Before, we were loading a palette (again and again) even if the
same was requested. Even if the font only had one palette...
For a font like NotoColorEmoji that has over 5000 colors in its
palette, this was dominating the COLRv1 loading times for HarfBuzz
(and I believe all other clients) because they have to set the
palette to get access to the colors.
* src/base/ftcolor.c (FT_Palette_Select): Check the current palette.
* src/bdf/bdflib.c (bdf_parse_start_): Reject fonts with initial
COMMENTs.
(bdf_parse_properties_): Skip COMMENTs so that...
(bdf_add_property_): Do not make exception for COMMENT.
(bdf_parse_glyphs_, bdf_add_comments): Updated.
* src/bdf/bdflib.c (bdf_parse_start_, bdf_add_property_): Delay handling
of the default char until...
* src/bdf/bdfdrivr.c (BDF_Face_Init): ... handles all other properties.
* src/bdf/bdf.h (bdf_font_t): Remove `monowidth`.
* src/bdf/bdflib.c (bdf_parse_start_, bdf_add_property_): Delay handling
of the font spacing until...
* src/bdf/bdfdrivr.c (BDF_Face_Init): ... handles all other properties.
* src/bdf/bdflib.c (bdf_parse_start_, bdf_add_property_): Delay handling
of the font ascent and descent until...
* src/bdf/bdfdrivr.c (BDF_Face_Init): ... handles all other properties.
Instead of cumbersome field list mamangement, we will tokenize input
using custom `bdf_strtok_`.
* src/bdf/bdflib.c (bdf_list_t_, bdf_list_init_, bdf_list_ensure_,
bdf_list_shift_, bdf_list_join_, bdf_list_split_,
bdf_set_default_spacing_): Removed.
(bdf_strtok_): New function which NUL-terminates the first token at
the delimiter position and returns the next token that follows
consequtive delimiters.
(bdf_parse_*_, bdf_load_font): Updated.
* docs/CHANGES: Claim overall 75% performance improvement.
* src/bdf/bdf.h (bdf_options_t): Removed.
(BDF_KEEP_COMMENTS, BDF_KEEP_UNENCODED, BDF_CORRECT_METRICS): Updated
to avoid collisions with the other parsing flags.
* src/bdf/bdfdrivr.c (BDF_Face_Init): Use default flags explicitly.
* src/bdf/bdflib.c (bdf_load_font): Use flags as an argument to
initialize the parser.
(bdf_set_fedault_spacing_): Default to BDF_PROPORTIONAL explicitly.
(bdf_parse_*_): Updated.
A dedicated parser of bitmap data with minimal error checking and
no string comparisons helps to improve performance.
* src/bdf/bdflib.c (bdf_parse_bitmap_): New dedicated bitmap parser.
(bdf_parse_glyphs_): Pass to `bdf_parse_bitmap_` on BITMAP.
This resolves build problems with newer libPNG versions by switching
to using IMPORTED CMake targets for common libraries where provided.
To do this the required CMake version is raised to version 3.12.
Doing this seems justified as using IMPORTED targets is a cleaner
and more powerful solution that can leverage advanced and CMake-specific
features that may be utilized by the referred dependencies' CMake modules.
Resolves#1311.
After the previous commit, cff_glyph_load is no longer called
without first checking for valid size and glyph objects in
FT_Glyph_Load and these checks can be removed downstream.
* src/cff/cffdrivr.c (cff_glyph_load): Remove `size` and `glyph` checks.
* src/cff/cffgload.c (cff_slot_load): Ditto.
Otherwise, let TT_Get_Advances fall back on cff_load_glyph to do
slow advances. This avoids unchecked access to cff_load_glyph and
this is how tt_get_advances is implemented.
This assignmets used to be done prematurely before errors were checked
and outlines were actually loaded. Delaying the assignment provides
certain protection against careless usage of malformed input that
should now remain FT_GLYPH_FORMAT_NONE.
* src/cff/cffgload.c (cff_slot_load): Ditto.
* src/cid/cidgload.c (cid_slot_load_glyph): Ditto.
* src/pfr/pfrobjs.c (pfr_slot_load): Ditto.
* src/truetype/ttgload.c (TT_Load_Glyph): Ditto.
* src/type1/t1gload.c (T1_Load_Glyph): Ditto.
This is already done by `ft_glyphslot_clear`.
* src/cff/cffgload.c (cff_slot_load): Do not zero outline fields.
* src/cid/cidgload.c (cid_slot_load_glyph: Ditto.
* src/pfr/pfrobjs.c (pfr_slot_load): Ditto.
* src/truetype/ttgload.c (load_sbit_image): Ditto.
* src/type1/t1gload.c (T1_Load_Glyph): Ditto.
For consistency with other bitmap-only fonts, we should ignore this
flag of the font is not scalable.
* src/truetype/ttgload.c (TT_Load_Glyph): Check if face is scalable
when checking for FT_LOAD_NO_BITMAP.
* docs/CHANGES: Start new chapter wtith this change.
In `FT_Stroker_ConicTo` and `FT_Stroker_CubicTo` there is a `bez_stack`.
`arc` is initialized with `arc = bez_stack` and is never set to point
into any different object. The main loop looks like `while ( arc >=
bez_stack )` which is depending on a later `arc -= 2` (or `arc -= 3`) to
make `arc` point to before `bez_stack`. However, using pointer
subtraction to make `arc` point outside the array is undefined behavior,
and attempting to use the value in the loop predicate is "very"
undefined behavior. (C99 "Additive operators" 6.5.6.8.)
This particular undefined behavior was discovered as either hangs or
MemorySantizer issues after "[InstCombine] Infer nuw for gep inbounds
from base of object" [0]. With this change, clang can infer that `arc`
must always point into the `bez_stack` object and therefore cannot be at
a "negative index" so the predicate is always true.
[0] e21ab4d16b
* src/base/ftstroke.c (FT_Stroker_ConicTo, FT_Stroker_CubicTo): test
loop exit condition (there are no more arcs to process) before
decrementing `arc`
Fixes: #1307
When `sfnt->load_face` succeeds it has already loaded any (optional)
cmap data. As a result, a subsequent call to `sfnt->load_cmap` will
overwrite the cmap data pointer with a new copy of the data but not free
the old, leading to a leak.
This is a fix for "* src/cff/cffobjs.c (cff_face_init): Better handling
of Type0 fonts.". This still allows the cmap to be missing but avoids
the leak by only calling `sfnt->load_cmap` when there is no `head`
table (the font data is not being loaded as OpenType/CFF).
* src/cff/cffobjs.c (cff_face_init): Fix leak
Fixes: #1306
This issue relates to the PDF specification and its usage of CFF-based
embedded Type0 fonts. For Type0 fonts containing CID-keyed descendant
CFF-based fonts, the glyph index is the CID present in the content stream.
As a result, there is no requirement for a 'cmap' table to be present in the
font since the glyph indices are derived directly from the CIDs. FreeType
throws an error when it doesn’t find a 'cmap' table for such an Open Type
font containing CFF outlines. This commit relaxes this requirement for a
'cmap' table for such fonts.
