Re: [Gcl-devel] Logs of failed build (ANSI) on WinXP -- configure process

[Gabriel Dos Reis]

Camm Maguire <address@hidden> writes:

| Gabriel Dos Reis <address@hidden> writes:
| 
| > Camm Maguire <address@hidden> writes:
| >
| > | Greetings!
| > | 
| > | Gabriel Dos Reis <address@hidden> writes:
| > | 
| > | > Donald Winiecki <address@hidden> writes:
| > | >
| > | > [...]
| > | >
| > | > | checking build system type... i686-pc-mingw32
| > | > | checking host system type... i686-pc-mingw32
| > | > | host=i686-pc-mingw32
| > | > | The host is canonicalised to i386-pc-mingw32
| > | >
| > | > Hi Camm,
| > | >
| > | > I have been wondering why GCL deliberately recategorizes a
| > | > i686-pc-mingw32 host as a i386-pc-mingw32 host?  Later on, it uses that
| > | > to configure GMP, which might lead to suboptimal code selection on a
| > | > host with greater capabilities..
| > | >
| > | 
| > | This was put in by Mike Thomas a long time ago, and arguably should be
| > | removed.  He ran into a situation where he tried to move the binary
| > | from one machine to another and got an Illegal instruction fault.  gmp
| > | is built in (i.e. not a shared library) on these systems.  He put in a
| > | configure switch --enable/--disable-common-binary governing this.
| >
| > Indeed, I think we can/should remove that now.
| 
| So in other words, no one moves windows binaries across machines
| anymore?

I think people are still moving binaries around.  However, I don't think
i368 is the common denominators of modern machines running Windows.

The `system requirements' of OS supported by MS are such that, if your
machine can only run i386 coes, then you are already out of luck.  
I would be surprised if that is the case of the vast majority of GCL users.

| > | Even when this is disabled, host=mingw32 defaults to i386.  One can do
| > | --host=i686-pc-mingw32 to get a higher capability build if desired.
| > | This is of course cross compiling so no cpu detection is pertinent.
| > | On a separate front, am a little puzzled why --build fails in place of
| > | --host. 
| >
| > That is puzzling, because `--host' is supposed to default to `--build'.
| > I think it would be welcome to remove the `recanonalization' to i386.
| >
| 
| OK, I'll look into this.

With the cross toolchain I set up on my x86_64 running openSUSE-11.2, I
am able to configure and build, install and use vanilla GMP-4.3.x. 


-- Gaby