Re: Git v2.13.1 SHA1 very broken

[Ramsay Jones <ramsay@ramsayjones.plus.com>]

On 05/06/17 22:05, Ævar Arnfjörð Bjarmason wrote:
> On Mon, Jun 5, 2017 at 10:34 PM, Adam Dinwoodie <adam@dinwoodie.org> wrote:
>> I'm trying to compile Git v2.13.1 to release for Cygwin, but it appears
>> a010391 ("sha1dc: update from upstream", 2017-05-20) is breaking a very
>> significant number of test cases in both 32-bit and 64-bit Cygwin
>> builds.
>>
>> The first failure is t0000.46 "validate object ID of a known tree"; output with
>> -x and -v is below, although it's not very interesting:
>>
>>     expecting success:
>>             test "$tree" = 7bb943559a305bdd6bdee2cef6e5df2413c3d30a
>>
>>     ++ test ceb282701536fe61bea01075664405caa7d6343f = 7bb943559a305bdd6bdee2cef6e5df2413c3d30a
>>     + test_eval_ret_=1
>>     + want_trace
>>     + test t = t
>>     + test t = t
>>     + set +x
>>     error: last command exited with $?=1
>>     not ok 46 - validate object ID of a known tree
>>     #
>>     #               test "$tree" = 7bb943559a305bdd6bdee2cef6e5df2413c3d30a
>>     #
>>
>> I have no idea where to even begin debugging this, but I'm happy to take
>> pointers / try things out on my box.
> 
> That looks scary, can you please comment out this:
> 
>     #define SHA1DC_ALLOW_UNALIGNED_ACCESS

No, that doesn't fix it.

> 
> In sha1dc/sha1.c and see if that helps, alternatively comment out the
> ifdefs guarded by "#ifdef _MSC_VER" calls in sha1dc/sha1.c

This can't possibly make a difference! ;-)

However, rebuilding with:

    $ make OPENSSL_SHA1=YesPlease >out2 2>&1

... make t0000-basic.sh pass just fine, so ...

ATB,
Ramsay Jones