Re: [xml] libxslt test suite broken by recent libxml2 commit

[Nick Wellnhofer <wellnhofer aevum de>]

On Oct 4, 2014, at 08:36 , Daniel Veillard <veillard redhat com> wrote:

> > I successfully updated the libxslt test suite to the current docbook-xsl version 1.78.1 from 1.40 and this 
> > indeed fixes the problem. But the resulting commit is quite large:
> >
> > 551 files changed, 207340 insertions(+), 59510 deletions(-)
> >
> > The libxslt tarball would grow by about 1 MB. Do you think this is acceptable?
>
>  Hummm,  The goal is to do regression tests :-) maybe it's a bit too
>  much for just avoiding a warning n generated output,
>
>    so not completely convinced, any other benefit I missed ?

Sure, it’s possible that we don’t catch some regressions that only occur with the old docbook-xsl version but 
in the long run, I think the newer version would catch much more issues. Version 1.40 is really ancient (from 
2001 or even older) and the current version uses many more complex XSLT constructs. It’s probably the most 
sophisticated XSL library out there. The goal of the docbook tests is not only regression testing but also 
integration testing so a somewhat up-to-date version is essential.

There are also bug reports like this one which probably need a current version of docbook-xsl to reproduce 
and test:

    https://bugzilla.gnome.org/show_bug.cgi?id=736077

I'd also propose to run the full docbook test suite when running ‘make tests’. Currently, only a subset of 
the docbook tests are executed and the output of the tests that weren't run before hasn’t been adjusted for 
years making these tests pretty much useless (‘make full’ in tests/docbook). Updating these tests already 
helped me to catch a problem with one of my earlier commits.

Additionally, I’d like to change the way we handle the problem of non-determistic ids in the test output 
created with generate-id(). Currently, we simply ignore all lines that contain the string “id”. So we throw 
away large and essential parts of the test output. It would be better to run the output through sed to only 
remove the id attributes.

I wouldn’t want to make all these changes on top of the ancient docbook-xsl version.

Nick