It would fail if parent is a template class

[ryzzn]

m.css/doxygen/dox2html5.py

Line 1750 in eb2ee6f

assert compound.name.startswith(prefix)

Hi,
Prefix won't contain parent's template parameters list but compound.name does contain, so the assertion would fail.

[mosra]

Hi,

sorry, I have trouble reproducing this. Can you provide a minimal code snippet that triggers this assert? Thank you!

[ryzzn]

I'm sorry I haven't described the issue clearly.

Here's the code:

template <class T>
struct S
{
};

template <class T>
struct S<T*>
{
  struct SS {};
};

S<T*> is partial template specialization and failure message is:

 Traceback (most recent call last):
  File "./dox2html5.py", line 2897, in <module>
    run(doxyfile, os.path.abspath(args.templates), args.wildcard, args.index_pages, search_merge_subtrees=not args.search_no_subtree_merging, search_add_lookahead_barriers=not args.search_no_lookahead_barriers, search_merge_prefixes=not args.search_no_prefix_merging)
  File "./dox2html5.py", line 2796, in run
    postprocess_state(state)
  File "./dox2html5.py", line 1752, in postprocess_state
    assert compound.name.startswith(prefix)
AssertionError

Then I print the compound and prefix variables

namespace(brief='', children=[], has_details=<Element 'detaileddescription' at 0x7f0a9cf61048>, id='structS_3_01T_01_5_01_4_1_1SS', is_deprecated=False, kind='struct', name='S&lt;T*&gt;::SS', parent='structS', templates=None, url='structS_3_01T_01_5_01_4_1_1SS.html')

S::

[mosra]

Ahh, great, partial template specialization, that's what I missed. :)

As crazy as it might sound, it's not a problem in m.css but in fact related to a Doxygen bug I reported back in December: https://bugzilla.gnome.org/show_bug.cgi?id=791942 (a bug that appears to be completely unrelated). As noted in the bug description, I "worked around" that by reverting the particular offending commit in my Doxygen fork (see mosra/doxygen@5ac9b11), but that unfortunately brings back some other problems (but minor problems, compared to this).

With that commit reverted, the prefix is correctly S<T*>:: instead of just S:: and everything is great. If you can apply the revert on your side, awesome, if not, I have an (ugly) workaround for you:

The problem boils down to the * and & characters, among others, so if the specialization contains them, it will fail. If you write the following, for example, it will work:

template <class T>
struct S<typename std::add_pointer<T>::type>
{
  struct SS {};
};

Yes, I know, it's really ugly, but can do the job. Unfortunately this is not possible to fix in m.css, it's a problem in Doxygen itself and all I can do is just consume its output. But I'll push on that bug report, asking when it can be fixed :)

[mosra]

Added a comment on the Doxygen issue.

[ryzzn]

T_T, Thanks for reasoning this in detail.

Finally I'll choose revert that commit to make work, because my project isn't small enough to change all code relating to partial template specialization. But I haven't tried it. If there is any result, I'll reply to you. Thanks so much.

[mosra]

Let's keep this open until the Doxygen issue is fixed. It causes numerous problems to me as well. There was some activity on it yesterday, so maybe this bug gets some attention too.

I would attempt to fix that myself, but this particular code is totally impenetrable for me and I gave up on trying to understand what is happening in it. Also with the lack of tests there I fear I would break other things by accident :/

[mosra]

For reference, doxygen/doxygen#6299 is the related Doxygen issue on the GitHub bugtracker, still no progress there :/

[mosra]

Fix for this issue is in doxygen/doxygen@449a7e2, scheduled to be released as 1.8.15 most probably at the end of the year.