GitLab

* commit '1dccc9f3':
  Fix access to freed memory

* commit '372648a7':
  Fix access to freed memory

Source should be freed after all jpeg operations completed
since it is used by these operations.

Change-Id: Ieb0741e1d219dc0513133a94d47df86b5f95cf29

It looks like Id8c91ae0be6cad8a7ef77a0cd5803676290986c1 upgraded this
message from an actual error notification into a full-fledged adb log
spammer. Oops! This CL moves it back to the location it should be at,
and requires SK_DEBUG to see it even there.

Bug: 7038312
Change-Id: Iefc0b001a9874e0f177507d43231e6fdb4a1d92a

Android/Skia supports pre-multiplied Alpha. WebP image format supports
Alpha via color modes MODE_RGBA & MODE_RGBA_4444 (corresponding ones for
premultiplied alpha are MODE_rgbA & MODE_rgbA_4444).
The pre-requisite change for this one is the libwebp patch:
https://googleplex-android-review.googlesource.com/#/c/219333/

Change-Id: I30b0d8d4b0325eb3c925371ac8f3a9b7d4801639

Cherry-pick Id8c91ae0be6cad8a7ef77a0cd5803676290986c1 from master.

During font initialization, create a seperate fallback list for each
locale. At runtime, use the fallbacklist associated with the locale
set in the paint object. Fallback files are associated with locales in
fallback_fonts.xml. Multiple files can be associated with the same
locale, ordering within that langauge and apart from that language in
the fallback order is preserved.

This changelist also includes some refactoring, notably of the
functions that call getNextContext().

Change-Id: I121f0e491a522c4a8558a0066b2d8969fb8a3667

* commit '5f421caf':
  Forward-compatibility stubs

* commit 'c2ad1992':
  Forward-compatibility stubs

Change-Id: I9c5bf5ce38827ced91d8912a1fa49adbd14a46b8

bug: 6987261
Change-Id: I8cb5b6da2347a1fa88a13aa3548312a7773831e5

* commit 'dccf2260':

Change-Id: Ia62ca8b048a34639d11f7838b03c74420b3b3868

The underlying problem is that no SkScalerContext objects existed at the
time shapeFontRun is called immediately after a locale change from en to
ja (apparently the dumping of the cache caused all these to be
deallocated), so gFTLibrary was null (and the call tio ref_ft_face
assumes that it's initialized).

There's a pattern for calls which might not necessarily be called from a
scaler context (GetAdvancedTypefaceMetrics is one such), to explicitly
check for an uninitialized library, and create one for the length of the
call if so. This patch changes GetUnitsPerEm to follow this pattern.

Change-Id: I19a4b6fa49fad0aeacc04bf971101aacca6bc94f

* commit '5ff3803b':
  Reduce embolden effect. DO NOT MERGE

* commit '1d6fa961':
  Reduce embolden effect. DO NOT MERGE

Cherry-pick of Iec1c478171e33751be1cddbaae6a8ec30bb6cbab

Certain CJK glyphs become essentially unreadable when emboldened at
sizes used by common apps like News/Weather. Reducing embolden
strength slightly makes these glyphs much more readable without
sacrificing bold/normal weight differentiation.

Bug: 6712857
Change-Id: I1025f72b5ed07f0a65924ba58aa19511bc417eb1

Certain CJK glyphs become essentially unreadable when emboldened at
sizes used by common apps like News/Weather. Reducing embolden
strength slightly makes these glyphs much more readable without
sacrificing bold/normal weight differentiation.

Bug: 6712857
Change-Id: Iec1c478171e33751be1cddbaae6a8ec30bb6cbab

The advance values computed by generateAdvance() were inconsistent with
those computed by generateMetrics, because the fMatrix22 was being
applied in the latter but not former case. Since fMatrix22.xx is 1.0
almost all the time (the exception is only when skew and scale are both
applied), failures were rare.

This patch is an alternative to ag/#/c/207887, in that they both fix the
problem, but in different ways. That patch made fMatrix22.xx equal to
1.0 in almost all cases, while this one gives correct advance values
for arbitrary fMatrix22 values.

Change-Id: Iedfa36e884e3e3e2f078a5d4edfc82004a54e895

After this checkin, SkCreateTypefaceForScript refs each typeface
that it creates before it returns it.  Additionally, all supported harfbuzz
codepoints are added to the unicode mapping array.  This patch also fixes a test crash (see bug)

Bug:6756432

Change-Id: Ida3276b76e745b817f9e63be54945e411516dd03

This reverts commit 81957ccb

This reverts commit 421654d4

* commit '456fc154':
  Cherrypick change to skia

* commit '8c1c7cfa':
  Cherrypick change to skia

Change-Id: I7514b6931a11a0a79ec3abd4068f72998fcdf99d

    Fonts can be marked with elegant or compact in fallback_fonts.xml.
Webkit uses elegant fonts, Textview uses compact fonts (the default),
unmarked fonts are used by both.

    Bug: 6649136

Change-Id: I6ab6173c2efc50eba8fcc488c89dfdd083fb9c06

* commit 'c27ee87c':

Change-Id: I10b0459ad51c4a2d7d62ad73d2a8fd3e51828f9c

* commit 'e8fff9d9':
  DO NOT MERGE Skia solution to fake-bold fallback fonts.

* commit 'ec1ad08b':
  DO NOT MERGE Skia solution to fake-bold fallback fonts.

This version of the fake-bold-for-fallback-fonts fix addresses the
issue succinctly inside of Skia, with no need for Skia's clients to
be aware of the issue. Attempts to draw a bold parent font flag the
embolden flag in the associated SkScalerContext while leaving the
SkPaint's fake bold flag unmodified.

This depends upon change Icdd13f6b, which arranges for FreeType's
FontHost to ignore redundant embolden requests.

This is a cherry-pick of change I4132a232 from master.

Bug: 6629786
Change-Id: I389d6793e5472d664f6ac7b894927dbfd811be12