L4 fixes

Lib/fontTools/varLib/instancer/__init__.py

@@ -204,8 +204,8 @@
         default = None
         if n == 2:
             minimum, maximum = v
-        elif n == 3:
-            minimum, default, maximum = v
+        elif n >= 3:
+            return cls(*v)
         else:
             raise ValueError(f"expected sequence of 2 or 3; got {n}: {v!r}")
         return cls(minimum, default, maximum)
@@ -234,12 +234,14 @@
 
 
 @dataclasses.dataclass(frozen=True, order=True, repr=False)
-class NormalizedAxisTriple(AxisTriple):
+class NormalizedAxisTripleAndDistances(AxisTriple):
     """A triple of (min, default, max) normalized axis values."""
 
     minimum: float
     default: float
     maximum: float
+    distanceNegative: Optional[float] = 1
+    distancePositive: Optional[float] = 1
 
     def __post_init__(self):
         if self.default is None:
@@ -250,6 +252,50 @@
                 f"minimum={self.minimum:g}, default={self.default:g}, maximum={self.maximum:g})"
             )
 
+    def reverse_negate(self):
+        v = self
+        return self.__class__(-v[2], -v[1], -v[0], v[4], v[3])
+
+    def normalizeValue(self, v, extrapolate=True):
+        lower, default, upper, distanceNegative, distancePositive = self
+        assert lower <= default <= upper
+
+        if not extrapolate:
+            v = max(lower, min(upper, v))
+
+        if v == default:
+            return 0
+
+        if default < 0:
+            return -self.reverse_negate().normalizeValue(-v, extrapolate=extrapolate)
+
+        # default >= 0 and v != default
+
+        if v > default:
+            return (v - default) / (upper - default)
+
+        # v < default
+
+        if lower >= 0:
+            return (v - default) / (default - lower)
+
+        # lower < 0 and v < default
+
+        totalDistance = distanceNegative * -lower + distancePositive * default
+
+        if v >= 0:
+            vDistance = (default - v) * distancePositive
+        else:
+            vDistance = -v * distanceNegative + distancePositive * default
+
+        if totalDistance == 0:
+            # This happens
+            if default == 0:
+                return -v / lower
+            return 0  # Shouldn't happen
+
+        return -vDistance / totalDistance
+
 
 class _BaseAxisLimits(Mapping[str, AxisTriple]):
     def __getitem__(self, key: str) -> AxisTriple:
@@ -334,8 +380,13 @@
         normalizedLimits = {}
 
         for axis_tag, triple in axes.items():
+            distanceNegative = triple[1] - triple[0]
+            distancePositive = triple[2] - triple[1]
+
             if self[axis_tag] is None:
-                normalizedLimits[axis_tag] = NormalizedAxisTriple(0, 0, 0)
+                normalizedLimits[axis_tag] = NormalizedAxisTripleAndDistances(
+                    0, 0, 0, distanceNegative, distancePositive
+                )
                 continue
 
             minV, defaultV, maxV = self[axis_tag]
@@ -344,8 +395,10 @@
                 defaultV = triple[1]
 
             avarMapping = avarSegments.get(axis_tag, None)
-            normalizedLimits[axis_tag] = NormalizedAxisTriple(
-                *(normalize(v, triple, avarMapping) for v in (minV, defaultV, maxV))
+            normalizedLimits[axis_tag] = NormalizedAxisTripleAndDistances(
+                *(normalize(v, triple, avarMapping) for v in (minV, defaultV, maxV)),
+                distanceNegative,
+                distancePositive,
             )
 
         return NormalizedAxisLimits(normalizedLimits)
@@ -358,7 +411,7 @@
         self._data = data = {}
         for k, v in dict(*args, **kwargs).items():
             try:
-                triple = NormalizedAxisTriple.expand(v)
+                triple = NormalizedAxisTripleAndDistances.expand(v)
             except ValueError as e:
                 raise ValueError(f"Invalid axis limits for {k!r}: {v!r}") from e
             data[k] = triple
@@ -442,7 +495,7 @@
 
 
 def changeTupleVariationAxisLimit(var, axisTag, axisLimit):
-    assert isinstance(axisLimit, NormalizedAxisTriple)
+    assert isinstance(axisLimit, NormalizedAxisTripleAndDistances)
 
     # Skip when current axis is missing (i.e. doesn't participate),
     lower, peak, upper = var.axes.get(axisTag, (-1, 0, 1))
@@ -505,7 +558,7 @@
                         "Instancing accross VarComposite axes with variation is not supported."
                     )
                 limits = axisLimits[tag]
-                loc = normalizeValue(loc, limits)
+                loc = limits.normalizeValue(loc, extrapolate=False)
                 newLocation[tag] = loc
             component.location = newLocation
 
@@ -925,15 +978,21 @@
             mappedMax = floatToFixedToFloat(
                 piecewiseLinearMap(axisRange.maximum, mapping), 14
             )
+            mappedAxisLimit = NormalizedAxisTripleAndDistances(
+                mappedMin,
+                mappedDef,
+                mappedMax,
+                axisRange.distanceNegative,
+                axisRange.distancePositive,
+            )
             newMapping = {}
             for fromCoord, toCoord in mapping.items():
-
                 if fromCoord < axisRange.minimum or fromCoord > axisRange.maximum:
                     continue
-                fromCoord = normalizeValue(fromCoord, axisRange)
+                fromCoord = axisRange.normalizeValue(fromCoord)
 
                 assert mappedMin <= toCoord <= mappedMax
-                toCoord = normalizeValue(toCoord, (mappedMin, mappedDef, mappedMax))
+                toCoord = mappedAxisLimit.normalizeValue(toCoord)
 
                 fromCoord = floatToFixedToFloat(fromCoord, 14)
                 toCoord = floatToFixedToFloat(toCoord, 14)

Lib/fontTools/varLib/instancer/featureVars.py

@@ -1,5 +1,4 @@
 from fontTools.ttLib.tables import otTables as ot
-from fontTools.varLib.models import normalizeValue
 from copy import deepcopy
 import logging
 
@@ -41,7 +40,9 @@ def _limitFeatureVariationConditionRange(condition, axisLimit):
         # condition invalid or out of range
         return
 
-    return tuple(normalizeValue(v, axisLimit) for v in (minValue, maxValue))
+    return tuple(
+        axisLimit.normalizeValue(v, extrapolate=False) for v in (minValue, maxValue)
+    )
 
 
 def _instantiateFeatureVariationRecord(
@@ -50,9 +51,9 @@ def _instantiateFeatureVariationRecord(
     applies = True
     shouldKeep = False
     newConditions = []
-    from fontTools.varLib.instancer import NormalizedAxisTriple
+    from fontTools.varLib.instancer import NormalizedAxisTripleAndDistances
 
-    default_triple = NormalizedAxisTriple(-1, 0, +1)
+    default_triple = NormalizedAxisTripleAndDistances(-1, 0, +1, 0, 0)
     for i, condition in enumerate(record.ConditionSet.ConditionTable):
         if condition.Format == 1:
             axisIdx = condition.AxisIndex

Lib/fontTools/varLib/instancer/solver.py

@@ -1,4 +1,4 @@
-from fontTools.varLib.models import supportScalar, normalizeValue
+from fontTools.varLib.models import supportScalar
 from fontTools.misc.fixedTools import MAX_F2DOT14
 from functools import lru_cache
 
@@ -12,15 +12,17 @@ def _reverse_negate(v):
 
 
 def _solve(tent, axisLimit, negative=False):
-    axisMin, axisDef, axisMax = axisLimit
+    axisMin, axisDef, axisMax, distanceNegative, distancePositive = axisLimit
     lower, peak, upper = tent
 
     # Mirror the problem such that axisDef <= peak
     if axisDef > peak:
         return [
             (scalar, _reverse_negate(t) if t is not None else None)
             for scalar, t in _solve(
-                _reverse_negate(tent), _reverse_negate(axisLimit), not negative
+                _reverse_negate(tent),
+                axisLimit.reverse_negate(),
+                not negative,
             )
         ]
     # axisDef <= peak
@@ -98,9 +100,8 @@ def _solve(tent, axisLimit, negative=False):
     #                           |
     #                      crossing
     if gain > outGain:
-
         # Crossing point on the axis.
-        crossing = peak + ((1 - gain) * (upper - peak) / (1 - outGain))
+        crossing = peak + (1 - gain) * (upper - peak)
 
         loc = (axisDef, peak, crossing)
         scalar = 1
@@ -116,7 +117,7 @@ def _solve(tent, axisLimit, negative=False):
         # the drawing above.
         if upper >= axisMax:
             loc = (crossing, axisMax, axisMax)
-            scalar = supportScalar({"tag": axisMax}, {"tag": tent})
+            scalar = outGain
 
             out.append((scalar - gain, loc))
 
@@ -147,84 +148,73 @@ def _solve(tent, axisLimit, negative=False):
 
             # Eternity justify.
             loc2 = (upper, axisMax, axisMax)
-            scalar2 = supportScalar({"tag": axisMax}, {"tag": tent})
+            scalar2 = 0
 
             out.append((scalar1 - gain, loc1))
             out.append((scalar2 - gain, loc2))
 
-    # Case 3: Outermost limit still fits within F2Dot14 bounds;
-    # we keep deltas as is and only scale the axes bounds. Deltas beyond -1.0
-    # or +1.0 will never be applied as implementations must clamp to that range.
-    #
-    # A second tent is needed for cases when gain is positive, though we add it
-    # unconditionally and it will be dropped because scalar ends up 0.
-    #
-    # TODO: See if we can just move upper closer to adjust the slope, instead of
-    # second tent.
-    #
-    #            |           peak |
-    #  1.........|............o...|..................
-    #            |           /x\  |
-    #            |          /xxx\ |
-    #            |         /xxxxx\|
-    #            |        /xxxxxxx+
-    #            |       /xxxxxxxx|\
-    #  0---|-----|------oxxxxxxxxx|xo---------------1
-    #    axisMin |  lower         | upper
-    #            |                |
-    #          axisDef          axisMax
-    #
-    elif axisDef + (axisMax - axisDef) * 2 >= upper:
-
-        if not negative and axisDef + (axisMax - axisDef) * MAX_F2DOT14 < upper:
-            # we clamp +2.0 to the max F2Dot14 (~1.99994) for convenience
-            upper = axisDef + (axisMax - axisDef) * MAX_F2DOT14
-            assert peak < upper
-
+    else:
         # Special-case if peak is at axisMax.
         if axisMax == peak:
             upper = peak
 
-        loc1 = (max(axisDef, lower), peak, upper)
-        scalar1 = 1
-
-        loc2 = (peak, upper, upper)
-        scalar2 = 0
-
-        # Don't add a dirac delta!
-        if axisDef < upper:
-            out.append((scalar1 - gain, loc1))
-        if peak < upper:
-            out.append((scalar2 - gain, loc2))
+        # Case 3:
+        # We keep delta as is and only scale the axis upper to achieve
+        # the desired new tent if feasible.
+        #
+        #                        peak
+        #  1.....................o....................
+        #                       / \_|
+        #  ..................../....+_.........outGain
+        #                     /     | \
+        #  gain..............+......|..+_.............
+        #                   /|      |  | \
+        #  0---|-----------o |      |  |  o----------1
+        #    axisMin    lower|      |  |   upper
+        #                    |      |  newUpper
+        #              axisDef      axisMax
+        #
+        newUpper = peak + (1 - gain) * (upper - peak)
+        assert axisMax <= newUpper  # Because outGain >= gain
+        if newUpper <= axisDef + (axisMax - axisDef) * 2:
+            upper = newUpper
+            if not negative and axisDef + (axisMax - axisDef) * MAX_F2DOT14 < upper:
+                # we clamp +2.0 to the max F2Dot14 (~1.99994) for convenience
+                upper = axisDef + (axisMax - axisDef) * MAX_F2DOT14
+                assert peak < upper
+
+            loc = (max(axisDef, lower), peak, upper)
+            scalar = 1
 
-    # Case 4: New limit doesn't fit; we need to chop into two tents,
-    # because the shape of a triangle with part of one side cut off
-    # cannot be represented as a triangle itself.
-    #
-    #            |   peak |
-    #  1.........|......o.|...................
-    #            |     /x\|
-    #            |    |xxy|\_
-    #            |   /xxxy|  \_
-    #            |  |xxxxy|    \_
-    #            |  /xxxxy|      \_
-    #  0---|-----|-oxxxxxx|        o----------1
-    #    axisMin | lower  |        upper
-    #            |        |
-    #          axisDef  axisMax
-    #
-    else:
+            out.append((scalar - gain, loc))
 
-        loc1 = (max(axisDef, lower), peak, axisMax)
-        scalar1 = 1
+        # Case 4: New limit doesn't fit; we need to chop into two tents,
+        # because the shape of a triangle with part of one side cut off
+        # cannot be represented as a triangle itself.
+        #
+        #            |   peak |
+        #  1.........|......o.|....................
+        #  ..........|...../x\|.............outGain
+        #            |    |xxy|\_
+        #            |   /xxxy|  \_
+        #            |  |xxxxy|    \_
+        #            |  /xxxxy|      \_
+        #  0---|-----|-oxxxxxx|        o----------1
+        #    axisMin | lower  |        upper
+        #            |        |
+        #          axisDef  axisMax
+        #
+        else:
+            loc1 = (max(axisDef, lower), peak, axisMax)
+            scalar1 = 1
 
-        loc2 = (peak, axisMax, axisMax)
-        scalar2 = supportScalar({"tag": axisMax}, {"tag": tent})
+            loc2 = (peak, axisMax, axisMax)
+            scalar2 = outGain
 
-        out.append((scalar1 - gain, loc1))
-        # Don't add a dirac delta!
-        if peak < axisMax:
-            out.append((scalar2 - gain, loc2))
+            out.append((scalar1 - gain, loc1))
+            # Don't add a dirac delta!
+            if peak < axisMax:
+                out.append((scalar2 - gain, loc2))
 
     # Now, the negative side
 
@@ -295,7 +285,7 @@ def rebaseTent(tent, axisLimit):
     If tent value is None, that is a special deltaset that should
     be always-enabled (called "gain")."""
 
-    axisMin, axisDef, axisMax = axisLimit
+    axisMin, axisDef, axisMax, distanceNegative, distancePositive = axisLimit
     assert -1 <= axisMin <= axisDef <= axisMax <= +1
 
     lower, peak, upper = tent
@@ -305,7 +295,7 @@ def rebaseTent(tent, axisLimit):
 
     sols = _solve(tent, axisLimit)
 
-    n = lambda v: normalizeValue(v, axisLimit, extrapolate=True)
+    n = lambda v: axisLimit.normalizeValue(v)
     sols = [
         (scalar, (n(v[0]), n(v[1]), n(v[2])) if v is not None else None)
         for scalar, v in sols