Add reporting of build platform

[landstander668]

@dscho This is my attempt to add build platform information to the output of git version --build-options for Issue #846. For now it only understands x86 and x86_64, but I plan to extend the process to include other platforms as well (so that this hopefully won't end up being a Git for Windows specific item).

At present all of the 32-bit variants are collapsed into x86, which made sense to me at the time but may or may not be the most desirable approach. For example:

$ uname -m
i686

$ git version --build-options
git version 2.9.3.windows.2.826.gd486be9
sizeof-long: 4
machine: x86

[dscho]

Nice! Could you Sign-off on the patch, please?

[landstander668]

@dscho I thought about that just after submitting the PR, but wasn't too concerned since this will almost certainly need to be rebased and force-pushed at least once before it's ready for formal inclusion. :) I'll re-push with signoff later today, and try to not to skip that step even for the interim patches going forward.

[dscho]

Well, I'd be comfortable merging it. Upstream Git is due this Friday, so... ;-)

[landstander668]

@dscho In that case, you're quite welcome to it. :) I've just re-pushed the branch with sign-off included (no code changes were made).

I'll followup with support for other platforms either via another pull request, or directly with upstream Git if that's more appropriate. Mainly I just need to find a list of the various GCC platform-specific defines, so I can determine how to identify the beasties.

[dscho]

Thanks!

As to GCC defines, I guess it would be safer to add $(uname_M) to the CFLAGS in config.mak.uname...

[landstander668]

@dscho Wouldn't that give inaccurate results in the case of cross-compiling to a different target platform? That's the main reason I was using the built-in platform defines, although I certainly could have been missing something.

[landstander668]

On the plus side, I think I've found a pretty comprehensive list of the GCC platform defines here. Should be able to run with that, unless updating CFLAGS (per your previous reply) is deemed to be clearly superior.

[dscho]

Wouldn't that give inaccurate results in the case of cross-compiling to a different target platform?

I should have made myself clearer, sorry. I was wondering whether we could not introduce a CFLAG, say, -DBUILD_ARCHITECTURE="$(BUILD_ARCHITECTURE)" that would be passed from the Makefile. If not set explicitly, e.g. via config.mak or via the command-line, it would be set to $(uname_M), falling back to (unknown) via ifeq (,$(uname_M)).

[landstander668]

@dscho Ah, now I'm with you. I'll go ahead and start looking into that... we can always revert the existing implementation if it works out better.