This also fixes unreported issues unrelated to the 7.4.0 release:
- (now old style "lightbox") admonitions render badly in tabulary,
- "heavybox" admonitions (they are all now but warning et al. were already) cause PDF crash if in tabulary and rendered badly if in tabular or longtable.
- figure in a table cell is not properly separated from immediately preceding text.
Co-authored-by: Adam Turner <9087854+aa-turner@users.noreply.github.com>
This commit adds a `level` option to the `rubric` directive, which propagates a `level` attribute to the `rubric` node,
and allows renderers to select a specific heading level.
Logic for this attribute has been added to the HTML5 and LaTeX builder.
Consider the following piece of reST::
.. automodule:: sphinx.ext.autosummary
:members:
.. autosummary::
sphinx.ext.autosummary.Autosummary
This inserts an autosummary after the module docstring, but before
the members of the module. Without the change in this commit, this
would fail because `import_by_name` would attempt to import::
sphinx.ext.autosummary.sphinx.ext.autosummary.Autosumary
because the prefix (from the parent) is `sphinx.ext.autosummary`,
and the name is `sphinx.ext.autosummary.Autosummary`, which is able
to be imported from `sphinx.ext.autosummary`, but is not the way
that anyone would want to refer to it.
Co-authored-by: Adam Turner <9087854+AA-Turner@users.noreply.github.com>
Currently there is no mechanism to identify totally undocumented modules
in the coverage builder, unlike with partially documented modules.
Resolve this by introducing a new ``coverage_modules`` config option.
This is a list of modules that should be documented somewhere
within the documentation tree.
Any modules that are specified in the configuration value but
are not documented anywhere will result in a warning.
Likewise, any modules that are not in the config option but
are documented somewhere will result in a warning.
Signed-off-by: Stephen Finucane <stephen@that.guru>
- Remove now-removed configuration values
- Use obvious 'unset' defaults for ``project`` and ``author``
- Prefer 'e.g.' to 'ex.'
- Allow non-list sequence types in various configuration values
- Add types and defaults to every confval directive
- Add test for basic literal block & literalinclude translation
- Add gettext builder test for litteral blocks additional targets
- Add i18n test for xref roles in titles
- Add i18n tests for strange markup
The key issue this commit seeks to address, is that existing tools / documentation often lead users to mistakenly use object types and not role names, a classic example being `function` not `func`
Previously, the warning message for using e.g. `` external:function`target` `` (with `py` as the default domain), would be:
```
WARNING: role for external cross-reference not found: function
```
This gives no information to the user on how to fix the issue, even though there is actually quite an easy fix
This commit adds logic to create more specific / helpful warning messages, e.g.
```
WARNING: role for external cross-reference not found in domains 'py', 'std': 'function' (perhaps you meant one of: 'py:func', 'py:obj')
```
This goes through the same original logic but, if the role is not found, it will look if the role name is actually an available object type on the domain(s), and then suggest its related roles.
This commit also reverts #12133, which provided a (silent) fallback to auto convert an object type to a role name.
This commit fixes the issue of `objects.inv` denoting object names, whilst the `external` role only allows for role names.
As an example, take the `objects.inv` for the sphinx documentation, which contains:
```
py:function
compile : usage/domains/python.html#compile
```
A user might understandably expect that they could reference this using `` :external:py:function:`compile` ``, but actually this would previously error with:
```
WARNING: role for external cross-reference not found: py:function
```
this is because, `function` is the object type, yet `external` expects the related role name `func`.
It should not be necessary for the user to know about this distinction,
so in this commit, we add logic, to first look if the name relates to a role name (as previous, to not be back-breaking) but, if not, then also look if the name relates to an object that has a known role and, if so, use that.
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
If importing with ``TYPE_CHECKING is True`` fails, reset the state of ``sys.modules``
so that the attempt with ``TYPE_CHECKING is False`` may succeed.
Co-authored-by: Adam Turner <9087854+aa-turner@users.noreply.github.com>
