFIXED_BLOCK_ENTERLEAVE" works. The entire operation is wrapped in an "ENTER"/"LEAVE" pair.

These are intended just for use of "XPK_SETUP" pieces as prefixes. Any other pieces which actually parse real input are likely to cause overly-complex, subtle, or outright ambiguous grammars, and should be avoided.

atomic, can probe, emits sv.

A bareword identifier name is expected, and passed as an SV containing a PV in the sv field. An identifier is not permitted to contain a double colon ("::").

atomic, can probe, emits sv.

A bareword package name is expected, and passed as an SV containing a PV in the sv field. A package name is similar to an identifier, except it permits double colons in the middle.

atomic, emits sv.

XPK_LEXVARNAME(kind)

A lexical variable name is expected, and passed as an SV containing a PV in the sv field. The "kind" argument specifies what kinds of variable are permitted, and should be a bitmask of one or more bits from "XPK_LEXVAR_SCALAR", "XPK_LEXVAR_ARRAY" and "XPK_LEXVAR_HASH". A convenient shortcut "XPK_LEXVAR_ANY" permits all three.

atomic, emits i followed by more args.

A list of ":"-prefixed attributes is expected, in the same format as sub or variable attributes. An optional leading ":" indicates the presence of attributes, then one or more of them are parsed. Attributes may be optionally separated by additional ":"s, but this is not required.

Each attribute is expected to be an identifier name, followed by an optional value wrapped in parentheses. Whitespace is NOT permitted between the name and value, as per standard Perl parsing rules.

:attrname
:attrname(value)

The i field indicates how many attributes were found. That number of additional arguments are then passed, each containing two SVs in the attr.name and attr.value fields. This number may be zero.

It is not an error for there to be no attributes present, or for the optional colon to be missing. In this case i will be set to zero.

atomic, can probe, emits sv.

A version string is expected, of the form "v1.234" including the leading "v" character. It is passed as a version SV object in the sv field.

atomic, emits padix.

XPK_LEXVAR(kind)

A lexical variable name is expected and looked up from the current pad. The resulting pad index is passed in the padix field. No error happens if the variable is not found; the value "NOT_IN_PAD" is passed instead.

The "kind" argument specifies what kinds of variable are permitted, as per "XPK_LEXVARNAME".

atomic, emits padix.

XPK_LEXVAR_MY(kind)

A lexical variable name is expected, added to the current pad as if specified in a "my" expression, and passed as the pad index in the padix field.

The "kind" argument specifies what kinds of variable are permitted, as per "XPK_LEXVARNAME".

atomic, can probe, emits nothing.

A literal character (",", ":" or "=") is expected. No argument value is passed.

atomic, emits nothing.

A literal semicolon (";") as a statement terminator is optionally expected. If the next token is a closing brace to indicate the end of a block, then a semicolon is not required. If anything else is encountered an error will be raised.

This piece type is the same as specifying the "XPK_FLAG_AUTOSEMI". It is useful to put at the end of a sequence that forms part of a choice of syntax, where some forms indicate a statement ending in a semicolon, whereas others may end in a full block that does not need one.

atomic, can probe, emits infix.

An infix operator as recognised by XS::Parse::Infix. The returned pointer points to a structure allocated by "XS::Parse::Infix" describing the operator.

Various versions of the macro are provided, each using a different selection filter to choose certain available infix operators:

XPK_INFIX_RELATION         # any relational operator
XPK_INFIX_EQUALITY         # an equality operator like `==` or `eq`
XPK_INFIX_MATCH_NOSMART    # any sort of "match"-like operator, except smartmatch
XPK_INFIX_MATCH_SMART      # XPK_INFIX_MATCH_NOSMART plus smartmatch

atomic, can probe, emits nothing.

XPK_LITERAL("literal")

A literal string match is expected. No argument value is passed.

This form should generally be avoided if at all possible, because it is very easy to abuse to make syntaxes which confuse humans and code tools alike. Generally it is best reserved just for the first component of a "XPK_OPTIONAL" or "XPK_REPEATED" sequence, to provide a "secondary keyword" that such a repeated item can look out for.

atomic, can probe, emits nothing.

XPK_KEYWORD("keyword")

A literal string match is expected. No argument value is passed.

This is similar to "XPK_LITERAL" except that it additionally checks that the following character is not an identifier character. This ensures that the expected keyword-like behaviour is preserved. For example, given the input "keyword", the piece XPK_LITERAL("key") would match it, whereas XPK_KEYWORD("key") would not because of the subsequent "w" character.

atomic, emits nothing.

Calls the core perl intro_my() function immediately. No input is consumed and no output value is generated. This is often useful after "XPK_LEXVAR_MY".

atomic, emits nothing.

XPK_WARNING("message here")

Emits a warning by calling the core perl warn() function on the given string literal. This is equivalent to simply calling warn() from the build function, except that it is emitted immediately at parse time, so line numbering will be more accurate. Also, by placing it as part of an optional or choice sequence, the warning will only be emitted conditionally if that part of the grammar structure is encountered.

Several variants of "XPK_WARNING" exist that are conditional on particular warning categories being enabled. These are ones that are likely to be useful at parse time:

XPK_WARNING_AMBIGUOUS
XPK_WARNING_DEPRECATED
XPK_WARNING_EXPERIMENTAL
XPK_WARNING_PRECEDENCE
XPK_WARNING_SYNTAX

structural, might support probe, emits nothing.

XPK_SEQUENCE(pieces ...)

A structural type which contains a number of pieces. This is normally equivalent to simply placing the pieces in sequence inside their own container, but it is useful inside "XPK_CHOICE" or "XPK_TAGGEDCHOICE".

An "XPK_SEQUENCE" supports probe if its first contained piece does; i.e. is transparent to probing.

structural, emits i.

XPK_OPTIONAL(pieces ...)

A structural type which may expects to find its contained pieces, or is happy not to. This will pass an argument whose i field contains either 1 or 0, depending whether the contents were found. The first piece type within must support probe.

structural, emits i.

XPK_REPEATED(pieces ...)

A structural type which expects to find zero or more repeats of its contained pieces. This will pass an argument whose i field contains the count of the number of repeats it found. The first piece type within must support probe.

structural, can probe, emits i.

XPK_CHOICE(options ...)

A structural type which expects to find one of a number of alternative options. An ordered list of types is provided, all of which must support probe. This will pass an argument whose i field gives the index of the first choice that was accepted. The first option takes the value 0.

As each of the options is interpreted as an alternative, not a sequence, you should use "XPK_SEQUENCE" if a sequence of multiple items should be considered as a single alternative.

It is not an error if no choice matches. At that point, the i field will be set to -1.

If you require a failure message in this case, set the final choice to be of type "XPK_FAILURE". This will cause an error message to be printed instead.

XPK_FAILURE("message string")

structural, can probe, emits i.

XPK_TAGGEDCHOICE(choice, tag, ...)

A structural type similar to "XPK_CHOICE", except that each choice type is followed by an element of type "XPK_TAG" which gives an integer. It is that integer value, rather than the positional index of the choice within the list, which is passed in the i field.

XPK_TAG(value)

As each of the options is interpreted as an alternative, not a sequence, you should use "XPK_SEQUENCE" if a sequence of multiple items should be considered as a single alternative.

structural, might support probe, emits i.

XPK_COMMALIST(pieces ...)

A structural type which expects to find one or more repeats of its contained pieces, separated by literal comma (",") characters. This is somewhat similar to "XPK_REPEATED", except that it needs at least one copy, needs commas between its items, but does not require that the first contained piece support probe (the comma itself is sufficient to indicate a repeat).

An "XPK_COMMALIST" supports probe if its first contained piece does; i.e. is transparent to probing.

structural, can probe, emits nothing.

XPK_PARENS(pieces ...)

A structural type which expects to find a sequence of pieces, all contained in parentheses as "( ... )". This will pass no extra arguments.

structural, emits nothing.

XPK_ARGS(pieces ...)

A structural type similar to "XPK_PARENS", except that the parentheses themselves are optional; much like Perl's parsing of calls to known functions.

If parentheses are encountered in the input, they will be consumed by this piece and it will behave identically to "XPK_PARENS". If there is no open parenthesis, this piece will behave like "XPK_SEQUENCE" and consume all the pieces inside it, without expecting a closing parenthesis.

structural, can probe, emits nothing.

XPK_BRACKETS(pieces ...)

A structural type which expects to find a sequence of pieces, all contained in square brackets as "[ ... ]". This will pass no extra arguments.

structural, can probe, emits nothing.

XPK_BRACES(pieces ...)

A structural type which expects to find a sequence of pieces, all contained in braces as "{ ... }". This will pass no extra arguments.

Note that this is not necessary to use with "XPK_BLOCK" or "XPK_ANONSUB"; those will already consume a set of braces. This is intended for special constrained syntax that should not just accept an arbitrary block.

structural, can probe, emits nothing.

XPK_CHEVRONS(pieces ...)

A structural type which expects to find a sequence of pieces, all contained in angle brackets as "< ... >". This will pass no extra arguments.

Remember that expressions like "a > b" are valid term expressions, so the contents of this scope shouldn't allow arbitrary expressions or the closing bracket will be ambiguous.

structural, can probe, emits i.

XPK_PARENS_OPT(pieces ...)
XPK_BRACKETS_OPT(pieces ...)
XPK_BRACES_OPT(pieces ...)
XPK_CHEVERONS_OPT(pieces ...)

Each of the four contained structure macros above has an optional variant, whose name is suffixed by "_OPT". These pass an argument whose i field is either true or false, indicating whether the scope was found, followed by the values from the scope itself.

This is a convenient shortcut to nesting the scope within a "XPK_OPTIONAL" macro.

XPK_SEQUENCE_pieces(ptr)
XPK_OPTIONAL_pieces(ptr)
...

For each of the "XPK_..." macros that takes a variable-length list of pieces, there is a variant whose name ends with "..._pieces", taking a single pointer argument directly. This must point at a "const XSParseKeywordPieceType []" array whose final element is the zero element.

Normally hand-written C code of a fixed grammar would be unlikely to use these forms, but they may be useful in dynamically-generated cases.

Paul Evans <leonerd@leonerd.org.uk>

2024-09-13 perl v5.40.0