Improve horizontal baseline detection.

src/core/SkScalerContext.cpp

 
 /*
  * Copyright 2006 The Android Open Source Project
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 
 #include "SkScalerContext.h"
@@ skipping 809 matching lines @@
     }
 
     // The 'remainingWithoutRotation' matrix GsA is the non-rotational part of A without the scale.
     if (GsA) {
         *GsA = GA;
          // G is rotational so reorders with the scale.
         GsA->preScale(SkScalarInvert(s->fX), SkScalarInvert(s->fY));
     }
 }
 
-SkAxisAlignment SkComputeAxisAlignmentForHText(const SkMatrix& matrix) {
-    SkASSERT(!matrix.hasPerspective());
-
-    if (0 == matrix[SkMatrix::kMSkewY]) {
-        return kX_SkAxisAlignment;
-    }
-    if (0 == matrix[SkMatrix::kMSkewX]) {

[bungeman-skia, 2016/02/29 18:00:27]

This was incorrect, or at least doesn't do what the documentation says it does.
If checking for when the x axis is mapped to the y axis, the correct check would
have been '0 == matrix[SkMatrix::kMScaleX]'. This is actually checking if the y
axis stays the y axis (which would seem to be a job for a hypothetical
SkComputeAxisAlignmentForVText).

-        return kY_SkAxisAlignment;
+SkAxisAlignment SkScalerContext::computeAxisAlignmentForHText() {
+    if (fRec.getHinting() != SkPaint::kNo_Hinting) {

[bungeman-skia, 2016/02/29 18:00:27]

This is causing a lot of churn in gold because most of the gms end up using the
portable font manager, which always sets kNo_Hinting in onFilterRec. This may
argue that this test should be looking at the hinting request (from the paint)
as opposed to the hinting resolution (from the scaler context). It's hard to say
how to handle this at all when the primary font does not hint, but a fallback
font will, making things not necessarily line up. There will need to be some
more thinking on this, but for now let's fix the existing bugs.

+        if (0 == fRec.fPost2x2[1][0]) {

[bungeman-skia, 2016/02/29 18:00:27]

To answer the question, does this need to use getSingleMatrix? That multiplies
in fTextSize (a scale, which should have no effect on either test), fPreScaleX
(a scale in x which would have no effect), and fPreSkewX (which would also have
no effect). In other words, making the text bigger, stretching it along the
horizontal axis, or fake italicizing it does not move the baseline. This will be
made into a comment.

The one difference that does exist here (with or with out getSingleMatrix) is
that this is testing against fPost2x2 which is the sk_relax version of the
matrix. On the other hand, this is somewhat 'more correcter-er' for stating that
if hinting is happening then the baselines should be snapped.

+            return kX_SkAxisAlignment;
+        }
+        if (0 == fRec.fPost2x2[0][0]) {
+            return kY_SkAxisAlignment;
+        }
     }
     return kNone_SkAxisAlignment;
 }
 
 ///////////////////////////////////////////////////////////////////////////////
 
 class SkScalerContext_Empty : public SkScalerContext {
 public:
     SkScalerContext_Empty(SkTypeface* face, const SkDescriptor* desc)
         : SkScalerContext(face, desc) {}
@@ skipping 24 matching lines @@
 
 SkScalerContext* SkTypeface::createScalerContext(const SkDescriptor* desc,
                                                  bool allowFailure) const {
     SkScalerContext* c = this->onCreateScalerContext(desc);
 
     if (!c && !allowFailure) {
         c = new SkScalerContext_Empty(const_cast<SkTypeface*>(this), desc);
     }
     return c;
 }