libmspub should have tests

[Michael Terry]

Neither upstream nor the packaging introduces tests for libmspub. We should really have them!

[Björn Michaelsen (bjoern-michaelsen)]

The Microsoft Publisher import in libmspub could profit of tests against regressions of the conversion by comparing known good files to the expected result. Caolán McNamaras convwatch:

 http://www.openoffice.org/qa/qadevOOo_doc/convwatch/convwatch.html

shows a possible starting point for the concept as applied to a different part of the product.

[Dtardon (dtardon)]

The latest rewrite of convwatch is http://cgit.freedesktop.org/libreoffice/core/tree/bin/convwatch.py

[Markus Mohrhard (moggi)]

Just a few comments:

convwatch looks like a bad idea and maybe a bit unreliable for libmspub, libvisio and libcdr

But since all three of them implement interfaces they could all easily use mocking similar to libcmis and orcus. The advantage with mocking compared to a visual comparison that I see is that as long as the libs are under heavy development you have a high rate of false positives while mocking allows you to concentrate on the features that you wish to test.

[Dtardon (dtardon)]

FWIW, we do have regression tests for most of the libraries. They use "raw" output format (which just prints the API calls and their arguments) and flat ODF for comparison. And we run them before every release. There is none for libmspub yet, because noone has invested the time to create a representative set of documents. (Well, we need a better set of documents for libcdr and libvisio too. With a clear license, so we can actually make them public.)