Re: [PATCH] Fix gdbserver build with x86_64-w64-mingw32 -m32

[Zach Welch <zwelch@codesourcery.com>]

On 01/13/2015 01:17 AM, Joel Brobecker wrote:
>>> Intuitively, I would say that the proper way to achive a 32bit
>>> gdbserver is by configuring it using a 32bit triplet, no?
>>> What happens if you do:
>>>
>>>     ./configure --build=i686-pc-mingw32 CFLAGS='-m32'
>>
>> Actually, I am using i686-pc-linux-gnu as $build.  I think you meant to
>> ask for --host.  I am using --host=x86_64-w64-mingw32 CFLAGS=-m32, which
>> is distinctly different than i686-pc-mingw32.  For example, they use
>> completely different runtime libraries.  I think it would be an error to
>> conflate the two toolchains.  Thus, I stand by my patch as the correct
>> solution for this issue.
> 
> Actually, I did mean --build=. This option is meant to override
> what the config.guess script would normally guess as being your
> operating system. On Windows, at least in my case, config.guess
> guesses i686-pc-mingw32, which is not what you want, because it's
> cygwin. That's why you need --build=i686-pc-mingw32.
> 
> Unless I misunderstood you, I suspect you're gettting confused by
> the meaning of the --host, --build and --target.

I believe that you have misunderstood my scenario.

The build server's host OS is x86_64-linux-gnu; however, I am using a
i686-pc-linux-gnu toolchain (thus, --build=i686-pc-linux-gnu) to produce
32-bit binaries (for maximum portability).  I am using that toolchain to
build a complete cross toolchain for x86_64-w64-mingw32 (--target).

Thus, the primary host for which I am building gdbserver is
x86_64-w64-mingw32 (--host=x86_64-w64-mingw32), but I am also using the
x86_64-w64-mingw32 toolchain to build a multilib for 32-bit windows
(--host=x86_64-w64-mingw32 CFLAGS="-m32") as a secondary host. gdbserver
runs on the target (and is not itself a cross compiler), so --host ==
--target in both cases.

The primary host gdbserver build (--host=x86_64-w64-mingw32) works, but
the secondary host build (--host=x86_64-w64-mingw32 CFLAGS="-m32") fails
without my provided patch.

In conclusion, I feel fairly confident that I am using the correct
configure options.  The i686-pc-mingw32 triplet should not enter into
the equation at all in my situation.

-- 
Zach Welch
Mentor Graphics Corporation
zwelch@codesourcery.com
(503) 685-5434