If we modify the default instance coordinates, then the inferred deltas that
are left in gvar are no longer valid, so we need to calculate them using the
original default coordinates.
They are then re-optimized using the modified default coordinates.
Also, the default deltas returned from instantiateTupleVariationStore are now
already rounded to integer.
The instantiateTupleVariationStore function now groups TupleVariation
tables that have the same axes 'tents', then merges them into a single
TupleVariation by summing their deltas. The rounding to integer happens
after summing the scaled deltas as floats, to reduce off-by-one errors.
To be able to sum gvar TupleVariation, it needs to calculate the inferred
deltas so it now takes two optional lists (origCoords and endPts) that
are passed on to iup_delta function. These only make sense for gvar
type of TupleVariation, of course, and are unused for cvar tuples.
It also run iup_delta_optimize on the gvar deltas that are left after
partial instancing and whose inferred deltas had to be interpolated.
This can be disabled with --no-optimize CLI option.
Also added calcInferredDeltas and optimize methods to TupleVariation
class, which use functions from varLib.iup module, plus tests
that exercise them.
Somehow we forgot to replace round -> otRound in arrayTools.calcIntBounds.
This function is used by glyf table to compute the glyphs' bounding boxes.
We already use otRound (aka 'int(math.floor(v + .5))') to round glyph
coordinates upon compiling glyf table. So the use of python3's round
in calcIntBounds was producing inconsistent roundings between the glyph
coordinates and the glyph bbox (sometimes, i.e. only when the glyf table
contains float coordinates, e.g. after instantiating with varLib.mutator).
update VarData.VarRegionCount
also set StartCoord and EndCoord to 0 (same end result as only setting PeakCoord
to 0, but this produces less noise when inspeciting the generated XML dump)
I added an MVAR table to the PartialInstancer-VF.ttx test font with made-up deltas for
OS/2.yStrikeoutSize, post.underlinePosition and post.underlineThickness.
I defined 3 regions, one with only wght, one with only wdth, and one with both wdth and wght axes.
Unlike simple glyphs, which always have 'program' attribute (even when empty, with no instructions)
in composite glyphs, it's the presence of 'program' attribute that determines whether
the optional WE_HAVE_INSTRUCTIONS component flag is set or not.
Thus, the trim method (with remove_hinting=True) should delete the attribute for
composite glyphs.
Fixes https://github.com/fonttools/fonttools/issues/1550
