1 ----------------------------------------------------------------------
\r
4 -- Summary: parser generator. Collection of higher order functors,
\r
5 -- which allow to build and combine parsers. Relies on a lexer
\r
6 -- that supports the same API as the one exposed in mll.lua.
\r
8 ----------------------------------------------------------------------
\r
10 -- Copyright (c) 2006-2008, Fabien Fleutot <metalua@gmail.com>.
\r
12 -- This software is released under the MIT Licence, see licence.txt
\r
15 ----------------------------------------------------------------------
\r
17 --------------------------------------------------------------------------------
\r
21 -- Parser generators:
\r
22 -- * [gg.sequence()]
\r
23 -- * [gg.multisequence()]
\r
26 -- * [gg.onkeyword()]
\r
27 -- * [gg.optkeyword()]
\r
29 -- Other functions:
\r
30 -- * [gg.parse_error()]
\r
31 -- * [gg.make_parser()]
\r
32 -- * [gg.is_parser()]
\r
34 --------------------------------------------------------------------------------
\r
36 module("gg", package.seeall)
\r
38 -------------------------------------------------------------------------------
\r
39 -- parser metatable, which maps __call to method parse, and adds some
\r
40 -- error tracing boilerplate.
\r
41 -------------------------------------------------------------------------------
\r
42 local parser_metatable = { }
\r
43 function parser_metatable.__call (parser, lx, ...)
\r
44 --printf ("Call parser %q of type %q", parser.name or "?", parser.kind)
\r
45 if mlc.metabugs then
\r
46 return parser:parse (lx, ...)
\r
47 --local x = parser:parse (lx, ...)
\r
48 --printf ("Result of parser %q: %s",
\r
49 -- parser.name or "?",
\r
50 -- _G.table.tostring(x, "nohash", 80))
\r
53 local li = lx:lineinfo_right() or { "?", "?", "?", "?" }
\r
54 local status, ast = pcall (parser.parse, parser, lx, ...)
\r
55 if status then return ast else
\r
56 error (string.format ("%s\n - (l.%s, c.%s, k.%s) in parser %s",
\r
57 ast:strmatch "gg.lua:%d+: (.*)" or ast,
\r
58 li[1], li[2], li[3], parser.name or parser.kind))
\r
63 -------------------------------------------------------------------------------
\r
64 -- Turn a table into a parser, mainly by setting the metatable.
\r
65 -------------------------------------------------------------------------------
\r
66 function make_parser(kind, p)
\r
68 if not p.transformers then p.transformers = { } end
\r
69 function p.transformers:add (x)
\r
70 table.insert (self, x)
\r
72 setmetatable (p, parser_metatable)
\r
76 -------------------------------------------------------------------------------
\r
77 -- Return true iff [x] is a parser.
\r
78 -- If it's a gg-generated parser, reutrn the name of its kind.
\r
79 -------------------------------------------------------------------------------
\r
80 function is_parser (x)
\r
81 return type(x)=="function" or getmetatable(x)==parser_metatable and x.kind
\r
84 -------------------------------------------------------------------------------
\r
85 -- Parse a sequence, without applying builder nor transformers
\r
86 -------------------------------------------------------------------------------
\r
87 local function raw_parse_sequence (lx, p)
\r
91 if type(e) == "string" then
\r
92 if not lx:is_keyword (lx:next(), e) then
\r
93 parse_error (lx, "Keyword '%s' expected", e) end
\r
94 elseif is_parser (e) then
\r
95 table.insert (r, e (lx))
\r
97 gg.parse_error (lx,"Sequence `%s': element #%i is not a string "..
\r
98 "nor a parser: %s",
\r
99 p.name, i, table.tostring(e))
\r
105 -------------------------------------------------------------------------------
\r
106 -- Parse a multisequence, without applying multisequence transformers.
\r
107 -- The sequences are completely parsed.
\r
108 -------------------------------------------------------------------------------
\r
109 local function raw_parse_multisequence (lx, sequence_table, default)
\r
110 local seq_parser = sequence_table[lx:is_keyword(lx:peek())]
\r
111 if seq_parser then return seq_parser (lx)
\r
112 elseif default then return default (lx)
\r
113 else return false end
\r
116 -------------------------------------------------------------------------------
\r
117 -- Applies all transformers listed in parser on ast.
\r
118 -------------------------------------------------------------------------------
\r
119 local function transform (ast, parser, fli, lli)
\r
120 if parser.transformers then
\r
121 for _, t in ipairs (parser.transformers) do ast = t(ast) or ast end
\r
123 if type(ast) == 'table'then
\r
124 local ali = ast.lineinfo
\r
125 if not ali or ali.first~=fli or ali.last~=lli then
\r
126 ast.lineinfo = { first = fli, last = lli }
\r
132 -------------------------------------------------------------------------------
\r
133 -- Generate a tracable parsing error (not implemented yet)
\r
134 -------------------------------------------------------------------------------
\r
135 function parse_error(lx, fmt, ...)
\r
136 local li = lx:lineinfo_left() or {-1,-1,-1, "<unknown file>"}
\r
137 local msg = string.format("line %i, char %i: "..fmt, li[1], li[2], ...)
\r
139 if li[3]>0 and src then
\r
140 local i, j = li[3], li[3]
\r
141 while src:sub(i,i) ~= '\n' and i>=0 do i=i-1 end
\r
142 while src:sub(j,j) ~= '\n' and j<=#src do j=j+1 end
\r
143 local srcline = src:sub (i+1, j-1)
\r
144 local idx = string.rep (" ", li[2]).."^"
\r
145 msg = string.format("%s\n>>> %s\n>>> %s", msg, srcline, idx)
\r
150 -------------------------------------------------------------------------------
\r
152 -- Sequence parser generator
\r
154 -------------------------------------------------------------------------------
\r
157 -- * [builder]: how to build an AST out of sequence parts. let [x] be the list
\r
158 -- of subparser results (keywords are simply omitted). [builder] can be:
\r
159 -- - [nil], in which case the result of parsing is simply [x]
\r
160 -- - a string, which is then put as a tag on [x]
\r
161 -- - a function, which takes [x] as a parameter and returns an AST.
\r
163 -- * [name]: the name of the parser. Used for debug messages
\r
165 -- * [transformers]: a list of AST->AST functions, applied in order on ASTs
\r
166 -- returned by the parser.
\r
168 -- * Table-part entries corresponds to keywords (strings) and subparsers
\r
169 -- (function and callable objects).
\r
171 -- After creation, the following fields are added:
\r
172 -- * [parse] the parsing function lexer->AST
\r
173 -- * [kind] == "sequence"
\r
174 -- * [name] is set, if it wasn't in the input.
\r
176 -------------------------------------------------------------------------------
\r
177 function sequence (p)
\r
178 make_parser ("sequence", p)
\r
180 -------------------------------------------------------------------
\r
182 -------------------------------------------------------------------
\r
183 function p:parse (lx)
\r
185 local fli = lx:lineinfo_right()
\r
186 local seq = raw_parse_sequence (lx, self)
\r
187 local lli = lx:lineinfo_left()
\r
189 -- Builder application:
\r
190 local builder, tb = self.builder, type (self.builder)
\r
191 if tb == "string" then seq.tag = builder
\r
192 elseif tb == "function" or builder and builder.__call then seq = builder(seq)
\r
193 elseif builder == nil then -- nothing
\r
194 else error ("Invalid builder of type "..tb.." in sequence") end
\r
195 seq = transform (seq, self, fli, lli)
\r
196 assert (not seq or seq.lineinfo)
\r
200 -------------------------------------------------------------------
\r
202 -------------------------------------------------------------------
\r
203 -- Try to build a proper name
\r
204 if not p.name and type(p[1])=="string" then
\r
205 p.name = p[1].." ..."
\r
206 if type(p[#p])=="string" then p.name = p.name .. " " .. p[#p] end
\r
208 p.name = "<anonymous>"
\r
215 -------------------------------------------------------------------------------
\r
217 -- Multiple, keyword-driven, sequence parser generator
\r
219 -------------------------------------------------------------------------------
\r
220 -- in [p], useful fields are:
\r
222 -- * [transformers]: as usual
\r
224 -- * [name]: as usual
\r
226 -- * Table-part entries must be sequence parsers, or tables which can
\r
227 -- be turned into a sequence parser by [gg.sequence]. These
\r
228 -- sequences must start with a keyword, and this initial keyword
\r
229 -- must be different for each sequence. The table-part entries will
\r
230 -- be removed after [gg.multisequence] returns.
\r
232 -- * [default]: the parser to run if the next keyword in the lexer is
\r
233 -- none of the registered initial keywords. If there's no default
\r
234 -- parser and no suitable initial keyword, the multisequence parser
\r
235 -- simply returns [false].
\r
237 -- After creation, the following fields are added:
\r
239 -- * [parse] the parsing function lexer->AST
\r
241 -- * [sequences] the table of sequences, indexed by initial keywords.
\r
243 -- * [add] method takes a sequence parser or a config table for
\r
244 -- [gg.sequence], and adds/replaces the corresponding sequence
\r
245 -- parser. If the keyword was already used, the former sequence is
\r
246 -- removed and a warning is issued.
\r
248 -- * [get] method returns a sequence by its initial keyword
\r
250 -- * [kind] == "multisequence"
\r
252 -------------------------------------------------------------------------------
\r
253 function multisequence (p)
\r
254 make_parser ("multisequence", p)
\r
256 -------------------------------------------------------------------
\r
257 -- Add a sequence (might be just a config table for [gg.sequence])
\r
258 -------------------------------------------------------------------
\r
260 -- compile if necessary:
\r
261 if not is_parser(s) then sequence(s) end
\r
262 if type(s[1]) ~= "string" then
\r
263 error "Invalid sequence for multiseq"
\r
264 elseif self.sequences[s[1]] then
\r
265 eprintf (" *** Warning: keyword %q overloaded in multisequence ***", s[1])
\r
267 self.sequences[s[1]] = s
\r
268 end -- </multisequence.add>
\r
270 -------------------------------------------------------------------
\r
271 -- Get the sequence starting with this keyword. [kw :: string]
\r
272 -------------------------------------------------------------------
\r
273 function p:get (kw) return self.sequences [kw] end
\r
275 -------------------------------------------------------------------
\r
276 -- Remove the sequence starting with keyword [kw :: string]
\r
277 -------------------------------------------------------------------
\r
278 function p:del (kw)
\r
279 if not self.sequences[kw] then
\r
280 eprintf("*** Warning: trying to delete sequence starting "..
\r
281 "with %q from a multisequence having no such "..
\r
282 "entry ***", kw) end
\r
283 local removed = self.sequences[kw]
\r
284 self.sequences[kw] = nil
\r
288 -------------------------------------------------------------------
\r
290 -------------------------------------------------------------------
\r
291 function p:parse (lx)
\r
292 local fli = lx:lineinfo_right()
\r
293 local x = raw_parse_multisequence (lx, self.sequences, self.default)
\r
294 local lli = lx:lineinfo_left()
\r
295 return transform (x, self, fli, lli)
\r
298 -------------------------------------------------------------------
\r
300 -------------------------------------------------------------------
\r
301 -- Register the sequences passed to the constructor. They're going
\r
302 -- from the array part of the parser to the hash part of field
\r
305 for i=1, #p do p:add (p[i]); p[i] = nil end
\r
307 -- FIXME: why is this commented out?
\r
308 --if p.default and not is_parser(p.default) then sequence(p.default) end
\r
310 end --</multisequence>
\r
313 -------------------------------------------------------------------------------
\r
315 -- Expression parser generator
\r
317 -------------------------------------------------------------------------------
\r
319 -- Expression configuration relies on three tables: [prefix], [infix]
\r
320 -- and [suffix]. Moreover, the primary parser can be replaced by a
\r
321 -- table: in this case the [primary] table will be passed to
\r
322 -- [gg.multisequence] to create a parser.
\r
324 -- Each of these tables is a modified multisequence parser: the
\r
325 -- differences with respect to regular multisequence config tables are:
\r
327 -- * the builder takes specific parameters:
\r
328 -- - for [prefix], it takes the result of the prefix sequence parser,
\r
329 -- and the prefixed expression
\r
330 -- - for [infix], it takes the left-hand-side expression, the results
\r
331 -- of the infix sequence parser, and the right-hand-side expression.
\r
332 -- - for [suffix], it takes the suffixed expression, and theresult
\r
333 -- of the suffix sequence parser.
\r
335 -- * the default field is a list, with parameters:
\r
336 -- - [parser] the raw parsing function
\r
337 -- - [transformers], as usual
\r
338 -- - [prec], the operator's precedence
\r
339 -- - [assoc] for [infix] table, the operator's associativity, which
\r
340 -- can be "left", "right" or "flat" (default to left)
\r
342 -- In [p], useful fields are:
\r
343 -- * [transformers]: as usual
\r
344 -- * [name]: as usual
\r
345 -- * [primary]: the atomic expression parser, or a multisequence config
\r
346 -- table (mandatory)
\r
347 -- * [prefix]: prefix operators config table, see above.
\r
348 -- * [infix]: infix operators config table, see above.
\r
349 -- * [suffix]: suffix operators config table, see above.
\r
351 -- After creation, these fields are added:
\r
352 -- * [kind] == "expr"
\r
353 -- * [parse] as usual
\r
354 -- * each table is turned into a multisequence, and therefore has an
\r
357 -------------------------------------------------------------------------------
\r
359 make_parser ("expr", p)
\r
361 -------------------------------------------------------------------
\r
363 -- In addition to the lexer, it takes an optional precedence:
\r
364 -- it won't read expressions whose precedence is lower or equal
\r
366 -------------------------------------------------------------------
\r
367 function p:parse (lx, prec)
\r
370 ------------------------------------------------------
\r
371 -- Extract the right parser and the corresponding
\r
372 -- options table, for (pre|in|suff)fix operators.
\r
373 -- Options include prec, assoc, transformers.
\r
374 ------------------------------------------------------
\r
375 local function get_parser_info (tab)
\r
376 local p2 = tab:get (lx:is_keyword (lx:peek()))
\r
377 if p2 then -- keyword-based sequence found
\r
378 local function parser(lx) return raw_parse_sequence(lx, p2) end
\r
380 else -- Got to use the default parser
\r
381 local d = tab.default
\r
382 if d then return d.parse or d.parser, d
\r
383 else return false, false end
\r
387 ------------------------------------------------------
\r
388 -- Look for a prefix sequence. Multiple prefixes are
\r
389 -- handled through the recursive [p.parse] call.
\r
390 -- Notice the double-transform: one for the primary
\r
391 -- expr, and one for the one with the prefix op.
\r
392 ------------------------------------------------------
\r
393 local function handle_prefix ()
\r
394 local fli = lx:lineinfo_right()
\r
395 local p2_func, p2 = get_parser_info (self.prefix)
\r
396 local op = p2_func and p2_func (lx)
\r
397 if op then -- Keyword-based sequence found
\r
398 local ili = lx:lineinfo_right() -- Intermediate LineInfo
\r
399 local e = p2.builder (op, self:parse (lx, p2.prec))
\r
400 local lli = lx:lineinfo_left()
\r
401 return transform (transform (e, p2, ili, lli), self, fli, lli)
\r
402 else -- No prefix found, get a primary expression
\r
403 local e = self.primary(lx)
\r
404 local lli = lx:lineinfo_left()
\r
405 return transform (e, self, fli, lli)
\r
407 end --</expr.parse.handle_prefix>
\r
409 ------------------------------------------------------
\r
410 -- Look for an infix sequence+right-hand-side operand.
\r
411 -- Return the whole binary expression result,
\r
412 -- or false if no operator was found.
\r
413 ------------------------------------------------------
\r
414 local function handle_infix (e)
\r
415 local p2_func, p2 = get_parser_info (self.infix)
\r
416 if not p2 then return false end
\r
418 -----------------------------------------
\r
419 -- Handle flattening operators: gather all operands
\r
420 -- of the series in [list]; when a different operator
\r
421 -- is found, stop, build from [list], [transform] and
\r
423 -----------------------------------------
\r
424 if (not p2.prec or p2.prec>prec) and p2.assoc=="flat" then
\r
425 local fli = lx:lineinfo_right()
\r
426 local pflat, list = p2, { e }
\r
428 local op = p2_func(lx)
\r
429 if not op then break end
\r
430 table.insert (list, self:parse (lx, p2.prec))
\r
431 local _ -- We only care about checking that p2==pflat
\r
432 _, p2 = get_parser_info (self.infix)
\r
434 local e2 = pflat.builder (list)
\r
435 local lli = lx:lineinfo_left()
\r
436 return transform (transform (e2, pflat, fli, lli), self, fli, lli)
\r
438 -----------------------------------------
\r
439 -- Handle regular infix operators: [e] the LHS is known,
\r
440 -- just gather the operator and [e2] the RHS.
\r
441 -- Result goes in [e3].
\r
442 -----------------------------------------
\r
443 elseif p2.prec and p2.prec>prec or
\r
444 p2.prec==prec and p2.assoc=="right" then
\r
445 local fli = e.lineinfo.first -- lx:lineinfo_right()
\r
446 local op = p2_func(lx)
\r
447 if not op then return false end
\r
448 local e2 = self:parse (lx, p2.prec)
\r
449 local e3 = p2.builder (e, op, e2)
\r
450 local lli = lx:lineinfo_left()
\r
451 return transform (transform (e3, p2, fli, lli), self, fli, lli)
\r
453 -----------------------------------------
\r
454 -- Check for non-associative operators, and complain if applicable.
\r
455 -----------------------------------------
\r
456 elseif p2.assoc=="none" and p2.prec==prec then
\r
457 parser_error (lx, "non-associative operator!")
\r
459 -----------------------------------------
\r
460 -- No infix operator suitable at that precedence
\r
461 -----------------------------------------
\r
462 else return false end
\r
464 end --</expr.parse.handle_infix>
\r
466 ------------------------------------------------------
\r
467 -- Look for a suffix sequence.
\r
468 -- Return the result of suffix operator on [e],
\r
469 -- or false if no operator was found.
\r
470 ------------------------------------------------------
\r
471 local function handle_suffix (e)
\r
472 -- FIXME bad fli, must take e.lineinfo.first
\r
473 local p2_func, p2 = get_parser_info (self.suffix)
\r
474 if not p2 then return false end
\r
475 if not p2.prec or p2.prec>=prec then
\r
476 --local fli = lx:lineinfo_right()
\r
477 local fli = e.lineinfo.first
\r
478 local op = p2_func(lx)
\r
479 if not op then return false end
\r
480 local lli = lx:lineinfo_left()
\r
481 e = p2.builder (e, op)
\r
482 e = transform (transform (e, p2, fli, lli), self, fli, lli)
\r
486 end --</expr.parse.handle_suffix>
\r
488 ------------------------------------------------------
\r
489 -- Parser body: read suffix and (infix+operand)
\r
490 -- extensions as long as we're able to fetch more at
\r
491 -- this precedence level.
\r
492 ------------------------------------------------------
\r
493 local e = handle_prefix()
\r
495 local x = handle_suffix (e); e = x or e
\r
496 local y = handle_infix (e); e = y or e
\r
499 -- No transform: it already happened in operators handling
\r
501 end --</expr.parse>
\r
503 -------------------------------------------------------------------
\r
505 -------------------------------------------------------------------
\r
506 if not p.primary then p.primary=p[1]; p[1]=nil end
\r
507 for _, t in ipairs{ "primary", "prefix", "infix", "suffix" } do
\r
508 if not p[t] then p[t] = { } end
\r
509 if not is_parser(p[t]) then multisequence(p[t]) end
\r
511 function p:add(...) return self.primary:add(...) end
\r
516 -------------------------------------------------------------------------------
\r
518 -- List parser generator
\r
520 -------------------------------------------------------------------------------
\r
521 -- In [p], the following fields can be provided in input:
\r
523 -- * [builder]: takes list of subparser results, returns AST
\r
524 -- * [transformers]: as usual
\r
525 -- * [name]: as usual
\r
527 -- * [terminators]: list of strings representing the keywords which
\r
528 -- might mark the end of the list. When non-empty, the list is
\r
529 -- allowed to be empty. A string is treated as a single-element
\r
530 -- table, whose element is that string, e.g. ["do"] is the same as
\r
533 -- * [separators]: list of strings representing the keywords which can
\r
534 -- separate elements of the list. When non-empty, one of these
\r
535 -- keyword has to be found between each element. Lack of a separator
\r
536 -- indicates the end of the list. A string is treated as a
\r
537 -- single-element table, whose element is that string, e.g. ["do"]
\r
538 -- is the same as [{"do"}]. If [terminators] is empty/nil, then
\r
539 -- [separators] has to be non-empty.
\r
541 -- After creation, the following fields are added:
\r
542 -- * [parse] the parsing function lexer->AST
\r
543 -- * [kind] == "list"
\r
545 -------------------------------------------------------------------------------
\r
547 make_parser ("list", p)
\r
549 -------------------------------------------------------------------
\r
551 -------------------------------------------------------------------
\r
552 function p:parse (lx)
\r
554 ------------------------------------------------------
\r
555 -- Used to quickly check whether there's a terminator
\r
556 -- or a separator immediately ahead
\r
557 ------------------------------------------------------
\r
558 local function peek_is_in (keywords)
\r
559 return keywords and lx:is_keyword(lx:peek(), unpack(keywords)) end
\r
562 local fli = lx:lineinfo_right()
\r
564 -- if there's a terminator to start with, don't bother trying
\r
565 if not peek_is_in (self.terminators) then
\r
566 repeat table.insert (x, self.primary (lx)) -- read one element
\r
568 -- First reason to stop: There's a separator list specified,
\r
569 -- and next token isn't one. Otherwise, consume it with [lx:next()]
\r
570 self.separators and not(peek_is_in (self.separators) and lx:next()) or
\r
571 -- Other reason to stop: terminator token ahead
\r
572 peek_is_in (self.terminators) or
\r
573 -- Last reason: end of file reached
\r
574 lx:peek().tag=="Eof"
\r
577 local lli = lx:lineinfo_left()
\r
579 -- Apply the builder. It can be a string, or a callable value,
\r
580 -- or simply nothing.
\r
581 local b = self.builder
\r
583 if type(b)=="string" then x.tag = b -- b is a string, use it as a tag
\r
584 elseif type(b)=="function" then x=b(x)
\r
586 local bmt = getmetatable(b)
\r
587 if bmt and bmt.__call then x=b(x) end
\r
590 return transform (x, self, fli, lli)
\r
591 end --</list.parse>
\r
593 -------------------------------------------------------------------
\r
595 -------------------------------------------------------------------
\r
596 if not p.primary then p.primary = p[1]; p[1] = nil end
\r
597 if type(p.terminators) == "string" then p.terminators = { p.terminators }
\r
598 elseif p.terminators and #p.terminators == 0 then p.terminators = nil end
\r
599 if type(p.separators) == "string" then p.separators = { p.separators }
\r
600 elseif p.separators and #p.separators == 0 then p.separators = nil end
\r
606 -------------------------------------------------------------------------------
\r
608 -- Keyword-conditionned parser generator
\r
610 -------------------------------------------------------------------------------
\r
612 -- Only apply a parser if a given keyword is found. The result of
\r
613 -- [gg.onkeyword] parser is the result of the subparser (modulo
\r
614 -- [transformers] applications).
\r
616 -- lineinfo: the keyword is *not* included in the boundaries of the
\r
617 -- resulting lineinfo. A review of all usages of gg.onkeyword() in the
\r
618 -- implementation of metalua has shown that it was the appropriate choice
\r
623 -- * [name]: as usual
\r
625 -- * [transformers]: as usual
\r
627 -- * [peek]: if non-nil, the conditionning keyword is left in the lexeme
\r
628 -- stream instead of being consumed.
\r
630 -- * [primary]: the subparser.
\r
632 -- * [keywords]: list of strings representing triggering keywords.
\r
634 -- * Table-part entries can contain strings, and/or exactly one parser.
\r
635 -- Strings are put in [keywords], and the parser is put in [primary].
\r
637 -- After the call, the following fields will be set:
\r
639 -- * [parse] the parsing method
\r
640 -- * [kind] == "onkeyword"
\r
644 -------------------------------------------------------------------------------
\r
645 function onkeyword (p)
\r
646 make_parser ("onkeyword", p)
\r
648 -------------------------------------------------------------------
\r
650 -------------------------------------------------------------------
\r
651 function p:parse(lx)
\r
652 if lx:is_keyword (lx:peek(), unpack(self.keywords)) then
\r
653 --local fli = lx:lineinfo_right()
\r
654 if not self.peek then lx:next() end
\r
655 local content = self.primary (lx)
\r
656 --local lli = lx:lineinfo_left()
\r
657 local fli, lli = content.lineinfo.first, content.lineinfo.last
\r
658 return transform (content, p, fli, lli)
\r
659 else return false end
\r
662 -------------------------------------------------------------------
\r
664 -------------------------------------------------------------------
\r
665 if not p.keywords then p.keywords = { } end
\r
666 for _, x in ipairs(p) do
\r
667 if type(x)=="string" then table.insert (p.keywords, x)
\r
668 else assert (not p.primary and is_parser (x)); p.primary = x end
\r
670 assert (p.primary, 'no primary parser in gg.onkeyword')
\r
675 -------------------------------------------------------------------------------
\r
677 -- Optional keyword consummer pseudo-parser generator
\r
679 -------------------------------------------------------------------------------
\r
681 -- This doesn't return a real parser, just a function. That function parses
\r
682 -- one of the keywords passed as parameters, and returns it. It returns
\r
683 -- [false] if no matching keyword is found.
\r
685 -- Notice that tokens returned by lexer already carry lineinfo, therefore
\r
686 -- there's no need to add them, as done usually through transform() calls.
\r
687 -------------------------------------------------------------------------------
\r
688 function optkeyword (...)
\r
690 if type (args[1]) == "table" then
\r
691 assert (#args == 1)
\r
694 for _, v in ipairs(args) do assert (type(v)=="string") end
\r
695 return function (lx)
\r
696 local x = lx:is_keyword (lx:peek(), unpack (args))
\r
697 if x then lx:next(); return x
\r
698 else return false end
\r
703 -------------------------------------------------------------------------------
\r
705 -- Run a parser with a special lexer
\r
707 -------------------------------------------------------------------------------
\r
709 -- This doesn't return a real parser, just a function.
\r
710 -- First argument is the lexer class to be used with the parser,
\r
711 -- 2nd is the parser itself.
\r
712 -- The resulting parser returns whatever the argument parser does.
\r
714 -------------------------------------------------------------------------------
\r
715 function with_lexer(new_lexer, parser)
\r
717 -------------------------------------------------------------------
\r
718 -- Most gg functions take their parameters in a table, so it's
\r
719 -- better to silently accept when with_lexer{ } is called with
\r
720 -- its arguments in a list:
\r
721 -------------------------------------------------------------------
\r
722 if not parser and #new_lexer==2 and type(new_lexer[1])=='table' then
\r
723 return with_lexer(unpack(new_lexer))
\r
726 -------------------------------------------------------------------
\r
727 -- Save the current lexer, switch it for the new one, run the parser,
\r
728 -- restore the previous lexer, even if the parser caused an error.
\r
729 -------------------------------------------------------------------
\r
730 return function (lx)
\r
731 local old_lexer = getmetatable(lx)
\r
733 setmetatable(lx, new_lexer)
\r
734 local status, result = pcall(parser, lx)
\r
736 setmetatable(lx, old_lexer)
\r
737 if status then return result else error(result) end
\r