M120 public #210
M120 public #210
Conversation
…t/0.38.2-2393-g746dbd8c61 (m120)
Milestone 120
-------------
`SkTypeface::UniqueID()` has been removed - clients should use the method instead of this static
function.
… removed
Milestone 120
-------------
* `SkFont::refTypefaceOrDefault` and `SkFont::getTypefaceOrDefault()` have been removed from the
public API.
Milestone 120
-------------
* `GrBackendSemaphore::initGL` and `GrBackendSemaphore::glSync` have been removed
from the public API.
Milestone 120
-------------
* `GrDirectContext::MakeVulkan...` has been moved to `GrDirectContexts::MakeVulkan...` which are defined
in `include/gpu/ganesh/vk/GrVkDirectContext.h`
Header diff:
$ diff -u skia-m119/src/core/SkMipmapBuilder.h skia-m120/src/core/SkMipmapBuilder.h
--- skia-m119/src/core/SkMipmapBuilder.h 2023-10-06 23:22:04.000000000 +0100
+++ skia-m120/src/core/SkMipmapBuilder.h 2023-11-01 22:34:27.000000000 +0000
@@ -27,7 +27,7 @@
* If these levels are compatible with src, return a new Image that combines src's base level
* with these levels as mip levels. If not compatible, this returns nullptr.
*/
- sk_sp<SkImage> attachTo(sk_sp<const SkImage> src);
+ sk_sp<SkImage> attachTo(const sk_sp<const SkImage>& src);
private:
sk_sp<SkMipmap> fMM;
The circular dependencies are only in Windows and Linux, from the OT-SVG hook.
Comments on SVG test, and make it pass
… defaults to skia_use_freetype &&) skia_use_freetype=true is not automatic with skia_enable_fontmgr_custom_empty=true Apparently it still tries to read headers from system locations on non-linux! Need skia_use_system_freetype2=false also.
Increase tolerance to make test_fontmgr_custom_svg_blob_bounds pass on windows
… flag = 10.13. This is a follow-up to an earlier change - we should keep compile-flag and link-flag minimum version in sync. commit ea0e2cd Author: Hin-Tak Leung <[email protected]> Date: Fri Aug 4 06:50:32 2023 +0100 CI failure log says Skia internals start to require Mac OS X 10.13.
…sh/SkImageGanesh.h These should have been part of the earlier commit (adding a misssing new header): commit deebf369ba0cdc1b2bc825e3ebbdbe0499377a3f Author: Jonathan Hogg <[email protected]> Date: Mon Aug 14 14:50:21 2023 +0100 Get `Image.MakeFromTexture` working again
Conflicts: tests/test_shader.py
|
The segfaults were caused by the withdrawn GL-related API, and actually GrBackendSemaphore only got SK_API status with m120, and GL-backend with this wiil never work. so having the GL-based methods earlier was a mistake. That said, apparently it should be possible if vulkan build is enabled, and should work if the GrBackendSemaphore is initialized with a Vulkan backend: #219 . @kyamagu I think this is missing a README.m120 but code-wise okay to go, as m121 should be out soon. I'll get on with a README summary soon. Not sure if we should include #219 in this? |
…texts::MakeVulkan...`
Milestone 120
-------------
* `GrDirectContext::MakeVulkan...` has been moved to `GrDirectContexts::MakeVulkan...` which are defined
in `include/gpu/ganesh/vk/GrVkDirectContext.h`
Adjust for recent m118/120 vulkan changes
Milestone 118
-------------
* Vulkan-specific calls are being removed from GrBackendSurface.h. Clients should use the
equivalents found in `include/gpu/ganesh/vk/GrVkBackendSurface.h"`
GrBackendFormats::MakeVk now has an optional argument "willUseDRMFormatModifiers".
…kend is compiled in
Milestone 98
------------
* GrBackendSemaphore only includes methods that match the GPU backend that Skia was compiled for.
For example, initVulkan and vkSemaphore are not defined unless the Vulkan backend is compiled
into Skia.
|
@kyamagu for vulkan related tests, it is mostly just re-enabling them. However, before m98 they were available for mac os (but don't work) , now they are not available on mac os x. HinTak/skia-m1xx-python@3d69f9b so I expect those should pass on Linux (and it does on my machine) , and Windows (let's see), but fails on mac os. So those REVISIT would get converted to "if (mac os) skip/xfail" eventually - hence running the tests on CI on my playground repo. |
…sr/include Locally, it is the "vulkan-headers" package.
|
@kyamagu added README.120 , and ready to go. I have an extra 14 tests in my m120-try branch but the extra tests require extra test fonts not in skia/resources/fonts (AFAIC - I could check that again eventually) so I don't want to add them here. Let me know if you want them (one is a 10MB font file...) - otherwise I'll just keep them as my own private addition. Ie. The new API are still tested, but not as often as the other tests officially added here. A bit curious about the state of vulkan support at m87 - mostly I am just re-enabling routines and tests already in m87 which became optional/ dependent on actual skia compiled option in m98. So I think they were present but not necessarily functional, and the tests themselves weren't/aren’t really testing the correct things, as in how people use those practically. I'd say add more vulkan tests but also occasionally backport the tests to m87 and run them on CI and see how they did in m87. Upstream has a sentence saying vulkan build is feature comparable to GL build but somewhat buggy - as mentioned in the other issue, Linux has both, Windows seems to be vulkan only (at least on Github CI), mac with newer python is opengl only, mac with older python have neither - this 4 different behavior is according to current CI. |
The commit which disabled it, and the one which made it start to work, are below: commit 15dd24a Author: Hin-Tak Leung <[email protected]> Date: Mon Jul 31 21:50:05 2023 +0100 These attributes are not (yet) available under m116; may never be. RE-VISIT!!! commit 0a62144 Author: Hin-Tak Leung <[email protected]> Date: Tue Aug 1 04:43:05 2023 +0100 m113: `SkImage::CompressionType` has been renamed to `SkTextureCompressionType` Milestone 113: * `SkImage::CompressionType` has been renamed to `SkTextureCompressionType` and moved to `include/core/SkTextureCompressionType.h`
|
@HinTak Thanks for the tough work. As for the Vulkan support in m87, the implementation was experimental and never expected to work as expected. Those were there to check the possibility of supporting multiple backends. I appreciate your rework on those APIs |
|
As of now, Compared to v87.5 |
|
Damn - it is "RTL" languages in the release note! |
The code changes from m119 to m120 are all obvious, most of them actually mentioned in the top-level release notes. So almost all the changes in the tests are obvious too, just skipping tests involving withdrawn APIs. I don't think those need revisiting, as they are all explicitly mentioned as withdrawn in the release note. Whether we want to emulate them is another matter.
However, somehow there are two new segfaults. I think they are indirectly related / dependent on those withdrawn ones, so they needs re-visiting, either enable them or mark them as indirectly withdrawn.
So this should pass CI, but I'd say those two segfaults need to be either properly skip as withdrawn or re-enabled, before this is merged.