Age | Commit message (Collapse) | Author |
|
https://codereview.chromium.org/2158023002/ )
Reason for revert:
Causes heap-use-after-free. See crbug.com/647612.
Original issue's description:
> Fix memory leaking on ClosePage.
> CFX_FontCache refactoring:
> after this CL: Only one global CFX_FontCache used. Any cached items from it, are released, when its are not used.
>
> BUG=79367,48791
>
> The fonts was not cleared after unloading pages.
>
> Test pdf:
>
> http://www.nasa.gov/pdf/750614main_NASA_FY_2014_Budget_Estimates-508.pdf
>
> For this file, we have ~5 fonts per page, which equal ~1 Mb per page.
> In this PDF we have 670 pages, as result after slow scrolling(reading) full document we have ~600 Mb fonts data in memory.
>
> memory usage of PDF Plugin:
> before this CL: ~660 Mb
> after this CL: ~100 Mb
>
> Committed: https://pdfium.googlesource.com/pdfium/+/cde5101eb15b24519e89fa500fe37038bc8e2201
TBR=tsepez@chromium.org,brucedawson@chromium.org,npm@chromium.org,art-snake@yandex-team.ru
# Not skipping CQ checks because original CL landed more than 1 days ago.
BUG=79367,48791
Review-Url: https://codereview.chromium.org/2350763002
|
|
CFX_FontCache refactoring:
after this CL: Only one global CFX_FontCache used. Any cached items from it, are released, when its are not used.
BUG=79367,48791
The fonts was not cleared after unloading pages.
Test pdf:
http://www.nasa.gov/pdf/750614main_NASA_FY_2014_Budget_Estimates-508.pdf
For this file, we have ~5 fonts per page, which equal ~1 Mb per page.
In this PDF we have 670 pages, as result after slow scrolling(reading) full document we have ~600 Mb fonts data in memory.
memory usage of PDF Plugin:
before this CL: ~660 Mb
after this CL: ~100 Mb
Review-Url: https://codereview.chromium.org/2158023002
|
|
This Cl makes the Get and Set methods consistenly use {G|S}et<Type>For.
BUG=pdfium:596
Review-Url: https://codereview.chromium.org/2334323005
|
|
BUG=pdfium:562
Review-Url: https://codereview.chromium.org/2257313002
|
|
Moved classes CFX_FontCache and CFX_AutoFontCache into a separate file.
Review-Url: https://codereview.chromium.org/2246223002
|
|
Recent Skia driver addition of 8 bit src / 8 bit dst workaround
needs to check for a nullptr bitmap.
R==dsinclair@chromium.org
Review-Url: https://codereview.chromium.org/2255763002
|
|
PDFs draw bitmaps with alpha by supplying an
image alpha mask along side the image color data.
The alpha mask may not be the same scale as the
image data. Skia doesn't have a direct mapping
for this draw call, but can come close by using
the internal SkCompositeShader class.
This scales the alpha mask to match the size of
the image data, then puts both in the composite
shader to draw.
There are more things to do:
- Allow Alpha8 or Gray8 to draw scaled to
a matching device (blit doesn't work today)
Temporary code marked with
a TODO implements this functionality.
- Refactor PDFium to allow SkComposeShader's
components to take a matrix.
The latter goal will defer drawing so that PDFium
doesn't resize the mask, discarding resolution,
before Skia has a chance to record it.
Additionally fix drawing text with a negative
font size.
R=reed@google.com,dsinclair@chromium.org,herb@chromium.org
Review-Url: https://codereview.chromium.org/2182763002
|
|
files.
This is the third CL to separate fx_ge into classes, one per file.
All fx_ge.h includes had to be replaced with new includes
The method definitions for CFX_FxgeDevice were not moved to a single file.
These methods are defined in two folders different from fxge/ge, so they were left untouched for now.
Review-Url: https://codereview.chromium.org/2223213002
|
|
This is the second CL in an attempt to split up the classes in fxge/include/fx_ge.h into their own files. CFX_ClipRgn is moved to core/fxge/ge because it is only used in core/fxge. The header for CFX_PathData is left in core/fxge/include since it is used elsewhere.
Review-Url: https://codereview.chromium.org/2216853004
|
|
This is the first CL in an attempt to split up the classes in fxge/include/fx_ge.h into their own files.
Review-Url: https://codereview.chromium.org/2217663002
|
|
The text matrix for Skia is still wrong.
The last fix allowed text to draw correctly when rotated,
but did not draw correctly when skewed.
With this edit, text draws correctly rotated, skewed
horizontally, skewed vertically, and rotated and skewed.
R=dsinclair@chromium.org
Review-Url: https://codereview.chromium.org/2189093002
|
|
My initial guess for text rotation worked for scaled
but not skewed or rotated.
R=dsinclair@chromium.org,reed@google.com
Review-Url: https://codereview.chromium.org/2178623002
|
|
Add back one function which is only used by Skia code. And make a couple small
fixes.
Review-Url: https://codereview.chromium.org/2167383002
|
|
Use unique_ptr for class owned member variables. Also clean up some
style issues such as removing unused functions and casting to raw pointer.
BUG=pdfium:518
Review-Url: https://codereview.chromium.org/2163103002
|
|
testing/corpus/fx/path/path_10_jd.pdf
sets the document to null. (It can't
be read at all by MacOS Preview.)
This causes the font cache to also
be null. In this case, get the font
cache from CFX_GEModule instead.
R=dsinclair@chromium.org
Review-Url: https://codereview.chromium.org/2128043004
|
|
The dash parameters where not copied; the copy
could point at random data and cause corpus
tests to hang when testing Skia.
PDFium measures text directly by calling FreeType.
Turn off hinting altogether in Skia so that drawn
text matches the metrics that PDFium measures.
Premultiply bits retrieved from images, and check
to see that the device bits are always
premultiplied.
Look for null graphics state and matrices.
R=thestig@chromium.org,dsinclair@chromium.org
BUG=pdfium:532
Review-Url: https://codereview.chromium.org/2120353004
|
|
Add checks to pass unit tests.
R=thestig@chromium.org,dsinclair@chromium.org
BUG=pdfium:525, pdfium:526, pdfium:527, pdfium:528, pdfium:529
Review-Url: https://codereview.chromium.org/2111553003
|
|
PDFium assumes the lowest common denominator
and draws many strings and paths that can
be accumulated.
Defer canvas->restore() calls until required
because the clip changed.
Defer text and path draws as long as subsequent
calls concatenate additional data.
Include debugging switch to allow disabling cache
at compile-time while bugs are shaken out.
Review-Url: https://codereview.chromium.org/2064753002
|
|
Review-Url: https://codereview.chromium.org/2075383002
|
|
Move ifx_renderdevicedriver class to its own file to share with
multiple devices to reduce code duplication.
Review-Url: https://codereview.chromium.org/2078783002
|
|
These changes are specific to Mac and Skia builds. They are
needed for these builds to compile with clang_use_chrome_plugin.
BUG=pdfium:469
Review-Url: https://codereview.chromium.org/2081523002
|
|
Add out-of-line implementations to Skia as well.
R=weili@chromium.org,dsinclair@chromium.org
Review-Url: https://codereview.chromium.org/2072493002
|
|
Review-Url: https://codereview.chromium.org/2059883004
|
|
Use SkCanvas::drawPosText() to render the whole string
rather than drawing glyphs one at a time.
R=dsinclair@chromium.org
Review-Url: https://codereview.chromium.org/2057343002
|
|
Review-Url: https://codereview.chromium.org/2032613003
|
|
This code has several disadvantages over a more modern graphics engine:
- no SIMD support
- no GPU support
- limited quality
Further, calling this code locks in the perceived resolution, so that
the output cannot be scaled without additional loss.
By directing all bitmap drawing through
CFX_SkiaDeviceDriver::StartDIBits, Skia can handle all appropriate
bitmap optimizations.
To that end, SetDIBits and StretchDIBits now call StartDIBits.
Other changes:
Skia's bitmaps are premultiplied. PDF contains bitmaps that are
unpremultiplied. PDFium appears to use premultiplied bitmaps sometimes,
and unpremultiplied bitmaps elsewhere. Add a debug check for
unpremultiplied bits in Skia's driver, and add a utility to premultiply
PDFium's bitmaps' bits.
PDFium supports a 24 bit RGB bitmap padded to a 32 bit word. Set the
high byte so that Skia can treat this as an ARGB bitmap.
Defer the application of the alpha value to the draw call rather than
calling MultiplyAlpha where possible.
Allow the destination bitmap to be alpha 8 or argb 32.
Review-Url: https://codereview.chromium.org/2025043002
|
|
Remove unused dithering code.
Review-Url: https://codereview.chromium.org/2010813003
|
|
Fix 2/4/8 bits per pixel bitmaps
that PDFium transposes into 8 bpp
indexed.
R=tsepez@chromium.org,dsinclair@chromium.org,reed@google.com
Review-Url: https://codereview.chromium.org/2011153002
|
|
Review-Url: https://codereview.chromium.org/2009803003
|
|
Review-Url: https://codereview.chromium.org/2017503002
|
|
Resolve changes to cpdf_shadingpattern.h that prevent
Skia from building.
R=thestig@chromium.org,dsinclair@chromium.org,reed@google.com
BUG=
Review-Url: https://codereview.chromium.org/2017463002
|
|
Match the matrix supplied by PDFium to the one expected
by Skia to draw bitmaps.
Set up the color if the bitmap is an image mask.
Additional work required to figure out when bitmap
filtering should be applied.
R=reed@google.com,dsinclair@chromium.org
BUG=
Review-Url: https://codereview.chromium.org/2004203003
|
|
Review-Url: https://codereview.chromium.org/1990153003
|
|
Review-Url: https://codereview.chromium.org/1996533004
|
|
This removes the last vestiges of antigrain from
a Skia-specific build.
R=dsinclair@chromium.org,tsepez@chromium.org,thestig@chromium.org,reed@google.com
Review-Url: https://codereview.chromium.org/1998623002
|
|
PDFium uses fxge to denote the graphics engine used
for general rendering. When pdf_use_skia=1 is set,
use Skia as the default engine instead of antigrain.
R=tsepez@chromium.org,dsinclair@chromium.org
Review-Url: https://codereview.chromium.org/1975143002
|
|
R=tsepez@chromium.org,thestig@chromium.org
Review-Url: https://codereview.chromium.org/1975823002
|
|
The Skia variant generates a couple of warnings on the linux build.
One of the warning silenced is an unused variable. Once PDFs are
identified that exercise this variable, the dummy code will be
replaced.
R=tsepez@chromium.org
Review-Url: https://codereview.chromium.org/1962683002
|
|
The first PDF attempted by the Skia infra team
uncovered three bugs. This gets past those bugs
crashing; more work to do to get the correct
output.
R=rmistry@google.com,dsinclair@chromium.org
BUG=skia:5183
Review URL: https://codereview.chromium.org/1887073002
|
|
Using these webpages as guides
http://www.globelegislators.org/pdfjs/test/pdfs/alphatrans.pdf
http://www.antennahouse.com/antenna1/wp-content/uploads/2015/07/background-image-gradient-1.pdf
flesh out the gradient shaders to include the PostScript Type 0
sampling function and to include radial gradients. This CL makes
rendering these pages agree with Adobe Reader output.
Some of these examples use an Extend array to clip the gradient.
Skia does not currently support this natively, so construct the
clip manually for now.
Other PDF pages may construct gradients using alternate mechanisms
-- this code will continue to be refactored as those come to light.
In particular, this CL sets up the clip and matrix differently for
axial gradients and radial gradients -- while it makes sense to do
it one way only, I prefer to defer until I have more examples to
work with.
R=dsinclair@chromium.org,tsepez@chromium.org
Review URL: https://codereview.chromium.org/1870463002
|
|
Review URL: https://codereview.chromium.org/1864153002
|
|
This CL is a straight move of the fxge includes into core/fxge/include.
Review URL: https://codereview.chromium.org/1868533002
|
|
Copy code from the agg driver so that the Skia
driver is no longer dependent on it.
The copied code doesn't in turn call
agg; it was in the agg driver for convenience.
Much of the copied code will eventually be replaced
by native Skia code, so there's not much point
in rewriting it to make it more efficient or
maintainable.
R=dsinclair@chromium.org,tsepez@chromium.org
Review URL: https://codereview.chromium.org/1865883002
|
|
One bit bitmaps may be black and white or may be alpha masks.
Treat alpha masks as transparent/opaque instead of black/white.
R=dsinclair@chromium.org
Review URL: https://codereview.chromium.org/1858613003
|
|
Reuse the Skia typeface on sucessive text draw calls.
This reduces the SKP size by 100x for some documents.
Note that this does not use a smart pointer for the
Skia typeface object. The downside of doing so is that
it requires all clients that include fx_font.h to also
have access to Skia.
In this specific case, it is preferable to have a
forward declared class to isolate Skia from the rest of PDFium.
R=dsinclair,tsepez@chromium.org
Review URL: https://codereview.chromium.org/1837113004
|
|
Review URL: https://codereview.chromium.org/1832173003
|
|
To draw paths that are stroked and filled with a pair of colors,
reduce the fill by the width of the stroke. This is implemented
with path ops subtracting the resolved stroke shape from the
fill shape. This permits rendering the result without requiring
an offscreen bitmap.
The implementation for stroke+fill requires a new entry into
the graphics engine, so a bit was added to device caps for that.
Extract the gradient information out of the axial gradient
function descriptions, and when possible, use Skia to draw
the linear gradient directly. This requires making the function
descriptions non-opaque, and adding a bit to device caps for
another entry into the graphics engine.
BUG=
Review URL: https://codereview.chromium.org/1828283002
|
|
Use FreeType instead of Direct Write on Windows. Since PDFium
currently uses the graphics engine to draw but not to measure,
the graphics engine needs to always use FreeType to get
consistent metrics.
Add a translation from PDF's blend modes to Skia's xfer modes.
Turn on subpixel positioning for text to get more precise
glyph placement.
Pass the alpha through to get proper bitmap transparency.
Address earlier nits.
R=dsinclair@chromium.org, tsepez@chromium.org, dsinclair
Review URL: https://codereview.chromium.org/1812263002 .
|
|
More Skia driver cleanup.
Fix the GN build for Skia.
Remove unused functions from header.
Remove agg setup that is not required.
Change path construction to use the SkCanvas matrix to position the path
rather than converting the points directly.
Draw stroked paths using Skia rather than generated the filled path.
Pin the minimum stroke width to 1 px in device space to mimic PDF's stroke
dropout control.
Factor out flipped and non-flipped matrices.
Add some debugging code.
Set the bitmap filter quality to high. This helps a lot with 1 bit sources.
R=dsinclair@chromium.org, dsinclair, tsepez
Review URL: https://codereview.chromium.org/1806843002 .
|
|
This CL adds the necessary pieces so you can add pdf_use_skia=1 to GYP_DEFINES
and do a standalone build.
R=caryclark@google.com
Review URL: https://codereview.chromium.org/1812563002 .
|