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