X-Git-Url: https://git.saurik.com/bison.git/blobdiff_plain/704a47c475e0c9abc2ba8624de6b3e360abd1ad1..b98ec53ee42edc8c709505472faca84d1cd4b4c3:/TODO?ds=sidebyside diff --git a/TODO b/TODO index 80a7cf8c..43405335 100644 --- a/TODO +++ b/TODO @@ -1,8 +1,9 @@ -*- outline -*- -* Prologue + +* URGENT: Prologue The %union is declared after the user C declarations. It can be -a problem if YYSTYPE is declared after the user part. [] +a problem if YYSTYPE is declared after the user part. Actually, the real problem seems that the %union ought to be output where it was defined. For instance, in gettext/intl/plural.y, we @@ -29,6 +30,222 @@ have: Where the first part defines struct expression, the second uses it to define YYSTYPE, and the last uses YYSTYPE. Only this order is valid. +Note that we have the same problem with GCC. + +I suggest splitting the prologue into pre-prologue and post-prologue. +The reason is that: + +1. we keep language independance as it is the skeleton that joins the +two prologues (there is no need for the engine to encode union yystype +and to output it inside the prologue, which breaks the language +independance of the generator) + +2. that makes it possible to have several %union in input. I think +this is a pleasant (but useless currently) feature, but in the future, +I want a means to %include other bits of grammars, and _then_ it will +be important for the various bits to define their needs in %union. + +When implementing multiple-%union support, bare the following in mind: + +- when --yacc, this must be flagged as an error. Don't make it fatal + though. + +- The #line must now appear *inside* the definition of yystype. + Something like + + { + #line 12 "foo.y" + int ival; + #line 23 "foo.y" + char *sval; + } + +* Language independent actions + +Currently bison, the generator, transforms $1, $$ and so forth into +direct C code, manipulating the stacks. This is problematic, because +(i) it means that if we want more languages, we need to update the +generator, and (ii), it forces names everywhere (e.g., the C++ +skeleton would be happy to use other naming schemes, and actually, +even other accessing schemes). + +Therefore we want + +1. the generator to replace $1, etc. by M4 macro invocations + (b4_dollar(1), b4_at(3), b4_dollar_dollar) etc. + +2. the skeletons to define these macros. + +But currently the actions are double-quoted, to protect them from M4 +evaluation. So we need to: + +3. stop quoting them + +4. change the [ and ] in the actions into @<:@ and @:>@ + +5. extend the postprocessor to maps these back onto [ and ]. + +* Coding system independence +Paul notes: + + Currently Bison assumes 8-bit bytes (i.e. that UCHAR_MAX is + 255). It also assumes that the 8-bit character encoding is + the same for the invocation of 'bison' as it is for the + invocation of 'cc', but this is not necessarily true when + people run bison on an ASCII host and then use cc on an EBCDIC + host. I don't think these topics are worth our time + addressing (unless we find a gung-ho volunteer for EBCDIC or + PDP-10 ports :-) but they should probably be documented + somewhere. + +* Using enums instead of int for tokens. +Paul suggests: + + #ifndef YYTOKENTYPE + # if defined (__STDC__) || defined (__cplusplus) + /* Put the tokens into the symbol table, so that GDB and other debuggers + know about them. */ + enum yytokentype { + FOO = 256, + BAR, + ... + }; + /* POSIX requires `int' for tokens in interfaces. */ + # define YYTOKENTYPE int + # endif + #endif + #define FOO 256 + #define BAR 257 + ... + +* Output directory +Akim: + +| I consider this to be a bug in bison: +| +| /tmp % mkdir src +| /tmp % cp ~/src/bison/tests/calc.y src +| /tmp % mkdir build && cd build +| /tmp/build % bison ../src/calc.y +| /tmp/build % cd .. +| /tmp % ls -l build src +| build: +| total 0 +| +| src: +| total 32 +| -rw-r--r-- 1 akim lrde 27553 oct 2 16:31 calc.tab.c +| -rw-r--r-- 1 akim lrde 3335 oct 2 16:31 calc.y +| +| +| Would it be safe to change this behavior to something more reasonable? +| Do you think some people depend upon this? + +Jim: + +Is it that behavior documented? +If so, then it's probably not reasonable to change it. +I've Cc'd the automake list, because some of automake's +rules use bison through $(YACC) -- though I'll bet they +all use it in yacc-compatible mode. + +Pavel: + +Hello, Jim and others! + +> Is it that behavior documented? +> If so, then it's probably not reasonable to change it. +> I've Cc'd the automake list, because some of automake's +> rules use bison through $(YACC) -- though I'll bet they +> all use it in yacc-compatible mode. + +Yes, Automake currently used bison in Automake-compatible mode, but it +would be fair for Automake to switch to the native mode as long as the +processed files are distributed and "missing" emulates bison. + +In any case, the makefiles should specify the output file explicitly +instead of relying on weird defaults. + +> | src: +> | total 32 +> | -rw-r--r-- 1 akim lrde 27553 oct 2 16:31 calc.tab.c +> | -rw-r--r-- 1 akim lrde 3335 oct 2 16:31 calc.y + +This is not _that_ ugly as it seems - with Automake you want to put +sources where they belong - to the source directory. + +> | This is not _that_ ugly as it seems - with Automake you want to put +> | sources where they belong - to the source directory. +> +> The difference source/build you are referring to is based on Automake +> concepts. They have no sense at all for tools such as bison or gcc +> etc. They have input and output. I do not want them to try to grasp +> source/build. I want them to behave uniformly: output *here*. + +I realize that. + +It's unfortunate that the native mode of Bison behaves in a less uniform +way than the yacc mode. I agree with your point. Bison maintainters may +want to fix it along with the documentation. + + +* Unit rules +Maybe we could expand unit rules, i.e., transform + + exp: arith | bool; + arith: exp '+' exp; + bool: exp '&' exp; + +into + + exp: exp '+' exp | exp '&' exp; + +when there are no actions. This can significantly speed up some +grammars. + +* Stupid error messages +An example shows it easily: + +src/bison/tests % ./testsuite -k calc,location,error-verbose -l +GNU Bison 1.49a test suite test groups: + + NUM: FILENAME:LINE TEST-GROUP-NAME + KEYWORDS + + 51: calc.at:440 Calculator --locations --yyerror-verbose + 52: calc.at:442 Calculator --defines --locations --name-prefix=calc --verbose --yacc --yyerror-verbose + 54: calc.at:445 Calculator --debug --defines --locations --name-prefix=calc --verbose --yacc --yyerror-verbose +src/bison/tests % ./testsuite 51 -d +## --------------------------- ## +## GNU Bison 1.49a test suite. ## +## --------------------------- ## + 51: calc.at:440 ok +## ---------------------------- ## +## All 1 tests were successful. ## +## ---------------------------- ## +src/bison/tests % cd ./testsuite.dir/51 +tests/testsuite.dir/51 % echo "()" | ./calc +1.2-1.3: parse error, unexpected ')', expecting error or "number" or '-' or '(' + +* yyerror, yyprint interface +It should be improved, in particular when using Bison features such as +locations, and YYPARSE_PARAMS. For the time being, it is recommended +to #define yyerror and yyprint to steal internal variables... + +* read_pipe.c +This is not portable to DOS for instance. Implement a more portable +scheme. Sources of inspiration include GNU diff, and Free Recode. + +* Memory leaks in the generator +A round of memory leak clean ups would be most welcome. Dmalloc, +Checker GCC, Electric Fence, or Valgrind: you chose your tool. + +* Memory leaks in the parser +The same applies to the generated parsers. In particular, this is +critical for user data: when aborting a parsing, when handling the +error token etc., we often throw away yylval without giving a chance +of cleaning it up to the user. + * --graph Show reductions. [] @@ -63,3 +280,197 @@ skeleton muscles. [] * testsuite ** tests/pure-parser.at [] New tests. + +* Debugging parsers + +From Greg McGary: + +akim demaille writes: + +> With great pleasure! Nonetheless, things which are debatable +> (or not, but just `big') should be discuss in `public': something +> like help- or bug-bison@gnu.org is just fine. Jesse and I are there, +> but there is also Jim and some other people. + +I have no idea whether it qualifies as big or controversial, so I'll +just summarize for you. I proposed this change years ago and was +surprised that it was met with utter indifference! + +This debug feature is for the programs/grammars one develops with +bison, not for debugging bison itself. I find that the YYDEBUG +output comes in a very inconvenient format for my purposes. +When debugging gcc, for instance, what I want is to see a trace of +the sequence of reductions and the line#s for the semantic actions +so I can follow what's happening. Single-step in gdb doesn't cut it +because to move from one semantic action to the next takes you through +lots of internal machinery of the parser, which is uninteresting. + +The change I made was to the format of the debug output, so that it +comes out in the format of C error messages, digestible by emacs +compile mode, like so: + +grammar.y:1234: foo: bar(0x123456) baz(0x345678) + +where "foo: bar baz" is the reduction rule, whose semantic action +appears on line 1234 of the bison grammar file grammar.y. The hex +numbers on the rhs tokens are the parse-stack values associated with +those tokens. Of course, yytype might be something totally +incompatible with that representation, but for the most part, yytype +values are single words (scalars or pointers). In the case of gcc, +they're most often pointers to tree nodes. Come to think of it, the +right thing to do is to make the printing of stack values be +user-definable. It would also be useful to include the filename & +line# of the file being parsed, but the main filename & line# should +continue to be that of grammar.y + +Anyway, this feature has saved my life on numerous occasions. The way +I customarily use it is to first run bison with the traces on, isolate +the sequence of reductions that interests me, put those traces in a +buffer and force it into compile-mode, then visit each of those lines +in the grammar and set breakpoints with C-x SPACE. Then, I can run +again under the control of gdb and stop at each semantic action. +With the hex addresses of tree nodes, I can inspect the values +associated with any rhs token. + +You like? + +* input synclines +Some users create their foo.y files, and equip them with #line. Bison +should recognize these, and preserve them. + +* BTYacc +See if we can integrate backtracking in Bison. Contact the BTYacc +maintainers. + +* Automaton report +Display more clearly the lookaheads for each item. + +* RR conflicts +See if we can use precedence between rules to solve RR conflicts. See +what POSIX says. + +* Precedence +It is unfortunate that there is a total order for precedence. It +makes it impossible to have modular precedence information. We should +move to partial orders. + +* Parsing grammars +Rewrite the reader in Bison. + +* Problems with aliases +From: "Baum, Nathan I" +Subject: Token Alias Bug +To: "'bug-bison@gnu.org'" + +I've noticed a bug in bison. Sadly, our eternally wise sysadmins won't let +us use CVS, so I can't find out if it's been fixed already... + +Basically, I made a program (in flex) that went through a .y file looking +for "..."-tokens, and then outputed a %token +line for it. For single-character ""-tokens, I reasoned, I could just use +[%token 'A' "A"]. However, this causes Bison to output a [#define 'A' 65], +which cppp chokes on, not unreasonably. (And even if cppp didn't choke, I +obviously wouldn't want (char)'A' to be replaced with (int)65 throughout my +code. + +Bison normally forgoes outputing a #define for a character token. However, +it always outputs an aliased token -- even if the token is an alias for a +character token. We don't want that. The problem is in /output.c/, as I +recall. When it outputs the token definitions, it checks for a character +token, and then checks for an alias token. If the character token check is +placed after the alias check, then it works correctly. + +Alias tokens seem to be something of a kludge. What about an [%alias "..."] +command... + + %alias T_IF "IF" + +Hmm. I can't help thinking... What about a --generate-lex option that +creates an .l file for the alias tokens used... (Or an option to make a +gperf file, etc...) + +* Presentation of the report file +From: "Baum, Nathan I" +Subject: Token Alias Bug +To: "'bug-bison@gnu.org'" + +I've also noticed something, that whilst not *wrong*, is inconvienient: I +use the verbose mode to help find the causes of unresolved shift/reduce +conflicts. However, this mode insists on starting the .output file with a +list of *resolved* conflicts, something I find quite useless. Might it be +possible to define a -v mode, and a -vv mode -- Where the -vv mode shows +everything, but the -v mode only tells you what you need for examining +conflicts? (Or, perhaps, a "*** This state has N conflicts ***" marker above +each state with conflicts.) + + +* $undefined +From Hans: +- If the Bison generated parser experiences an undefined number in the +character range, that character is written out in diagnostic messages, an +addition to the $undefined value. + +Suggest: Change the name $undefined to undefined; looks better in outputs. + +* Default Action +From Hans: +- For use with my C++ parser, I transported the "switch (yyn)" statement +that Bison writes to the bison.simple skeleton file. This way, I can remove +the current default rule $$ = $1 implementation, which causes a double +assignment to $$ which may not be OK under C++, replacing it with a +"default:" part within the switch statement. + +Note that the default rule $$ = $1, when typed, is perfectly OK under C, +but in the C++ implementation I made, this rule is different from +$$ = $1. I therefore think that one should implement +a Bison option where every typed default rule is explicitly written out +(same typed ruled can of course be grouped together). + +* Pre and post actions. +From: Florian Krohm +Subject: YYACT_EPILOGUE +To: bug-bison@gnu.org +X-Sent: 1 week, 4 days, 14 hours, 38 minutes, 11 seconds ago + +The other day I had the need for explicitly building the parse tree. I +used %locations for that and defined YYLLOC_DEFAULT to call a function +that returns the tree node for the production. Easy. But I also needed +to assign the S-attribute to the tree node. That cannot be done in +YYLLOC_DEFAULT, because it is invoked before the action is executed. +The way I solved this was to define a macro YYACT_EPILOGUE that would +be invoked after the action. For reasons of symmetry I also added +YYACT_PROLOGUE. Although I had no use for that I can envision how it +might come in handy for debugging purposes. +All is needed is to add + +#if YYLSP_NEEDED + YYACT_EPILOGUE (yyval, (yyvsp - yylen), yylen, yyloc, (yylsp - yylen)); +#else + YYACT_EPILOGUE (yyval, (yyvsp - yylen), yylen); +#endif + +at the proper place to bison.simple. Ditto for YYACT_PROLOGUE. + +I was wondering what you think about adding YYACT_PROLOGUE/EPILOGUE +to bison. If you're interested, I'll work on a patch. + +----- + +Copyright (C) 2001, 2002 Free Software Foundation, Inc. + +This file is part of GNU Autoconf. + +GNU Autoconf is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 2, or (at your option) +any later version. + +GNU Autoconf is distributed in the hope that it will be useful, +but WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +GNU General Public License for more details. + +You should have received a copy of the GNU General Public License +along with autoconf; see the file COPYING. If not, write to +the Free Software Foundation, Inc., 59 Temple Place - Suite 330, +Boston, MA 02111-1307, USA.