-*- outline -*- * URGENT: Prologue The %union is declared after the user C declarations. It can be 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 have: %{ ... #include "gettextP.h" ... %} %union { unsigned long int num; enum operator op; struct expression *exp; } %{ ... static int yylex PARAMS ((YYSTYPE *lval, const char **pexp)); ... %} 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. * 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 ... > I'm in favor of > > %token FOO 256 > %token BAR 257 > > and Bison moves error into 258. Yes, I think that's a valid extension too, if the user doesn't define the token number for error. * 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. [] * Broken options ? ** %no-lines [ok] ** %no-parser [] ** %pure-parser [] ** %semantic-parser [] ** %token-table [] ** Options which could use parse_dquoted_param (). Maybe transfered in lex.c. *** %skeleton [ok] *** %output [] *** %file-prefix [] *** %name-prefix [] ** Skeleton strategy. [] Must we keep %no-parser? %token-table? *** New skeletons. [] * src/print_graph.c Find the best graph parameters. [] * doc/bison.texinfo ** Update informations about ERROR_VERBOSE. [] ** Add explainations about skeleton muscles. [] %skeleton. [] * 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.