SmartEiffel / gcc 4 bug

3 messages Options
Embed this post
Permalink
dubois.georges

SmartEiffel / gcc 4 bug

Reply Threaded More More options
Print post
Permalink
Hello,

I found a minimal modification to compile SmartEiffel with gcc 4  -O2 was to
ensure good string hash_code (hash_code >= 0) after optimization (with the
addition of a line like: Result := Result.hash_code at the end of
STRING.hash_code).

Best regards,
Georges Dubois
Guillaume Lemaître-2

Re: SmartEiffel / gcc 4 bug

Reply Threaded More More options
Print post
Permalink
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Georges Dubois a écrit :
> Hello,
>
> I found a minimal modification to compile SmartEiffel with gcc 4  -O2 was to
> ensure good string hash_code (hash_code >= 0) after optimization (with the
> addition of a line like: Result := Result.hash_code at the end of
> STRING.hash_code).
>
> Best regards,
> Georges Dubois

  Hello Georges,

Unfortunately, I added the patch you proposed in STRING.e, and it still
doesn't work on Fedora10/x86_64, that embeds a gcc 4.3.2. I also tried
to change make_release.sh to have boost mode associated with -O2 instead
of -Os, with no more success.

Even though I agree with you, there seems to have something with hash
codes (I have to dig into that with gdb)

        Guillaume
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)

iEYEARECAAYFAkl8mhMACgkQGl8D2UjaREBWxACg0japVpoyPfCAYxW6QL+/5uAj
tOYAoMS9rXdopyCou/8uhlM6CQr080lv
=8qDe
-----END PGP SIGNATURE-----
dubois.georges

Re: SmartEiffel / gcc 4 bug

Reply Threaded More More options
Print post
Permalink
Guillaume Lemaître a écrit :

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Georges Dubois a écrit :
>  
>> Hello,
>>
>> I found a minimal modification to compile SmartEiffel with gcc 4  -O2 was to
>> ensure good string hash_code (hash_code >= 0) after optimization (with the
>> addition of a line like: Result := Result.hash_code at the end of
>> STRING.hash_code).
>>
>> Best regards,
>> Georges Dubois
>>    
>
>   Hello Georges,
>
> Unfortunately, I added the patch you proposed in STRING.e, and it still
> doesn't work on Fedora10/x86_64, that embeds a gcc 4.3.2. I also tried
> to change make_release.sh to have boost mode associated with -O2 instead
> of -Os, with no more success.
>
> Even though I agree with you, there seems to have something with hash
> codes (I have to dig into that with gdb)
>
> Guillaume
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
>
> iEYEARECAAYFAkl8mhMACgkQGl8D2UjaREBWxACg0japVpoyPfCAYxW6QL+/5uAj
> tOYAoMS9rXdopyCou/8uhlM6CQr080lv
> =8qDe
> -----END PGP SIGNATURE-----
>  
Hello Guillaume,

The patch works on Windows XP with gcc 4.3.2-tdm-2 mingw32 (64-bits
systems appear to cause other problems...).

Best regards,
Georges Dubois