Non-RST source text is parsed as RST after translation

[jpmckinney]

Describe the bug

a12399d introduced a regression: If the source text is in Markdown, it gets parsed as RST after translation.

In Sphinx 1.6.7 (and earlier), in publish_msgstr, reader.read() (i.e. SphinxBaseReader.read()) would select the parser based on the source_path suffix. After that commit, the parser is instead set by:

parser = app.registry.create_source_parser(app, '')

This means that it's always RST that parses the translated text. Obviously, if the translated text is in Markdown syntax, this will not result in a correct translation.

I'll submit a PR that determines the filetype based on the source_path. Ideally, this can be applied to 2.2.x.

To Reproduce

I have open PRs on CommonMark.py and recommonmark that also need to be merged to be able to translate Markdown in Sphinx, so it will be hard to create a 'small' example.

Expected behavior

I expect the translated text to be parsed as Markdown (not RST) if the source text is Markdown.

Your project

https://github.com/open-contracting-extensions/public-private-partnerships

Screenshots

N/A

Environment info

• OS: Mac
• Python version: 3.6.1
• Sphinx version: 2.2.1
• Sphinx extensions: recommonmark
• Extra tools: N/A

Additional context

This was originally fixed in aefab51 and then broken in a12399d

I have open PRs on CommonMark.py and recommonmark that also need to be merged to be able to translate Markdown in Sphinx:

• Don't garbage collect raw strings, which are needed for gettext translation by Sphinx readthedocs/commonmark.py#225
• Set rawsource value of each node to use as translation source for Sphinx i18n feature readthedocs/recommonmark#187

[tk0miya]

Fixed by #6927.
Thank you for reporting.