[Doxygen-users] Doxygen testing (was: hide undocumented enums)

[Dimitri v. H. <di...@st...>]

On Mon, Jun 11, 2001 at 04:49:18PM +0200, Volker Boerchers wrote:
> Hi all,
> 
> we see statements like this
> 
> [Peter Steiner]
> > I get undocumented #defines in the documentation (introduced somewhere
> > since 1.2.4 [I skipped 1.2.5 and 1.2.6, but 1.2.7, 1.2.8 and 1.2.8.1 all
> > have this bug]).
> 
> quite often in the last months in this list and i've the impression
> that they are still getting more. It seems as if most reported bugs
> are features that were added in version XXX but disappeared again in
> version XYZ.

Most of these happen because 
1) I'm restructuring things and
2) Things are a quite fragile (this is many because of the loosly parsing
   using flex, which is both a good and a bad thing.)
3) I (can) only do limited testing.

> This sounds as if there should be a regression test suite to protect
> the great features that doxygen provides from disappearing
> while adding further features. (As far as i can see no tests are part
> of the doxygen sources.)

That would certainly be nice.

> I have some experience with the great CppUnit test framework (see
> http://sourceforge.net/projects/cppunit, a port of JUnit) but i think
> it would be too much work to instrument doxygen with more or less
> complete unit tests.
> 
> Another, simpler possibility would be a collection of input files and
> approved output files to compare to with 'diff' each time a code
> change is made. To automate these tests shell scripts or dedicated
> tools like dejagnu could be used.

I think this would make the tests rather fragile. If I tweak something
on the output (for instance use a table instead of a list somewhere) 
that would invalidate all tests. Another problem is how to verify that the
generated diagrams contain the correct information.

Maybe verification on the XML output level would be more robost, but
with the current XML output a lot of information would be missed.

So, any ideas are welcome!

Regards,
  Dimitri