Language independent actions.

[bison.git] / TODO

-*- 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.

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
   ...

> 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 <akim.demaille@epita.fr> 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" <s0009525@chelt.ac.uk>
Subject: Token Alias Bug
To: "'bug-bison@gnu.org'" <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" <s0009525@chelt.ac.uk>
Subject: Token Alias Bug
To: "'bug-bison@gnu.org'" <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
$<type_name>$ = $<type_name>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 <florian@edamail.fishkill.ibm.com>
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.