ada vs identical flex

ada vs identical flex

[Andrew Cagney]

I'm comparing the generated ada-lex.c output from FLEX 2.5.2.  Yes, 
that's comparing the output from the same "identical" FLEX - one is 
found on NetBSD/PPC 1.6.2, and the other on FC3.

For yyrealloc, I'm finding that some cases are wrapped in #ifndef 
YY_USES_REJECT, while others are not vis:

                 NBSD     FC3
declaration     yes      no
definition      yes      no
reference       yes      yes
ada_flex_use    no       no

On NetBSD, since ada_flex_use tries to refer to a function wrapped in 
#ifndef YY_USE_REJECT (i.e., #ifdef'ed out) the build barfs.

On the other hand on FC3, the declaration / definition are always 
visible and the build succeeds.

Ideas?
Andrew

Re: ada vs identical flex

[Paul Hilfinger]

> I'm comparing the generated ada-lex.c output from FLEX 2.5.2.  Yes, 
> that's comparing the output from the same "identical" FLEX - one is 
> found on NetBSD/PPC 1.6.2, and the other on FC3.

We'll look at it, but that's a REALLY old flex (on GDB, I never built
with anything <2.5.4, and that was from ~1997; latest rev is more like
2.5.31).

Paul

Re: ada vs identical flex

[Andrew Cagney]

> I think I can get rid of the yyrealloc stuff altogether.  It seems that
> one of the rules in ada-lex is not really used, but causes the non-use
> of yyrealloc as a side-effect.  If you could, would you try the following
> patch with the problematic platform/flex version?  Thanks.

Built for me!  Thanks.

Andrew

> Paul Hilfinger
> 
> 
> Index: merge.224/gdb/ada-lex.l
> --- merge.224/gdb/ada-lex.l Wed, 06 Oct 2004 02:59:24 -0700 hilfingr
> +++ merge.224(w)/gdb/ada-lex.l Fri, 08 Oct 2004 03:18:25 -0700 hilfingr
> @@ -154,8 +154,6 @@ static int find_dot_all (const char *);
>  		   return CHARLIT;
>  		}
>  
> -\"{OPER}\"/{WHITE}*"(" { return processId (yytext, yyleng); }
> -
>  <INITIAL>\"	{
>  		   tempbuf_len = 0;
>  		   BEGIN IN_STRING;
> @@ -926,5 +924,5 @@ yywrap(void)
>  typedef void (*dummy_function) ();
>  dummy_function ada_flex_use[] = 
>  { 
> -  (dummy_function) yyrealloc, (dummy_function) yyunput
> +  (dummy_function) yyunput
>  };
> 

Re: ada vs identical flex

[Paul Hilfinger]

> For yyrealloc, I'm finding that some cases are wrapped in #ifndef 
> YY_USES_REJECT, while others are not vis:

>                  NBSD     FC3
> declaration     yes      no
> definition      yes      no
> reference       yes      yes
> ada_flex_use    no       no

> On NetBSD, since ada_flex_use tries to refer to a function wrapped in 
> #ifndef YY_USE_REJECT (i.e., #ifdef'ed out) the build barfs.

Andrew,

I think I can get rid of the yyrealloc stuff altogether.  It seems that
one of the rules in ada-lex is not really used, but causes the non-use
of yyrealloc as a side-effect.  If you could, would you try the following
patch with the problematic platform/flex version?  Thanks.

Paul Hilfinger


Index: merge.224/gdb/ada-lex.l
--- merge.224/gdb/ada-lex.l Wed, 06 Oct 2004 02:59:24 -0700 hilfingr
+++ merge.224(w)/gdb/ada-lex.l Fri, 08 Oct 2004 03:18:25 -0700 hilfingr
@@ -154,8 +154,6 @@ static int find_dot_all (const char *);
 		   return CHARLIT;
 		}
 
-\"{OPER}\"/{WHITE}*"(" { return processId (yytext, yyleng); }
-
 <INITIAL>\"	{
 		   tempbuf_len = 0;
 		   BEGIN IN_STRING;
@@ -926,5 +924,5 @@ yywrap(void)
 typedef void (*dummy_function) ();
 dummy_function ada_flex_use[] = 
 { 
-  (dummy_function) yyrealloc, (dummy_function) yyunput
+  (dummy_function) yyunput
 };