[ufoLib] userNameToFileName very slow for large existing lists

[madig]

Profiling https://github.com/madig/noto-amalgamated/blob/main/amalgamate-noto.py shows that ~2 mins of ~10 mins are spent in userNameToFileName, specifically

fonttools/Lib/fontTools/ufoLib/filenames.py

Line 101 in df2916a

if fullName.lower() in existing:

since existing is derived from self._existingFileNames.values() in

fonttools/Lib/fontTools/ufoLib/glifLib.py

Line 461 in df2916a

fileName = self.glyphNameToFileName(glyphName, self._existingFileNames.values())

.

writeGlyph is called per-glyph, I don't see an easy way to fix this. Changing the call to set(self._existingFileNames.values()) only makes it slower because now you make a set every time. And I'm not sure the list is guaranteed to be ordered?

[justvanrossum]

Good catch.

I think GlyphSet._existingFileNames needs to become a set. The only time its keys are used is in GlyphSet.deleteGlyph, and unless I'm missing something, I think we can write self._existingFileNames.remove(self.contents[glyphName]) there, instead of the current del self._existingFileNames[fileName].

(In other words, I think it's userNameToFileName()'s caller's responsibility to pass a set-like thing for the existing argument.)