[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Confused about mips integer types



Richard B. Kreckel wrote:
> Hi!
>
> Thiemo Seufer wrote:
>> FWIW, I did a test build on mips, which worked fine. I ran the build
>> under "linux32", which changes the kernel personality (an thus uname
>> from mips64 to mips).
>
> Thank you. So your suggestion is to help config.guess.
>
>> Additionally you may need to pass dpkg-architecture's DEB_HOST_GNU_TYPE
>> as target to configure.
>
>
> Aha. Your suggestion is to help config.guess.
>
> Julien Blanche wrote that configure was wrong setting __mips64__ because 
> the userland is 32 bit. Well, configure just I can help config.guess in any 
> of the two ways you suggest, and I'll do so if upgrading doesn't help, but 
> it stands in contradiction to what Julien said.

Upgrading config.guess to something recent is good advice in any case.

> Out of curiosity, I looked a little bit into config.guess and, indeed, it 
> doesn't appear to care about userland, only about what uname says. Did I 
> miss anything? Returning anything else but userland strikes me as wrong and 
> useless for a tool that is used as host detection device for configuring 
> programs that care only about, well, userland.

Unfortunately "uname" is the most widespread available indicator for
the userland type, so it is what the autotools check for.

The alternatives would be checking a random binary with "file" and hope
it has the same ABI as the user would like to see for the compilation,
or to check the default ABI of the compiler and hope that's the
intended one.

None of those techniques is particularily compelling for a multi-ABI
system. In Debian, however, we can rely on dpkg-architecture to provide
a stable mapping between Debian architectures and GNU triplets.

Passing the triplet explcitly has also the nice side-effect to make
your package more cross-compiler friendly.


Thiemo



Reply to: