When there are multiple \throws entries in a function documentation, the second
and further exception names were omitted from output and a combined description
was associated with the first exception. Fix this by producing separate
paragraphs, one per exception.
Additionally, support free form exception description when the exception type
is specified as "~". In this case, the exception name is not used, and the
entry description is used to initialize the paragraph in the output.
Updated test to verify these changes.
This makes the entries after "See Also:" to be displayed inline rather
than starting from the next line. This is consistent with other kinds
of paragraphs generated from Doxygen output.
Produce id attributes for all entities generated by Doxygen. For refs,
generate link tags referencing those ids. Preface the ids with "doxygen."
prefix to avoid potential clashes with ids that may be present in other
places of the resulting DocBook document. Also add the document refid to
the prefix to avoid potential clashes between multiple Doxygen-generated
documents with SHORT_NAMES=YES.
In order to pass through the generated ids without modification (to keep
links working in the output), change generate.id template so that it
returns the value of the id attribute, if one is present, instead of
generating a new id.
The end result is that all references generated by Doxygen (to methods,
classes, variables and so on) propagate to the final HTML output instead
of being stripped during Doxygen to BoostBook translation.
Comparing formatted XML files makes the comparison less sensitive
to whitespace and newline differences between XML documents, and
also makes the subsequent diff more informative, if any differences
are found.
Whatever version of xsltproc I'm not using generates even longer id
numbers, which means that the base id needs to be further truncated to
make space for them.
It turns out the '\code' blocks don't have this problem, so an easier
fix would be just to use them. But now I've implemented the workaround,
might as well leave it in, in case anyone wants to use <pre> blocks in
the future.
This appears to have removed the computeroutput tags from function links in
the hash documentation. I'm not sure if the solution is to retain them in
annotation mode or to be smarted about syntax highlighting programlisting.
[SVN r86505]
Turns out Doxygen doesn't like BOOST_NOEXCEPT_IF. Text parsing in XSL is an
almighty pain, it might be better to just combine whatever follows the
arguments in argsstring with the values from the attributes.
[SVN r86285]
This may seem like overkill, but it's going to be needed for handling
parameters to noexcept. Doesn't handle escapes because that my sanity is
already stretched far enough.
[SVN r86284]
Should really add proper support to boostbook for `=delete`, `=default` and
noexcept, but I'm not feeling that keen right now. I'm not sure I did the right
thing with `BOOST_CONSTEXPR_OR_CONST`.
[SVN r86277]
The ids are now a little more truncated than they used to be, this is partly
due to the fix in r77599, and partly due to the generate-id generating longer
ids in newer versions of xsltproc.
It might be better to improve the id normalisation so that it is more
consistent across versions.
[SVN r82891]
