Pavel Sanda wrote: > Georg Baum wrote: >> However, since LyX depends heavily on external tools, failures like >> this do not indicate a problem with the test, but with the spreadsheet >> support on this particular distro, so actually the test did its job. > > This depends how much you want to adhere to GNU standards. If some test > is not sensible for a given system (i.e. the distribution doesn't support > some 3rd party tool), the test is supposed to be skipped, not to fail > with autotools.
Sure, but this is not the case here: The test which fails for Kornel does make sense on his OS, so it should not be skipped. >> I don't care if I need to call 'make check' or 'make testall' or >> whatever, but IMHO it is also a bit naive of distro packagers to assume >> that a 'make check' target executes tests which are suitable for their >> needs. > > My feeling is that make - check - install is kind of standard mantra for > building autotools based stuff but I have no reference to prove it. > Certainly I can see that it's unusual that tests are explicitly disabled > on source based distribution. If this is some standard then lets execute the test by a different target. I did not know about that. >> Both the layout and tex2lyx tests have already found real bugs. The >> latest failing tex2lyx was my bad, I forgot to run the tests before >> committing 89b2c54c00. I fixed that now. > > I don't attack these tests per se. The question is whether they are for > our own regression testing or for people trying to install/pack it on > their computer. Second case implies we have unresolved bugs. Mainly for us, but could be useful for packagers as well. Georg
