[gdal-dev] autotest/ogr_mitab.py in trunk_mingw_w64

[Even Rouault]

Selon Kurt Schwehr <schwehr at gmail.com>:

> Even (or anyone else),
>
> Any idea what is up with the trunk_mingw_w64 failures with ogr_mitab?  I'm
> worried that I somehow caused this, but can't come up with how.
>
> https://trac.osgeo.org/gdal/ticket/6076
>

Kurt,

My first thinking would have been that it is due to
https://trac.osgeo.org/gdal/changeset/29535#file15
since the random failures started to appear at that point

But
1) I don't see anything wrong in the changes in ogr_mitab.py
2) when running isolated under Valgrind, I get nothing interesting
3) the error appears randomly

All that combined makes me think that it might be due to the tests running in
different order from one run to another one due to files being listed in
different order from one git clone to another one ( the tests are run in
os.listdir() order )

So I think the issue is in another test that doesn't properly close a dataset
/ cleanup after itself and let a ghost file in autotest/ogr/tmp which causes
the mitab driver to fail to open it afterwards, due to the presence of a file
that might be detected by another driver. The fact that it is only seen on
mingw targets is likely due to an attempt of deleting a file which isn't closed
before.

The error is perhaps in r29535 elsewhere...
If you don't find it by review, perhaps you could a debug trace to list the
files in tmp/ at the beginning ogr_mitab_3 ? That should give a good hint of
the test that cause that issue.

Even

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com