Skip all code blocks in the ensure_attribute_between_backticks_in_content rule - #2390

Not sure it's to error prone. Did you run it against all supported Symfony docs branches?

Good call: about 17,000 fake attributes later, you were right to be suspicious 😄

With each branch's .doctor-rst.yaml, the output on 6.4, 7.4, 8.0, 8.1 and 8.2 is identical to main, except for the messenger.rst false positive going away on 8.2.

Since the docs are clean, that alone proves little, so I also injected a bare #[ProbeN] line right after every code block (about 3,500 per branch). That found a real blind spot in DirectiveTrait::in(): it walks up to the first less indented directive, but doesn't stop at a less indented list item or definition term. So prose following a code block inside a #. item is taken as part of an earlier code block. main already has it for diff blocks only, and this PR would extend it to all of them (5 to 8 probes lost per branch).

The fix is a 3 line change: lower the reference indentation when a non-directive line with less indentation is met. With it, every probe is reported, the symfony-docs output doesn't change on any branch (8 rules use in()), and 3 new tests fail without it.

Since it touches 8 rules, I'll open an issue and a dedicated PR for in(); this one can wait for it.

So let's close here?

Thanks for merging #2392!

That one fixed in(), which attached prose to an earlier code block. This PR is a different defect: the rule only skips php and diff blocks, so an attribute inside any other block is still reported. On current main, with #2392 in, symfony-docs 8.2 still fails:

::error file=messenger.rst,line=4594::Please ensure to use backticks "          App\Message\DummyQuery (from #[AsMessage])"

With this PR on top, that line is gone and nothing else in the output moves: the three errors left on 8.2 are unrelated and were already there. So closing here would leave DOCtor-RST red on every symfony-docs 8.2 pull request, since the debug:messenger output in messenger.rst is a plain terminal block.

I just pushed it rebased on main, so the CI here runs against the merged in() fix, with the 2100 tests green. Happy to close instead if you would rather exempt code blocks another way, just tell me which way you prefer.