Re: [patch] sim/erc32/ max simulation time extended by using 64bit ints

[Doug Evans <dje@google.com>]

On Sun, May 16, 2010 at 6:57 PM, Joel Brobecker <brobecker@adacore.com> wrote:
> Hey guys,
>
>> Index: sim/erc32/sis.h
>> ===================================================================
>> RCS file: /cvs/src/src/sim/erc32/sis.h,v
>> retrieving revision 1.2
>> diff -u -r1.2 sis.h
>> --- sim/erc32/sis.h   9 Jun 2002 15:45:46 -0000       1.2
>> +++ sim/erc32/sis.h   4 May 2010 21:14:55 -0000
>> @@ -23,6 +23,7 @@
>>  #include "ansidecl.h"
>>  #include "gdb/callback.h"
>>  #include "gdb/remote-sim.h"
>> +#include <stdint.h>
>
> Unfortunately, this change breaks the build when stdint.h is not
> available (Eg. sparc-solaris).

How old is this solaris?

OTOH, gdb/defs.h unconditionally includes stdint.h (I think I checked
it at the time, so I approved the patch) so I'm confused.

> Apparently, this header was included for 2 things:
>  - define two 64bit types int64 and uint64;
>  - have UINT64_MAX.
>
> I assume that the requirement was for 64bit minimum, as opposed to
> exactly 64bit? Making that assumption, we can remove the need for
> including stdint.h by using long long instead of int64_t (same thing
> for the unsigned counterpart).  Similarly, UINT64_MAX has a well defined
> value reguardless of the platform, so it can easily be defined as well.
> Looking at the rest of the type definitions above, it's actually in line
> with what's been done so far.
>
> Joel: Would that work for you as well?
>
> Doug: Would that be OK to commit? I think that the cleanest thing to do
>      here would be to have some configury that would provide our own
>      stdint.h when missing.  We already do that for GDB by using gnulib
>      so perhaps one way to do so would be to share the gnulib between
>      GDB and the sim (probably meaning moving it to the root directory).

For reference sake, there's bfd_stdint.h generated by
config/stdint.m4.  I don't know all that it handles, but again gdb
itself just includes stdint.h unconditionally.

> PS: We have the same problem with rx sim, I believe. I haven't tried
>    building it, though.