The decoding loop in Decoder::Decode only pauses to report progress when it runs out of bytes to decode. So for long animated images where the network is keeping up with decoding it will be a relatively long time until we deliver the first frame complete notification and corresponding invalidation. In most cases this shouldn't be too expensive as it is just dispatching a runnable to the main thread from the decoding thread.
The decoding loop in Decoder::Decode only pauses to report progress when it runs out of bytes to decode. So for long animated images where the network is keeping up with decoding it will be a relatively long time until we deliver the first frame complete notification and corresponding invalidation. In most cases this shouldn't be too expensive as it is just dispatching a runnable to the main thread from the decoding thread.
The extra height of the test element (the embed) versus the reference element (a div with height 40px) can cause the iframe the test is run in to overflow, thus causing a scrollbar in the test but not the reference. But we are only interested in testing the frame of the contained animated gif.
This was caused by http://hg.mozilla.org/mozilla-central/rev/167ceb965079 (bug 1194059). Before that changeset mIsAnimated meant "we currently have more than one frame". After that changeset mIsAnimated was replaced with HasAnimation(). HasAnimation() just looks at the metadata to see if the image is animated. That changeset had the effect of always detected if an image is animated during the metadata decode. Therefore during a full decode we always know the image is animated, even before we've decoded two or more frames.
The fix is to go back to using the actual current frame count to manage invalidations.
This will allow MOZ_MUST_USE to be used for a different and more common case.
MozReview-Commit-ID: 4dQsdWjJfc6
--HG--
extra : rebase_source : 390ab56ef83d71eb6d28759a0195a79a78b153bd
This requires a change to how we process test manifests in the build system:
now, whenever we see a support file mentioned in a manifest, we require that
file isn't already in that test's support files, but if we see a support file
that was already seen in some other test, the entry is ignored, but it is not
an error. As a result of this change, several duplicate support-files entries
needed to be removed.
MozReview-Commit-ID: G0juyxzcaB8
--HG--
rename : testing/mozbase/manifestparser/tests/test_default_skipif.py => testing/mozbase/manifestparser/tests/test_default_overrides.py
If the image load is from the same document that cached the image we are required to use the cached version. Otherwise we should be free to ignore the cached version.
If the sync decoding the LookupFrame does encounters an error it will set mError on the RasterImage, which LookupFrame callers check before calling LookupFrame. But they've called LookupFrame before the error was encountered, so we check if the frame has had Abort called on it to determine if we should return it at all.
We only does this if one of the sync decode flags was passed in because IsAborted needs to get the imgFrame's monitor, so we don't want to block consumers that haven't asked for decoding.
This means that in RasterImage::LookupFrame when we are asked to do a sync decode (if needed) we use WaitUntilComplete to wait until the frame is finished decoding. But we would actually return after the next progressive pass notified the monitor to wake up. Thus, we would draw a not-fully-decoded image even though the sync decode flag was passed.
The change in FrameAnimator means that we won't draw the next frame in an animated image until all progressive passes of that image are complete. This seems like what we want anyways.
There is one real use of IsImageComplete left, in imgFrame::Draw, where we need to know if the decoded image data covers the whole image frame. (There are a couple of uses of IsImageComplete in asserts.)
The PNG decoder posts the size almost immediately, and later posts transparency (even for non-animated images).
It would be nice to still assert what this assert is intending (that transparency of non-animated images is posted during the metadata decode) but we don't have any easy way of telling when a metadata finishes here.
This is left over from the pre-SurfacePipe code that interacted directly with the Downscaler. It was calculating the size of the surface for the Downscaler to use, and then the Downscaler would handle putting transparent pixels inside that surface (and outside the framerect).
--HG--
extra : rebase_source : aad384fa8589f291254f0a18537a5d6674487182
The testcase has an svg-as-image inside an svg-as-image. At shutdown the viewer of the inner svg-as-image is destroyed (via the shutdown observer) first. Then the outer svg-as-image destroys its viewer which tries to unregister all image requests from the refresh driver. So it unregisters the inner svg-as-image, which calls GetAnimated.
When storing ms, 32 bit ints can hold 2^32/1000/60/60/24 ~= 49 days. It's quite conceivable that someone would leave a tab in the background for 50 days.
GetSingleLoopTime returns -1 on exceptional cases but we used an unsigned int to hold the return value in AdvanceFrame. So the |loopTime > 0| check would succeed. Fortunately the |delay.ToMilliseconds() > loopTime| check would fail because loopTime was MAX_UNIT32, so we didn't do anything incorrect.
http://hg.mozilla.org/mozilla-central/rev/263980931d1b (bug 890743) changed GetSingleLoopTime from returning 0 (and uint32_t) to -1 (and int32_t) on exceptional cases. But the caller of GetSingleLoopTime wasn't updated.
Before the previous patch we would (wrongly) loop through the decoded frames even though we didn't have all of the frames of the animation. This had the beneficial side effect of advancing mCurrentAnimationFrameTime to aTime (the current time). With the previous patch we stop at the last decoded frame and don't advance mCurrentAnimationFrameTime, so it can lag behind. The problem with this is that when we have finished decoding we will then try to catch mCurrentAnimationFrameTime up, and this will jump us to a random point in the animation. So we need to advance mCurrentAnimationFrameTime ourselves.
If we were blocked on network/decoding then displaying the last available decoded frame is the correct frame to be displaying. So we are up to date. So we advance mCurrentAnimationFrameTime to the current time.
mImage->GetNumFrames() is the current number of decoded frames (that the RasterImage knows about), so it only represents the last frame of the animation if we are done decoding.
If we are not fully decoded, and we are on the last decoded frame, just stay on the last decoded frame. When more frames get decoded (or we determine that we are the last frame of the animation) we will advance.
One might expect that if |nextFrameIndex == mImage->GetNumFrames()| then |GetRawFrame(nextFrameIndex)| would return a null surface. But that is not the case because the decoding thread can insert frames into the surface cache that the RasterImage hasn't acknowledged yet (because it has to do so on the main thread, which we are currently running on).
This is why moving animated images to the surface cache is likely the cause of this bug.
This introduces an issue that is explained in, and fixed by the next patch.
http://hg.mozilla.org/mozilla-central/rev/411f18fdffeb (bug 1186796) had a mistake in it.
It changed ImageSurfaceCache::LookupBestMatch to use a for loop instead of using a callback to iterate each entry of the hashtable. The callback was called with the surface key of its entry, and it used the name |aSurfaceKey| for that key. ImageSurfaceCache::LookupBestMatch uses the name |aSurfaceKey| for the key we are looking for. So when the code from the callback was moved into the for loop in ImageSurfaceCache::LookupBestMatch the meaning of |aSurfaceKey| changed, but the code was not updated.
In (non-animated) PNGs the image data is contained in IDAT chunks. In APNGs there are IDAT chunks, which contain the default image, and fDAT chunks, which contain frames of the animation. The default image is sometimes part of the animation (as the first frame), and sometimes not (displayed only by non-APNG aware viewers).
The default image must have the same size as in the PNG header chunk. But the fDAT images can be any (smaller) size. So the first frame of a PNG is allowed to be smaller than the whole image size so long as we are in an APNG and the first frame is from an fDAT chunk, not an IDAT chunk.
We post transparency if we encounter this case because we don't draw into those pixels on at least the first frame.
Be warned. Do not attemp to change the .js "test" source code in ./js
They are meant to check
- the outdated 0666 octal constant is still parsed correctly,
- the outdated 0666 octal constant raises syntax error flag
in strict mode, etc.
So leave them alone.
Also add crashtest with a corrupt animated PNG image that causes a leak to be reported (thanks to these MOZ_COUNT calls). This leak is fixed by the patch on separate bug 1237709.
This patch introduces TerminalState and changes LexerTransition::mNextState to
be a Variant<State, TerminalState>. This means that SUCCESS and FAILURE no
longer need to be part of State.
Some things to note:
- This simplifies the handling of Lex()'s return value, which is nice.
- The patch splits Terminate() into TerminateSuccess() and TerminateFailure().
- |const State& aNextState| wouldn't work for the first arg to
LexerTransition's ctor due to errors in Variant construction that I didn't
understand. I had to change it to |State aNextState|.
--HG--
extra : rebase_source : f405a67fdf0f1bb712409eafecb21ac9b59d6db0
This fixes failures for
image/test/reftest/bmp/bmpsuite/b/{badrle.bmp,shortfile.bmp} with the Skia
back-end.
--HG--
extra : rebase_source : 6c5b967cebf43cf5d49d0e532619bdd1c8ccc69e
gfxRect can be implicitly constructed from IntRect, which hides a number of
implicit conversion points, makes Moz2Dification harder, and has some
surprising effects.
This patch removes the implicit constructor and replaces it with an explicit
conversion function:
gfxRect ThebesRect(const IntRect&)
This is the obvious outcome of removing the constructor.
But there is also a second, less obvious outcome: currently we do a number of
IntRect-to-Rect conversions using ToRect(), which (surprisingly) works because
it turns into an implicit IntRect-to-gfxRect conversion (via the implicit
constructor) combined with an explicit gfxRect-to-Rect conversion (via
ToRect()). I.e. we do two conversions, going from a Moz2D type to a Thebes
type and back to a Moz2D type!
So this patch also changes these conversion. It moves this existing function:
Rect ToRect(const IntRect&)
from gfx2DGlue.h -- where it doesn't really belong because it doesn't involve
any Thebes types -- to gfx/2d/Rect.h, templatifying and renaming it as
IntRectToRect() in the process.
The rest of the patch deals with fall-out from these changes. The call sites
change as follows:
- IntRect-to-gfxRect conversions:
- old: implicit
- new: ThebesRect()
- IntRect-to-Rect conversions:
- old: ToRect()
- new: IntRectToRect()
--HG--
extra : rebase_source : e4e4c2ad10b36ecad4d57d1630158f3374e403be
This ID will be null for non-controlled documents and also for image cache
entries for which a document is not available, and it will be the numerical
value of the document pointer for controlled documents.
This effectively makes sure that a controlled document doesn't share its
image cache entries with anything else.
nsBMPEncoder produces either BMPv3 or BMPv5 files. (See the Version enum which
only has VERSION_3 and VERSION_5 values, and ParseOptions()'s handling of the
|version| parameter.
Nonetheless, there is some code to handle encoding of BMPv2 files. This patch
removes this.
--HG--
extra : rebase_source : eaa1ddd801872c14860e3bd81645bc99d56d8e4b
nsICODecoder's reading and writing of little-endian values can be simplified
greatly.
Also, ReadBPP() was highly dodgy: BMP's bpp field is 16-bit
but ReadBPP() read it as if it's 32-bit. I think this currently works because
the bpp field is followed by the 32-bit compression field which is usually 0
for BMPs within ICOs!
--HG--
extra : rebase_source : 5fd43dedc036dca5bc2ff79b029855dc76d62515
- GetBitsPerPixel() and GetWidth() are no longer used.
- GetHeight() is now only used within nsBMPDecoder and can be renamed and
inlined into the header.
- GetImageData() can be inlined into the header.
--HG--
extra : rebase_source : f902f7ce6513e54eaa7fe6210b4ff3ff6865c4bf
This requires delaying the creation of the BMP decoder used by the ICO decoder.
--HG--
extra : rebase_source : 629a2ac387a9c8ee1a520c70733adb10cc156aa8
The FileHeader and V5InfoHeader structs are shared by the BMP decoder and
encoder. But most of the fields within those structs are actually unused by the
decoder. It makes things clearer if we create a decoder-only struct that
contains the used fields, and then make FileHeader and V5InfoHeader only used
by the encoder. This patch does that.
This patch also renames BMPFileHeaders.h as BMPHeaders.h, which is now a better
name for it.
--HG--
rename : image/BMPFileHeaders.h => image/BMPHeaders.h
extra : rebase_source : 2227679b8aef25e48d3e8e7d38a3ba79a57c40d3
MakeUnique and its underlying |new| call will crash the program on
failure. This code was clearly written with fallible allocations in
mind, so let's make the allocations actually be fallible.
Currently we don't implement transparency at all in BMP images except for an
odd-duck case of BMPs within ICOs.
This patch does the following.
- It implements transparency properly for 16bpp and 32bpp images via bitfield
masking. (For 32bpp images this also requires handling colors via bitfield
masking.) The patch maintains the existing BMP-within-ICO transparency
handling.
- It also reworks BitFields::Value::Set().
* It now works correctly if the run of 1s goes all the way to bit 31 (the
old code didn't set mBitWidth).
* If the mask is 0, will give an mRightShift of 0 (old code gave 32, and
right-shifting by 32 is dodgy).
* It's now easier to read.
- It renames transparent.bmp as transparent-if-within-ico.bmp. Ironically
enough this file currently uses BITFIELDS compression and is WinBMPv5 format,
which means it contains well-specified alpha data. In order to use it to test
the hacky BMP-within-ICO transparency scheme the patch changes it to be
WinBMPv3 format with RGB compression (i.e. no compression). I left the
now-excess bytes (including the bitfields) in the info header in place
because that's allowed -- thanks to the start of the pixel data being
specified by the |dataoffset| field -- they'll just be ignored.
- It tweaks the naming of the relevant gtests and some of their finer details
to work with the new way of doing things.
This fixes all four remaining failures in bmpsuite.
--HG--
rename : image/test/gtest/transparent.bmp => image/test/gtest/transparent-if-within-ico.bmp
extra : rebase_source : 2f4838d04bbae4fac00cc69e8d75469105a5de3b
Currently we don't read BMP bitfields during metadata decoding. But we'll need
to in order implement alpha, because we need to know during metadata decoding
if alpha is used.
This patch moves code around to achieve this (and adds the required
mMayHaveTransparency field). The change has no noticeable effect for now.
--HG--
extra : rebase_source : 32106149bf064f0e44ec9dcf8f013612dceacbb7
The wrinkle here is that while we can pass a UniquePtr to
ConvertHostARGBRow, we can't pass UniquePtr to
EncodeImageDataRow{24,32}, as the latter get called with raw pointers
out of our control.
These functions are only used in nsBMPDecoder.cpp; we don't need them
anywhere else. The only other place they might get used would be the
BMP encoder, but the encoder appears to have its own routines for
setting pixel data, which don't overlap very well with the decoder's.
mTouchedTime is not appropriate for this as it is updated when an image
load re-uses the same imgRequest, especially as it has one second
granularity. A timestamp that is updated every time the backing file is
re-read should work better.
A millisecond granularity timestamp would be preferable, and would be
achievable on most or all supported platforms. But some older
filesystems have timestamp granularity of a second or worse, notably
ext3 and FAT32 (and even ext4 filesytems created with inode_size < 256
bytes, e.g. with 'mke2fs -t small' - see
https://ext4.wiki.kernel.org/index.php/Ext4_Disk_Layout#Inode_Timestamps
for details.)
The bulk of this commit was generated with a script, executed at the top
level of a typical source code checkout. The only non-machine-generated
part was modifying MFBT's moz.build to reflect the new naming.
CLOSED TREE makes big refactorings like this a piece of cake.
# The main substitution.
find . -name '*.cpp' -o -name '*.cc' -o -name '*.h' -o -name '*.mm' -o -name '*.idl'| \
xargs perl -p -i -e '
s/nsRefPtr\.h/RefPtr\.h/g; # handle includes
s/nsRefPtr ?</RefPtr</g; # handle declarations and variables
'
# Handle a special friend declaration in gfx/layers/AtomicRefCountedWithFinalize.h.
perl -p -i -e 's/::nsRefPtr;/::RefPtr;/' gfx/layers/AtomicRefCountedWithFinalize.h
# Handle nsRefPtr.h itself, a couple places that define constructors
# from nsRefPtr, and code generators specially. We do this here, rather
# than indiscriminantly s/nsRefPtr/RefPtr/, because that would rename
# things like nsRefPtrHashtable.
perl -p -i -e 's/nsRefPtr/RefPtr/g' \
mfbt/nsRefPtr.h \
xpcom/glue/nsCOMPtr.h \
xpcom/base/OwningNonNull.h \
ipc/ipdl/ipdl/lower.py \
ipc/ipdl/ipdl/builtin.py \
dom/bindings/Codegen.py \
python/lldbutils/lldbutils/utils.py
# In our indiscriminate substitution above, we renamed
# nsRefPtrGetterAddRefs, the class behind getter_AddRefs. Fix that up.
find . -name '*.cpp' -o -name '*.h' -o -name '*.idl' | \
xargs perl -p -i -e 's/nsRefPtrGetterAddRefs/RefPtrGetterAddRefs/g'
if [ -d .git ]; then
git mv mfbt/nsRefPtr.h mfbt/RefPtr.h
else
hg mv mfbt/nsRefPtr.h mfbt/RefPtr.h
fi
--HG--
rename : mfbt/nsRefPtr.h => mfbt/RefPtr.h
This commit was generated using the following script, executed at the
top level of a typical source code checkout.
# Don't modify select files in mfbt/ because it's not worth trying to
# tease out the dependencies currently.
#
# Don't modify anything in media/gmp-clearkey/0.1/ because those files
# use their own RefPtr, defined in their own RefCounted.h.
find . -name '*.cpp' -o -name '*.h' -o -name '*.mm' -o -name '*.idl'| \
grep -v 'mfbt/RefPtr.h' | \
grep -v 'mfbt/nsRefPtr.h' | \
grep -v 'mfbt/RefCounted.h' | \
grep -v 'media/gmp-clearkey/0.1/' | \
xargs perl -p -i -e '
s/mozilla::RefPtr/nsRefPtr/g; # handle declarations in headers
s/\bRefPtr</nsRefPtr</g; # handle local variables in functions
s#mozilla/RefPtr.h#mozilla/nsRefPtr.h#; # handle #includes
s#mfbt/RefPtr.h#mfbt/nsRefPtr.h#; # handle strange #includes
'
# |using mozilla::RefPtr;| is OK; |using nsRefPtr;| is invalid syntax.
find . -name '*.cpp' -o -name '*.mm' | xargs sed -i -e '/using nsRefPtr/d'
# RefPtr.h used |byRef| for dealing with COM-style outparams.
# nsRefPtr.h uses |getter_AddRefs|.
# Fixup that mismatch.
find . -name '*.cpp' -o -name '*.h'| \
xargs perl -p -i -e 's/byRef/getter_AddRefs/g'
This patch implements proper color-scaling, instead of bit-shifting, and uses
it for 16bpp images.
It also cleans up the code relating to color masking in the process, by making
BitFields a proper class and introducing the Value class within it.
This fixes sub-optimal handling of four images in bmpsuite.
Two-space indents is the Gecko standard, and it's what nsBMPDecoder.cpp uses.
This patch changes nsBMPEncoder.h to use two-space indents as well. This will
help avoid possible mis-indentation if code is moved from the .h to the .cpp or
vice versa (as subsequent patches in this bug will do).
Also, it changes some of the // comments on public methods to doxygen-style ///
comments.
We have 52 passes and 5 known fails. Three of the passes have higher fuzziness
allowances than they should, so really there are 8 files that we need to
improve on.
--HG--
extra : rebase_source : 23738272b38c7d03c90e425e8170fc3fabc4c021
This patch is a major overhaul of nsBMPDecoder.
The patch improves the code in the following ways.
- It converts nsBMPDecoder to use StreamingLexer, which makes it much easier to
read.
- It adds a detailed comment about the BMP format at the top of
nsBMPDecoder.cpp.
- It fixes lots of inconsistent indenting.
- It moves |bihsize| from |mBFH| to |mBIH| to match the file format and common
sense. The avoids the need for the confusing LENGTH/INTERNAL_LENGTH
distinction.
- It renames most of the types in BMPFileHeader.h, so they have better names,
in StudlyCaps form, and within the new |bmp| namespace.
- It removes the BMP_HEADER_LENGTH struct and inlines its values directly into
the two places they were used.
- It removes the MOZ_LOG logging done on some of the failure cases. (Most
failure cases lacked logging so why bother with some?)
- It removes over 200 lines of code, despite the addition of the big format
comment.
The patch changes the way BMPs are decoded as follows.
- It adds stricter testing of the InfoHeader length, rejecting files with bad
values.
- It moves all header sanity checking that can lead to file rejection into the
metadata decode phase. (Previously, bpp/compression consistency checking did
not occur during a metadata decode.)
- It removes BMPINFOHEADER::ALPHABITFIELDS, which was (a) a weird WinCE-only
thing, and (b) we didn't actually allow it, and (c) we used the value 4
instead of 6(!).
- It rejects the previously-accepted compression==RLE4 && bpp=1 combination
because it doesn't make sense.
- It removes a fudge in RLE absolute mode handling that permitted one pixel too
many in a row but only if the row's width was odd(!)
- It now rejects a file with a negative gap between the color table and the
pixel data.
The patch leaves the following problems unaddressed.
- If bpp==32 we totally ignore compression==BITFIELDS and treat it like
compression=RGB.
- Transparency as specified in WinBMPv{4,5} isn't handled at all.
These will be fixed in follow-ups.
All these changes affect (for the better) the results of the following tests
that will be added in part 2:
- g/pal8v4.bmp
- g/pal8v5.bmp
- q/pal8os2sp.bmp
- q/pal8os2v2.bmp
- q/pal8os2v2-16.bmp
- b/badheadersize.bmp
- b/badpalettesize.bmp
- b/badrle.bmp
--HG--
extra : rebase_source : 8ddc2f5fccce6998348097ff9f0a1072d273cdf4
|getReportsForThisProcess| differs from |getReports| in that it is limited to current process and is synchronous. When asynchronous memory reporters are added the function will no longer be able tobe synchronous. There isn't much utility in only measuring the current process, so we can remove the function and switch existing users to |getReports|.
The conversion is as follows:
- GraphicsFilter::FILTER_NEAREST == gfx::Filter::POINT
- GraphicsFilter::FILTER_GOOD == gfx::Filter::GOOD
- GraphicsFilter::FILTER_BEST == gfx::Filter::LINEAR
Also typedef GraphicsFilter to gfx::Filter; this will be removed in the next
patch.
These changes mean ToFilter() and ThebesFilter() are no longer needed.
Seth Fowler
Daniel Holbert
Carsten "Tomcat" Book
Seth Fowler
Jonathan Watt
Eric Rahm
vincentliu
Chris Peterson
Timothy Nikkel
Sebastian Hengst
JerryShih
Nathan Froyd
Christoph Kerschbaumer
Robert Longson
Joel Maher
Kyle Huey
Georg Fritzsche
Aryeh Gregor
Nicholas Nethercote
Christoph Kerschbaumer
Yoshi Huang
Xidorn Quan
David Newton
Kris Maglione
Dragana Damjanovic
Heiher
Milan Sreckovic
Chris Manchester
Ryan VanderMeulen
Kartikaya Gupta
Geoff Brown
Mike Hommey
Edwin Flores
Bogdan Postelnicu
CJKu
Matt Woodrow
Martijn Wargers
Mike Taylor
Mason Chang
ISHIKAWA, Chiaki
Aidin Gharibnavaz
Birunthan Mohanathas
Phil Ringnalda
Jeff Muizelaar
David Rajchenbach-Teller
Jan de Mooij
Steve Singer
Morris Tseng
Wes Kocher
Glenn Randers-Pehrson
sajitk
David Anderson
Nathan Froyd
Sotaro Ikeda
Ehsan Akhgari
Oliver Henshaw
Hiroyuki Ikezoe
Tom Klein