[FFmpeg-devel] Regression tests fail on MinGW

[Víctor Paesa]

Hi,

said Ramiro Polla
> Hello,
>
> Regression tests fail on MinGW since r11142. I've been procrastinating
> to find the cause for an incredible amount of time.
>
> The problem is that strtod() on Windows doesn't like hex strings. When
> av_get_string() was used instead of av_get_double(), it would pass hex
> strings to av_set_string() (which uses av_strtod(), which uses strtod())
>  for FF_OPT_TYPE_FLAGS. So -flags +bitexact wasn't doing anything, and
> hence the tests fail.
>
> This should be fixed on MinGW, and I sent a message about it on [0].
>
> I haven't checked, but I assume this doesn't affect Cygwin, because it
> reimplements almost everything to be compliant.

Right, Cygwin passes your strtod() check at [0].

On the other hand, Cygwin fails for wmav1/wmav2:
--- /home/Inma/src/ffmpeg-13030/tests/ffmpeg.regression.ref     2008-05-02
06:25:07.000000000 +0200
+++ tests/data/vsynth.regression        2008-05-02 09:39:25.656250000 +0200
@@ -197,11 +197,11 @@
 353368 ./tests/data/a-flac.flac
 c4228df189aad9567a037727d0e763e4 *./tests/data/flac.vsynth.out.wav
 stddev: 33.31 PSNR:65.87 bytes:1040384
-4435d87463cd6c5407bd88cca241ca56 *./tests/data/a-wmav1.asf
+a24bd1ac6f8046aad904f726f9ceccd6 *./tests/data/a-wmav1.asf
 106004 ./tests/data/a-wmav1.asf
-stddev:12251.50 PSNR:14.56 bytes:1056768
-stddev:2106.00 PSNR:29.85 bytes:1048576
-24ae726aef89e5d17e315c64182f4c61 *./tests/data/a-wmav2.asf
+stddev:12251.76 PSNR:14.56 bytes:1056768
+stddev:2105.99 PSNR:29.85 bytes:1048576
+b7a4528d36cb0eea601ad97b8295af09 *./tests/data/a-wmav2.asf
 106044 ./tests/data/a-wmav2.asf
-stddev:12255.92 PSNR:14.55 bytes:1056768
-stddev:2099.31 PSNR:29.88 bytes:1048576
+stddev:12255.91 PSNR:14.55 bytes:1056768
+stddev:2099.19 PSNR:29.88 bytes:1048576
make: *** [codectest] Error 1

I believe that is due to some rounding errors (PSNR are identical),
though I must confess I never tried to find the cause as I don't use
wmav1/wmav2.

> By the way, am I the only one running make test on MinGW and actually
> caring about the result? Doesn't anyone else that distributes builds do
> this?
>
> Ramiro Polla
> [0] http://article.gmane.org/gmane.comp.gnu.mingw.user/26356

Regards,
V?ctor