RFE: Support building without FFmpeg #3
Comments
|
Hi, good point. FFmpeg backend is integral part of Chromium and it will be challenging to disable it compile time, but we should definitely consider it. May be it could be done as a part of the refactoring once we introduce media process in upstream Chromium. |
|
I suppose the first step would be to set media_use_ffmpeg=0 and comment out the runtime backend selection, and see what breaks, right? |
|
(But that would still try to build a lot of code that isn't actually going to be used, wouldn't it? As far as I can see, your backend plugs in at a much higher level than where the MEDIA_DISABLE_FFMPEG defines are.) |
|
Yes, you are right. If you use MEDIA_DISABLE_FFMPEG it will disable ffmpeg decoder (FFmpegVideoDecoder) for the renderer and it shouldn't compile FFmpeg license code. But it can still use default media player (WebMediaPlayerImpl) with other supported decoders like gpu accelerated (GpuVideoDecoder). And as you said we plug in few levels up and provide our own implementation of blink::WebMediaPlayer that why I was talking about possible future refactoring which would split WebMediaPlayerImpl code into render and media process parts similar to ours. But it looks like you can already disable ffmpeg and use our backend. |
This CL parses the report-uri attribute on HPKP headers and stores them in TransportSecurityPersister. This is CL #1. CL #2: crrev.com/1212973002 (add net::CertificateReportSender) CL #3: crrev.com/1212613004 (build and send HPKP reports) BUG=445793 Committed: https://crrev.com/1320e36d908427d615357df1630348bfb38cb5c4 Cr-Commit-Position: refs/heads/master@{#339667} Review URL: https://codereview.chromium.org/1211363005 Cr-Commit-Position: refs/heads/master@{#340490}
Summary of changes available at: https://chromium.googlesource.com/skia/+log/9a5d1ab..a3a9ebc CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel [email protected] Commits in this roll: a3a9ebc [email protected] documentation: more API details and examples. 6e9aed9 [email protected] Fixing src rect constraint support for drawImage with SkPicture bca1400 [email protected] remove pixel assert from ctable validator e9d6095 [email protected] Cleanup Default Geo Proc API 2c32342 [email protected] No one calls SkXfermode::GetProc16 56b78a7 [email protected] Revert of Lay groundwork for SkOpts. (patchset #3 id:40001 of https://codereview.chromium.org/1255193002/) 513f265 [email protected] fix up GrImmediateDrawTarget.cpp 58fd2c8 [email protected] Remove sk_memcpy32 ce2c505 [email protected] Lay groundwork for SkOpts. Review URL: https://codereview.chromium.org/1258073004 Cr-Commit-Position: refs/heads/master@{#340579}
…40001 of https://codereview.chromium.org/1220223002/) Reason for revert: Looks like this doesn't quite work on Windows: http://build.chromium.org/p/tryserver.chromium.win/builders/win8_chromium_ng/builds/21206/steps/compile%20%28with%20patch%29/logs/stdio Original issue's description: > Add output reference to gen written files. > > By adding the outputs written at gn gen time to the build.ninja representation the build tool (ninja) can know how to recreate them if and when necessary. > > For example, if one or more of the <blah>.tmp files is removed, even as part of an entire hierarchy/subtree currently the build fails because ninja does not know how to re-make it. > > BUG=469621 > TEST=Build time only. See https://code.google.com/p/chromium/issues/detail?id=469621#c3 > > Committed: https://crrev.com/517ac52f86903beb206b19e6ddeedbc709b253a1 > Cr-Commit-Position: refs/heads/master@{#338929} [email protected],[email protected] NOPRESUBMIT=true NOTREECHECKS=true NOTRY=true BUG=469621 Review URL: https://codereview.chromium.org/1255113004 Cr-Commit-Position: refs/heads/master@{#340596}
net::CertificateReportSender contains code factored out of CertificateErrorReporter in //chrome. net::CertificateReportSender will be used for both HPKP violation reports and the Safe Browsing Extended Reporting cert reports that CertificateErrorReporter handles. CL #1: crrev.com/1211363005 (parse report-uri) This is CL #2. CL #3: crrev.com/1212613004 (build and send HPKP reports) BUG=445793 Review URL: https://codereview.chromium.org/1212973002 Cr-Commit-Position: refs/heads/master@{#340641}
This CL adds code to TransportSecurityState to build HPKP reports, and sends them with a CertificateReportSender constructed by ProfileIOData. Calls to CheckPublicKeyPins() indicate whether a report should be sent and pass necessary reporting information as arguments. CL #1: crrev.com/1211363005 (parse report-uri) CL #2: crrev.com/1212973002 (add net::CertificateReportSender) This is CL #3. BUG=445793 Review URL: https://codereview.chromium.org/1212613004 Cr-Commit-Position: refs/heads/master@{#340687}
…d:60001 of https://codereview.chromium.org/1239363002/) Reason for revert: Breaks various fast/events and fast/scrolling layout tests on Mac >= 10.8. https://code.google.com/p/chromium/issues/detail?id=514704 Original issue's description: > Pass NSUserDefaults over IPC to the renderer > > Blink can now accept the NSUserDefaults values via IPC. This change causes the main process to pass the default values over IPC using the new API. > > BUG=392141 > > Committed: https://crrev.com/54113566ce626a9fcfe5fd1dd46f0362d01ed8ed > Cr-Commit-Position: refs/heads/master@{#340695} [email protected],[email protected],[email protected],[email protected] NOPRESUBMIT=true NOTREECHECKS=true NOTRY=true BUG=392141 Review URL: https://codereview.chromium.org/1261933003 Cr-Commit-Position: refs/heads/master@{#340714}
In order to roll GN ec34450ee9..2f763b5fef (r338735:r340609) and pick up the following changes: 61fb682 Revert of Add output reference to gen written files. (patchset #3 id:40001 of https://codereview.chromium.org/1220223002/) 177c1e2 Add GN isolate support for a bunch of unittests. 295e072 Count data deps even if it was already a regular dep 45884a2 Make sure GN deps rolls include the optional GN bots. 97ec9ac Add a GN auto-roller script. 9550931 Remove legacy StartsWithASCII function. 925b216 Fix broken link and add cross-compiling to quick start a7ff1b2 Remove some legacy versions of StartsWith and EndsWith. ddbde8f Revert of Remove some legacy versions of StartsWith and EndsWith. (patchset #6 id:100001 of https://codereview.chromium.org/1239493005/) edce9a3 Remove some legacy versions of StartsWith and EndsWith. 517ac52 Add output reference to gen written files. d673f55 Add missing source to gn bootstrap. [email protected] CQ_EXTRA_TRYBOTS=tryserver.chromium.mac:mac_chromium_gn_rel,mac_chromium_gn_dbg;tryserver.chromium.win:win8_chromium_gn_dbg,win_chromium_gn_x64_rel Review URL: https://codereview.chromium.org/1260153002 Cr-Commit-Position: refs/heads/master@{#340781}
…s (patchset #1 id:1 of https://codereview.chromium.org/1244243003/) Reason for revert: It turns out this CL isn't the culprit of crashes. Let's reland. Original issue's description: > Revert of Call EnsureWebKitInitialized() before registering extensions (patchset #3 id:40001 of https://codereview.chromium.org/1182083006/) > > Reason for revert: > This seems the cause of many crashes > > Original issue's description: > > Call EnsureWebKitInitialized() before registering extensions > > > > Blink needs to be initialized before registering an extension because: > > - WebScriptController::registerExtension() allocates an WTF::Vector on the first > > call. > > - WTF::Vector uses PartitionAlloc. > > - PartitionAlloc's partitions needs to be initialized before allocating memory. > > Blink initialization does the job. > > > > Before this CL, partitions are initialized lazily but lazy initialization > > doesn't provide proper histogram function, which might be the cause of missing > > report for PartitionAlloc.CommittedSize UMA. > > > > BUG=501171 > > > > Committed: https://crrev.com/0d244fa5e106b3d146655a2c623539a3e1900831 > > Cr-Commit-Position: refs/heads/master@{#339200} > > [email protected],[email protected],[email protected],[email protected],[email protected],[email protected] > NOPRESUBMIT=true > NOTREECHECKS=true > NOTRY=true > BUG=501171 > > Committed: https://crrev.com/2b94b05fc252c9a006ae924d0fe75ed3abecbeb2 > Cr-Commit-Position: refs/heads/master@{#339777} [email protected],[email protected],[email protected],[email protected],[email protected],[email protected] NOPRESUBMIT=true NOTREECHECKS=true NOTRY=true BUG=501171 Review URL: https://codereview.chromium.org/1258103005 Cr-Commit-Position: refs/heads/master@{#340793}
It's currently possible to "lose" a tab-modal dialog (e.g. print preview) when dragging multiple tabs off a browser. Mac's ConstrainedWindowManager doesn't support reparenting tab-modal dialogs, so dragging of tabs with modal dialogs must be prevented. BUG=514937 TEST=Have 3 tabs, press Cmd+P in #3 get a print preview, Shift+Click tab #2 (release mouse button), now drag tab #2 off the window. The whole window should drag (tabs can not be detached with dialogs). Before this change, the tabs would come off and the dialog would be lost. Review URL: https://codereview.chromium.org/1261513005 Cr-Commit-Position: refs/heads/master@{#341239}
…es. (patchset #3 id:40001 of https://codereview.chromium.org/1254423006/) Reason for revert: Speculative revert. This seems to have broken BrandColorTest.testBrandColorInterstitial BrandColorTest.testNavigatingToBrandColorAndBack BrandColorTest.testNoBrandColor on Lollipop in Document Mde. Original issue's description: > Add getThemeColor to Tab and add plumbing for ChromeActivites. > > Moves all theme color computing logic to tab. Adds the necessary > update signals to ChromeActivity and adds an onUpdate call to be > overridden by extending classes. > > The main outcome of all this is, all ChromeActivities become color change aware. Doesn't add any visual changes. Yet. > > BUG=507340 > > Committed: https://crrev.com/25d23c547cd7661fe2e679d74e4272917dae2985 > Cr-Commit-Position: refs/heads/master@{#341226} [email protected],[email protected] NOPRESUBMIT=true NOTREECHECKS=true NOTRY=true BUG=507340 Review URL: https://codereview.chromium.org/1260523004 Cr-Commit-Position: refs/heads/master@{#341332}
Summary of changes available at: https://chromium.googlesource.com/skia/+log/0aff2fa..8db6fdc CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel [email protected] Commits in this roll: 8db6fdc [email protected] bump the size of the atlas id to 64 bits 1421aee [email protected] Make SkIsPow2 templated bdc91a7 [email protected] Revert of Modifying TextBlobCacheTest to use SkRandomScalerContext (patchset #3 id:40001 of https://codereview.chromium.org/1266003002/) adcdca8 [email protected] Modifying TextBlobCacheTest to use SkRandomScalerContext Review URL: https://codereview.chromium.org/1256703005 Cr-Commit-Position: refs/heads/master@{#341356}
Summary of changes available at: https://chromium.googlesource.com/skia/+log/8db6fdc..f96bee3 CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel [email protected] Commits in this roll: f96bee3 [email protected] disable SkOpts on x86 iOS (simulator) 853336c [email protected] adding gm to use random scaler context aa80e42 [email protected] Fix straggling SK_VIRTUAL_CONSTRAINT_TYPEs da9ccf1 [email protected] Revert of Move strike to subrun in GrAtlasTextContext (patchset #3 id:40001 of https://codereview.chromium.org/1257253005/) bdb34d0 [email protected] Move SkOpts.h back to src/core. eae6200 [email protected] Some cleanup in GrTextureProvider and GrResourceProvider. 77d89f7 [email protected] Move strike to subrun in GrAtlasTextContext 7568d0b [email protected] C API: add sk_xfermode.h, impl, test 85cd78d [email protected] Speculative fix for http://crbug.com/515966 ec994b6 [email protected] Documentation spelling error 82314e9 [email protected] Documentation: SkXfermode::Mode d1ebe06 [email protected] Another small fix to GrFontScaler 383ff10 [email protected] Remove unnecessary virtual destructor on SkTArray 490b615 [email protected] Port SkXfermode opts to SkOpts.h 685f277 [email protected] use SkNextID::ImageID for android's stable id 65e96b4 [email protected] Modifying TextBlobCacheTest to use SkRandomScalerContext 88c7b98 [email protected] Make ANGLE perf decisions be runtime rather than compile time 7eb0945 [email protected] Port SkUtils opts to SkOpts. 5119ac0 [email protected] Update Android Testing apps to support release mode b411b3b [email protected] Restore read pixels perf on ANGLE 96a52c8 [email protected] remove drawimagerect flags, as skia no longer respects them Review URL: https://codereview.chromium.org/1260793009 Cr-Commit-Position: refs/heads/master@{#341448}
Summary of changes available at: https://chromium.googlesource.com/skia/+log/e003680..1d183b4 CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel [email protected] Commits in this roll: 1d183b4 [email protected] Revert of SkCanvas::onDrawPicture() quick-reject (patchset #3 id:40001 of https://codereview.chromium.org/1264133003/ ) 48ed62b [email protected] SkCanvas::onDrawPicture() quick-reject Review URL: https://codereview.chromium.org/1268983004 Cr-Commit-Position: refs/heads/master@{#341682}
…app is already installed. (patchset #3 id:40001 of https://codereview.chromium.org/1259543002/ ) Reason for revert: breaks browser_tests on Win7: https://build.chromium.org/p/chromium.win/builders/Win7%20Tests%20(dbg)(1)/builds/40480/steps/browser_tests/logs/AppBannerDataFetcherBrowserTest.WebAppBannerCreatedDirect Original issue's description: > Make desktop banner prompts respect bypass flag when a web app is already installed. > > Desktop app banners are currently not shown if the web app has already > been added to the shelf. This should be overruled by > --bypass-app-banner-engagement-checks > > BUG=513582 > [email protected] > > Committed: https://crrev.com/98d5901a6cd71bfdca36e92db2f51015df000ba8 > Cr-Commit-Position: refs/heads/master@{#341671} [email protected],[email protected] NOPRESUBMIT=true NOTREECHECKS=true NOTRY=true BUG=513582 Review URL: https://codereview.chromium.org/1265233002 Cr-Commit-Position: refs/heads/master@{#341693}
…(patchset #3 id:40001 of https://codereview.chromium.org/1267243003/ ) Reason for revert: Test flaking on dowsntream ToT due to deferred startup not finishing. Original issue's description: > Add custom tabs tests using intents with non-null sessions > > All current tests either test the service side session creation without > launching intents or the UI customization with no sessions. This change > add a trivial test that uses a valid session to launch an intent and also > another more complex test that reloads a new URL in the same tab when the > same session is used inside the intent. > > Note that the service side is interfaced directly through CustomTabConnection > and this doesn't exactly replicate the actual scenario since all calls come > from the same process. > > Committed: https://crrev.com/a69c3f7ba69a30aef2143945b8b2bbf8e6421080 > Cr-Commit-Position: refs/heads/master@{#342131} [email protected] NOPRESUBMIT=true NOTREECHECKS=true NOTRY=true Review URL: https://codereview.chromium.org/1276993002 Cr-Commit-Position: refs/heads/master@{#342169}
|
Simply add: media_use_ffmpeg=0 media_use_libvpx=0 media_use_libwebm=0 to the GYP_DEFINES when executing the gyp configuration command. (I need to mention it in our README.md) Btw why FFmpeg is not allowed on Fedora ? |
FFmpeg contains certain algorithms (decoders, encoders) that can be patent encumbered which can't be packaged in Fedora either due to the licensing guidelines: https://fedoraproject.org/wiki/Licensing:Main?rd=Licensing#Patented_Software (yet you can get them installed via other repositories, such as RPMFusion) Additionally, another good reason to push for being able to build chromium without the internal FFmpeg it bundles is that Fedora has a strict policy when it comes to bundled libraries which makes it really hard to package Chromium, and not just because of FFmpeg, but also for the rest of libraries it bundles: https://fedoraproject.org/wiki/Packaging:No_Bundled_Libraries |
Because it is a patent minefield. In principle, the patent-encumbered stuff can be ripped out, but in practice, shipping that would be a disservice to our users because its monolithic structure does not allow plugging in additional codecs from third-party repositories, you have to recompile the whole library. Sadly, neither FFmpeg upstream nor Libav upstream are in any way interested in making codecs pluggable. There were even proof of concept patches for codec plugins that got submitted, they all got rejected because they don't like the idea as a principle. There are Chromium packages in Copr repositories (our version of PPAs) for Fedora, but those ship a bundled FFmpeg with almost all codecs ripped out. Bundled libraries are also not allowed in Fedora in most cases, which is why those packages are only in Copr repositories, and they also suffer from the "can't add more codecs" problem I described above. |
|
Thx for the detailed explanations. Make perfectly sense. I also like the argument about "Bundled libraries are also not allowed in Fedora in most cases" Then for chromium it is not only ffmpeg you should disable but at least also libvpx and libwebm. Theoretically it should be enough to add "media_use_ffmpeg=0 media_use_libvpx=0 media_use_libwebm=0" to the GYP_DEFINES build command. I tried but I got a few build failures which I fixed: 94e7010 So now you should be able to completely disable ffmpeg. (A side note about HW decoders in chromium, they are not available on desktop linux, only on some Embedded configurations, iOS, Android and ChromeOS. With our on-going zero-copy work our GStreamer backend will be able to use HW decoders that come from gstreamer plugins packages, like gstvaapidecode, gstomxvideodecoder and gstv4l2videodecoder, and having zero-copy working) |
|
Worth to mention that the patch mentioned in previous comment in under review here: https://codereview.chromium.org/1415793003 |
It should be possible to compile Chromium entirely without the FFmpeg backend (only with GStreamer), not only either with both (runtime-selectable) or only with FFmpeg. This is especially important for us in Fedora because FFmpeg is not allowed in Fedora.
The text was updated successfully, but these errors were encountered: