Re: PHP support

[Tim Landscheidt]

Akim Demaille <address@hidden> wrote:

>> I hope this list is the right place for this.

>>  In the past few weeks, I started working on "%language
>> PHP". You can browse the code at
>> <URI:https://github.com/scfc/bison-php> (bison's "master" is
>> "upstream").

> I am very curious about this.  Are you really wanting to _use_ a Bison parser 
> in PHP?  Or is this some kind of experimental toy project?  Unfortunately the 
> current maintainers don't spend a lot of time on Bison, and features that 
> might never be maintained will finish by hindering the development of the 
> whole project.

<excursus>

The starting point of this endeavour is
<URI:https://bugzilla.wikimedia.org/show_bug.cgi?id=17465>.
ATM, MediaWiki uses an external Ocaml program (generated by
ocamllex/ocamlyacc) to determine whether a "<math>" segment
contains only "safe" TeX by validating against a (subset of)
TeX grammar.

  So I looked for existing PHP scanner/parser generators and
found several, let's say, code dumps.  Few were working
code, actively maintained were none and grammars, if docu-
mented at all, looked rather funky and/or were lacking
flex's/Bison's functionalities.

  Rather than whipping one of them in usable shape, I adapt-
ed Bison's Java generator as grammar, concept & Co. are fa-
miliar to *everyone*, the infrastructure of testsuites,
mailing lists, bug trackers, etc. is already in place and it
would be much easier to port new features like GLR if the
need arises.

</excursus>

>>   But the second "is" should be a "$is". I tried some
>>   variants of "$$" and patsubst at different places, but
>>   unfortunately, m4's levels of quoting have always exceed-
>>   ed my imagination :-).

> Bison is not ready for this, not at all.  The easiest would be to 
> post-process the result as use some kind of new quadrigraph to denote $, say 
> @S|@ :)

> Autoconf uses this:

>       s/\@<:\@/[/g;
>       s/\@:>\@/]/g;
>       s/address@hidden:\@/(/g;
>       s/\@:\}\@/)/g;
>       s/address@hidden|\@/\$/g;
>       s/address@hidden:\@/#/g;
>       s/\@&t\@//g;

> A cleaner design requires more thinking.

Actually, the real problem lay elsewhere as Bison expected
"type identifier" as the argument to %lex-param and just si-
lently accepted, but "mistreated" "$identifier" :-).  What
do you think of the patch:

| diff --git a/data/php.m4 b/data/php.m4
| index 7095d8a..6ab2600 100644
| --- a/data/php.m4
| +++ b/data/php.m4
| @@ -262,7 +262,9 @@ m4_define([b4_lex_param_call],
|                      [$1])])
|  m4_define([b4_param_calls],
|            [m4_map([b4_param_call], address@hidden)])
| -m4_define([b4_param_call], [, $2])
| +# FIXME: This should probably better be dealt with in parse-gram.y's
| +#        add_param ().
| +m4_define([b4_param_call], [, m4_bpatsubst($1, [^.* ], [])])

that I posted in
<URI:news:address@hidden>?

  I would strongly disagree with your assumption that Bison
isn't ready for this, though.  The code at
<URI:https://github.com/scfc/bison-php> is already working
for simple examples.  The only real showstopper yet - on the
C side - is the use of "$variables" in actions for which I
have written a patch that I will post for discussion in the
next few days (which reminds that I still have to reply to
Bruno on bug-gnulib :-().

Tim