Two small changes that significantly speed up subsetting of large fonts
such as Noto Sans CJK:
1. When emptying a charstring, simply empty its program rather than
attempting to decompile it first. (Only relevant when retaining GIDs.)
2. When reindexing charstrings, swap an accidentally-quadratic
implementation for one that is linear in the number of retained glyphs.
* [subset CFF] Fix bug in de-subroutinizing when subroutines contain hints, issue 1493
The code was skipping executing a subroutine if it had already been desubroutinized. The initial set of vstemhm and hstemhm operators and values may be in a subroutine. If a charstring is being executed which calls such subroutines, they still need to be executed in order to count the number of hint values seen, so that the byte length of the hintmask can be calculated.
I fixed this bug by executing subroutines even if they have already been desubroutinized, as long as (we don't know yet if we are doing hintmasks) or ( we do need a hintmask, but have not yet seen it).
Clean up code per Cosimo's suggestions:
In arg list for stop_hint_count(), use *args to accept unused argument, rather than a dummy positional argument.
Change stop_hintcount_ops to a from a global variable to a class variable in _DesubroutinizingT2Decompiler.
Remove un-needed 'return' at line 387
Remove duplicate assignment of cs at line 437
Add patch for the bug where AttributeError is encountered when remove_hints is run after desubroutinize: remove lines deleting the GlobalSubrs for each FontDict. This always needed to be done only once, and is now in any case done in cff.GlobalSubrs.clear(), at the end of the desubroutinize() function.
Changed test case subset_test.py::'test_no_hinting_desubroutinize_CFF' to reference a font with a non-empty GlobalSubr, in order to trigger AttributeError traceback.
Fixes#1483
'remove_unused_subroutines' method expects a 'GlobalSubrs' attribute but the 'desubroutinize'
method deletes it. We don't have to call 'remove_unused_subroutines' at the end of 'desubroutinize'
method, we can just delete all the local subroutines, and clear the GlobalSubrsIndex.
Only call 'remove_unused_subroutines' method once, when we are not desubroutinizing.
@bedhad
Address issues raised in #1403
I do think setting the dummy CFF2 PrivateDict nominalWidthX and defaultWidthX to None, which leads to the charstring.width also being None, is a good idea. I originally set them to 0, which produces a charstring width of 0, in order to avoid problems with logic that assumes that the field is good for math. However, I now think that it is better to find errors around charstring type assumptions earlier than later.
"drop_hints()" is actually not wrong - I did look at this when making the changes. For CFF2 charstrings, self.width is always equal to self.private.defaultWidthX, so the width is never inserted. This is because in psCharstrings.py::T2WidthExtractor.popallWidth(), the test "evenOdd ^ (len(args) % 2)" is alway False. Left to myself, I would not change this code. If the CFF2 charstring is correct, there is not a problem. if the CFF2 charstring is not correct, then both in drop_hints() and in T2WidthExtractor.popallWidth(), the logic will stack dump. I did add asserts, but am not totally sure it is worth the extra calls.
Fixed psCharstrings so that calc_bounds will run. I would guess no-one has tried to use a BoundsPen on a CFF2 VF before - thanks to Chris Chapman. It now returns a result only for the default instance.
Fixed bug in subsetting: removed assert that a Subr is not empty after subsetting or de-hinting. CFF2 Charstrings do not have terminal "return' op.
