varLib merger dies on CFF->CFF2 conversion

[simoncozens]

I'm trying to merge some OTFs to VF because reasons. The first thing that varLib.merge does is convert the default master CFF table to a CFF2. With this font:

from fontTools.ttLib import TTFont
from fontTools.varLib.cff import convertCFFtoCFF2

f = TTFont("NotoSansCJKjp-Light.otf")
f.ensureDecompiled()
convertCFFtoCFF2(f)

dies with

  File "/Users/simon/others-repos/fonttools/Lib/fontTools/cffLib/__init__.py", line 1549, in _read
    return SubrsIndex(file, isCFF2=isCFF2)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/simon/others-repos/fonttools/Lib/fontTools/cffLib/__init__.py", line 878, in __init__
    super(GlobalSubrsIndex, self).__init__(file, isCFF2=isCFF2)
  File "/Users/simon/others-repos/fonttools/Lib/fontTools/cffLib/__init__.py", line 784, in __init__
    assert offSize <= 4, "offSize too large: %s" % offSize
           ^^^^^^^^^^^^
AssertionError: offSize too large: 7

[simoncozens]

The problem seems to be coming when calling .Subrs on a PrivateDict object as part of this little dance:

fonttools/Lib/fontTools/varLib/cff.py

Lines 122 to 128 in e724acd

	# At this point, the Subrs and Charstrings are all still T2Charstring class
	# easiest to fix this by compiling, then decompiling again
	cff.major = 2
	file = BytesIO()
	cff.compile(file, otFont, isCFF2=True)
	file.seek(0)
	cff.decompile(file, otFont, isCFF2=True)

[simoncozens]

The problem seems related to the presence of multiple FontDicts in an FDArray. I wonder if we're not descending into the FDArray properly when doing some conversion somewhere. (This is a guess, I know very little about CFF fonts.)

[simoncozens]

Here's an interesting thing. If I change the code in cff.py (line 95) to:

        for fontDict in fdArray:
            fontDict.setCFF2(True)
            print(fontDict.Private, fontDict.Private.rawDict.keys())

it dies. But if I say:

        for fontDict in fdArray:
            print(fontDict.Private, fontDict.Private.rawDict.keys())
            fontDict.setCFF2(True)

it succeeds.

[behdad]

That suggest some lazy-loading is not happening?