(Even) better merger errors

[simoncozens]

Fixes #2224 (as well as some other silent errors) and adds a test.

[simoncozens]

Thanks, good catches!

[madig]

I realize it's probably too late for a debate on principles, but is there a special reason why there's an enum and error classes instead of subclassing the base VarLibMergeError for each subcategory of VarLibMergeFailure? Would make instantiation and printing a bit more straightforward I suppose...

[simoncozens]

Maybe. I’ll take a look at doing it that way.

[simoncozens]

Done, but now the code is sufficiently different, you're going to have to review it again. :-)

[madig]

Hm, this scuppers CFF merge errors, they follow their own logic. Maybe the merge errors with a cause should be subclasses of a class VarLibDetailedMergeError(VarLibMergeError) for now. Or someone needs to adapt the CFF code... Also, could the __init__s use kwargs so one can do

UnsupportedFormat(self, subtable="pair positioning lookup")
# UnsupportedFormat(self, ({"subtable": "pair positioning lookup"},))

InconsistentGlyphOrder(self)
# InconsistentGlyphOrder(self, ({},))

[anthrotype]

@madig good catch about CFF-related VarLibMergeErrors (I'd expected the tests to fail?)

also 👍 on using **kwargs

[madig]

The tests don't fail, but the str makes little sense.

[simoncozens]

Yes, moving to kwargs was a good move; it allowed me to split up the reason from the traceback, which makes everything a lot tidier.

[madig]

Small nit: if your class has a docstring, you don't need a pass.