PATCH RFA: Build system: Use AC_SYS_LARGEFILE

On 11/09/2010 05:46 PM, Ian Lance Taylor wrote:
> David Edelsohn<dje.gcc@gmail.com>  writes:
>
>> With Paolo's patch, my bootstrap has progressed into stage 2.
>> Hopefully the Flex-generated file is the only case where header files
>> are included in the wrong order.  Paolo's patch definitely is a step
>> in the right direction.
>
> Paolo's patch makes me a little bit uncomfortable because it means that
> gengtype-lex.l is compiled with a different ABI than the rest of the
> compiler.  Of course it's a very simple file and it doesn't use any
> aspect of the ABI.  But it does call fopen and fclose.  If a call to
> fseeko ever sneaks in there somehow, it will break on AIX in a rather
> mysterious manner.

I agree.  I had considered this too but then went for the inferior patch 
for a reason:

  # Remake internationalization support.

The advantage of this, if anything, is that something like that might be 
wrapped in automake's ylwrap script.

However, Ian---depending on the order in which David tests the various 
approaches, your patch is okay if it works.

Paolo

Paolo's earlier patch #undef'ing _LARGE_FILES works and allows me to
bootstrap; still running testsuite.

I agree about the potential problems, although gengtype is fairly independent.

I am worried about defining _LARGE_FILES in CFLAGS causing other
problems because that will cause many other files that do not use
auto-hosts.h to be compiled with _LARGE_FILES support with unknown
affects.

Inserting bconfig.h first seems the most localized approach.

Thanks, David

On Tue, Nov 9, 2010 at 11:55 AM, Paolo Bonzini <bonzini@gnu.org> wrote:
> On 11/09/2010 05:46 PM, Ian Lance Taylor wrote:
>>
>> David Edelsohn<dje.gcc@gmail.com>  writes:
>>
>>> With Paolo's patch, my bootstrap has progressed into stage 2.
>>> Hopefully the Flex-generated file is the only case where header files
>>> are included in the wrong order.  Paolo's patch definitely is a step
>>> in the right direction.
>>
>> Paolo's patch makes me a little bit uncomfortable because it means that
>> gengtype-lex.l is compiled with a different ABI than the rest of the
>> compiler.  Of course it's a very simple file and it doesn't use any
>> aspect of the ABI.  But it does call fopen and fclose.  If a call to
>> fseeko ever sneaks in there somehow, it will break on AIX in a rather
>> mysterious manner.
>
> I agree.  I had considered this too but then went for the inferior patch for
> a reason:
>
> Index: Makefile.in
> ===================================================================
> --- Makefile.in (revision 166028)
> +++ Makefile.in (working copy)
> @@ -3949,7 +3949,11 @@ $(genprog:%=build/gen%$(build_exeext)):
>
>  # Generated source files for gengtype.
>  gengtype-lex.c : gengtype-lex.l
> -       -$(FLEX) $(FLEXFLAGS) -o$@ $<
> +       -$(FLEX) $(FLEXFLAGS) -o$@ $< && { \
> +         echo '#include "bconfig.h"' > $@.tmp; \
> +         cat $@ >> $@.tmp; \
> +         mv $@.tmp $@; \
> +       }
>
>  #
>
>  # Remake internationalization support.
>
> The advantage of this, if anything, is that something like that might be
> wrapped in automake's ylwrap script.
>
> However, Ian---depending on the order in which David tests the various
> approaches, your patch is okay if it works.
>
> Paolo
>

David Edelsohn <dje.gcc@gmail.com> writes:

> Inserting bconfig.h first seems the most localized approach.

I agree and I'd like to go with that approach if it works.

Thanks.

Ian

Paolo,

Your Makefile patch seems to work as well, I am past building
gengtype-lex step in bootstrap.

Thanks, David

On Tue, Nov 9, 2010 at 11:55 AM, Paolo Bonzini <bonzini@gnu.org> wrote:
> On 11/09/2010 05:46 PM, Ian Lance Taylor wrote:
>>
>> David Edelsohn<dje.gcc@gmail.com>  writes:
>>
>>> With Paolo's patch, my bootstrap has progressed into stage 2.
>>> Hopefully the Flex-generated file is the only case where header files
>>> are included in the wrong order.  Paolo's patch definitely is a step
>>> in the right direction.
>>
>> Paolo's patch makes me a little bit uncomfortable because it means that
>> gengtype-lex.l is compiled with a different ABI than the rest of the
>> compiler.  Of course it's a very simple file and it doesn't use any
>> aspect of the ABI.  But it does call fopen and fclose.  If a call to
>> fseeko ever sneaks in there somehow, it will break on AIX in a rather
>> mysterious manner.
>
> I agree.  I had considered this too but then went for the inferior patch for
> a reason:
>
> Index: Makefile.in
> ===================================================================
> --- Makefile.in (revision 166028)
> +++ Makefile.in (working copy)
> @@ -3949,7 +3949,11 @@ $(genprog:%=build/gen%$(build_exeext)):
>
>  # Generated source files for gengtype.
>  gengtype-lex.c : gengtype-lex.l
> -       -$(FLEX) $(FLEXFLAGS) -o$@ $<
> +       -$(FLEX) $(FLEXFLAGS) -o$@ $< && { \
> +         echo '#include "bconfig.h"' > $@.tmp; \
> +         cat $@ >> $@.tmp; \
> +         mv $@.tmp $@; \
> +       }
>
>  #
>
>  # Remake internationalization support.
>
> The advantage of this, if anything, is that something like that might be
> wrapped in automake's ylwrap script.
>
> However, Ian---depending on the order in which David tests the various
> approaches, your patch is okay if it works.
>
> Paolo
>

On 11/11/2010 05:05 PM, David Edelsohn wrote:
> Paolo,
>
> Your Makefile patch seems to work as well, I am past building
> gengtype-lex step in bootstrap.

Can you commit it if it passes?

Paolo