Detect local label assignments as directives on MIPS (#6392)

[jf2048]

For obscure reasons (see comment here) GCC MIPS inserts DWARF debug labels with the format $LXX = . instead of $LXX:. So this updates the regexes in that scenario using the existing MIPS detection, to explicitly consider that specific pattern as a directive instead of an assignment. That way they are filtered correctly in the assembly output with the "Directives" option like on all other architectures.

This shouldn't need any additional tests, just updating the existing ones to say those labels should now be filtered. Currently it causes test failures though, I couldn't find any testing docs or if there is an automatic way to regenerate the test files? If there is a recommended way, I can update them.

[partouf]

I would like to some caution in terms of the tests failing:

• I don't think https://github.com/compiler-explorer/compiler-explorer/actions/runs/9041257536/job/24878097174?pr=6467#step:5:121 it would be appropriate to hide the label here (all filters disabled), the labels are used for address and size calculations stored in data
• same here https://github.com/compiler-explorer/compiler-explorer/actions/runs/9041257536/job/24878097174?pr=6467#step:5:204

The rest of the failing tests it seems Ok to hide the labels.

We've had a bit of shuffling around as the tests are concerned, but you can run npx vitest test/filter-tests.ts --update (according to #6366)

[jf2048]

Good point 😃 I managed to fix those tests and find a much simpler solution. Instead of messing about with directives I made it so symbol = . is now detected as a label in all cases. Note that will also cause those lines to be filtered if someone inserts that as inline assembly, but that should be consistent with how unused symbol: lines are handled too if those are placed in inline assembly.

Edit: I guess if someone really wants to put these in manually and always show them, things like symbol = . + 0 could still be used as a workaround.