Fixing the strong imported async frame pointer flags #40035

etcwilde · 2021-11-03T17:25:22Z

The weakly-imported symbol was getting optimized out, then put back in
as a strongly-imported symbol. This is no good. The symbol is a
declaration though, so it can't be "used" directly. Instead, we assign
it to another global and "use" it. That avoids the optimizations and
should be fine. Even if that symbol is a nullptr because it doesn't
exist, we are taking the pointer to it, which should be fine for all
situations.

Fixes rdar://84877644.

etcwilde · 2021-11-03T17:25:29Z

@swift-ci please test

swift-ci · 2021-11-03T17:38:14Z

Build failed
Swift Test Linux Platform
Git Sha - 63599b7f9bd2080d2a594a060fa8aa34e3901e98

etcwilde · 2021-11-03T17:52:57Z

@swift-ci please test

lib/IRGen/IRGenModule.cpp

etcwilde · 2021-11-03T18:18:10Z

@swift-ci please test

swift-ci · 2021-11-03T19:30:20Z

Build failed
Swift Test Linux Platform
Git Sha - 5ffd77ea4f7a99530f896d8aa30f3b47a57625ed

etcwilde · 2021-11-03T19:34:14Z

Linux failures are due to the compile module being created by an older version of the compiler.

etcwilde · 2021-11-03T19:34:26Z

@swift-ci please clean test Linux

swift-ci · 2021-11-03T20:11:40Z

Build failed
Swift Test OS X Platform
Git Sha - 5ffd77ea4f7a99530f896d8aa30f3b47a57625ed

DougGregor · 2021-11-04T04:47:01Z

@swift-ci please test macOS

swift-ci · 2021-11-04T07:16:49Z

Build failed
Swift Test OS X Platform
Git Sha - 5ffd77ea4f7a99530f896d8aa30f3b47a57625ed

etcwilde · 2021-11-04T16:00:31Z

The macOS failure is due to other symbols being included in the used list. I really just want to verify that (i8*** @_swift_async_extendedFramePointerFlagsUser to i8*) is in the list.

The full list on the bot is

appending global [10 x i8*] [i8* bitcast ({ i32 }* @"\01l_entry_point" to i8*),
i8* bitcast (void (%swift.context*)* @"$s4main1fyyYaF" to i8*),
i8* bitcast (%swift.async_func_pointer* @"$s4main1fyyYaFTu" to i8*),
i8* bitcast (void (%swift.context*)* @"$s4main1gSSyYaF" to i8*),
i8* bitcast (%swift.async_func_pointer* @"$s4main1gSSyYaFTu" to i8*),
i8* bitcast (i16* @__swift_reflection_version to i8*),
i8* bitcast (void ()** @"_swift_FORCE_LOAD_$_swiftCompatibility51_$_main" to i8*),
i8* bitcast (void ()** @"_swift_FORCE_LOAD_$_swiftCompatibilityConcurrency_$_main" to i8*),
i8* bitcast (i8*** @_swift_async_extendedFramePointerFlagsUser to i8*),
i8* bitcast (i32 (i32, i8**)* @main to i8*)],
section "llvm.metadata"

compared to the following list that I had in the test

appending global [5 x i8*] [i8* bitcast ({ i32 }* @"\01l_entry_point" to i8*), 
i8* bitcast (i16* @__swift_reflection_version to i8*), 
i8* bitcast (void ()** @"_swift_FORCE_LOAD_$_swiftCompatibilityConcurrency_$_test" to i8*), 
i8* bitcast (i8*** @_swift_async_extendedFramePointerFlagsUser to i8*),
i8* bitcast (i32 (i32, i8**)* @main to i8*)],
section "llvm.metadata"

The weakly-imported symbol was getting optimized out, then put back in as a strongly-imported symbol. This is no good. The symbol is a declaration though, so it can't be "used" directly. Instead, we assign it to another global and "use" it. That avoids the optimizations and should be fine. Even if that symbol is a nullptr because it doesn't exist, we are taking the pointer to it, which should be fine for all situations.

Make the little variable LinkOnceODRLinkage so that it doesn't take up as much space.

etcwilde · 2021-11-04T16:03:55Z

@swift-ci please test

aschwaighofer

LGTM

swift-ci · 2021-11-04T16:39:08Z

Build failed
Swift Test Linux Platform
Git Sha - 3e2c5a6

etcwilde · 2021-11-04T16:40:13Z

@swift-ci please clean test Linux

swift-ci · 2021-11-04T18:41:23Z

Build failed
Swift Test Linux Platform
Git Sha - 3e2c5a6

swift-ci · 2021-11-04T19:03:40Z

Build failed
Swift Test OS X Platform
Git Sha - 3e2c5a6

DougGregor · 2021-11-05T04:02:29Z

@swift-ci please test

swift-ci · 2021-11-05T06:37:49Z

Build failed
Swift Test OS X Platform
Git Sha - 3e2c5a6

swift-ci · 2021-11-05T07:24:00Z

Build failed
Swift Test Linux Platform
Git Sha - 3e2c5a6

DougGregor · 2021-11-05T17:48:00Z

@swift-ci please test

etcwilde · 2021-11-05T17:59:55Z

The macOS bot may be signal. I'm trying to reproduce the lldb-api :: lang/swift/playgrounds/TestPlaygrounds.py failure locally because the log doesn't give very much information.

FAIL: LLDB (/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/buildbot_incremental/llvm-macosx-x86_64/bin/clang-x86_64) :: test_cross_module_extension_dsym (TestPlaygrounds.TestSwiftPlaygrounds)
UNSUPPORTED: LLDB (/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/buildbot_incremental/llvm-macosx-x86_64/bin/clang-x86_64) :: test_cross_module_extension_dwarf (TestPlaygrounds.TestSwiftPlaygrounds) (skipping due to the following parameter(s): debug info format) 
UNSUPPORTED: LLDB (/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/buildbot_incremental/llvm-macosx-x86_64/bin/clang-x86_64) :: test_cross_module_extension_dwo (TestPlaygrounds.TestSwiftPlaygrounds) (test case does not fall in any category of interest for this run) 
UNSUPPORTED: LLDB (/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/buildbot_incremental/llvm-macosx-x86_64/bin/clang-x86_64) :: test_cross_module_extension_gmodules (TestPlaygrounds.TestSwiftPlaygrounds) (skipping (gmodules only makes sense for clang tests)) 
======================================================================
FAIL: test_cross_module_extension_dsym (TestPlaygrounds.TestSwiftPlaygrounds)
   Test that playgrounds work
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/llvm-project/lldb/packages/Python/lldbsuite/test/lldbtest.py", line 1846, in test_method
    return attrvalue(self)
  File "/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/llvm-project/lldb/packages/Python/lldbsuite/test/decorators.py", line 150, in wrapper
    return func(*args, **kwargs)
  File "/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/llvm-project/lldb/packages/Python/lldbsuite/test/decorators.py", line 150, in wrapper
    return func(*args, **kwargs)
  File "/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/llvm-project/lldb/packages/Python/lldbsuite/test/decorators.py", line 150, in wrapper
    return func(*args, **kwargs)
  File "/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/llvm-project/lldb/test/API/lang/swift/playgrounds/TestPlaygrounds.py", line 79, in test_cross_module_extension
    self.do_test(True)
  File "/Users/buildnode/jenkins/workspace/swift-PR-macos/branch-main/llvm-project/lldb/test/API/lang/swift/playgrounds/TestPlaygrounds.py", line 150, in do_test
    self.assertTrue("=\\'23\\'" in playground_output)
AssertionError: False is not True

The only way I can see for that particular test to fail is if something fails to import.

swift-ci · 2021-11-05T21:44:34Z

Build failed
Swift Test OS X Platform
Git Sha - 3e2c5a6

etcwilde · 2021-11-09T19:55:56Z

swiftlang/llvm-project#3520

@swift-ci please test macOS

swift-ci · 2021-11-09T22:51:23Z

Build failed
Swift Test OS X Platform
Git Sha - 3e2c5a6

etcwilde · 2021-11-10T06:34:34Z

@swift-ci please test macOS

etcwilde requested review from DougGregor and aschwaighofer November 3, 2021 17:25

etcwilde force-pushed the ewilde/fix-weak-extendedFramePointerFlags branch from 63599b7 to 9258f3b Compare November 3, 2021 17:52

DougGregor reviewed Nov 3, 2021

View reviewed changes

lib/IRGen/IRGenModule.cpp Outdated Show resolved Hide resolved

etcwilde force-pushed the ewilde/fix-weak-extendedFramePointerFlags branch from 9258f3b to 5ffd77e Compare November 3, 2021 18:17

etcwilde mentioned this pull request Nov 3, 2021

5.5 fix weak extended frame pointer flags #40044

Merged

etcwilde added 2 commits November 4, 2021 09:02

Make it LinkOnceODRLinkage

3e2c5a6

Make the little variable LinkOnceODRLinkage so that it doesn't take up as much space.

etcwilde force-pushed the ewilde/fix-weak-extendedFramePointerFlags branch from 5ffd77e to 3e2c5a6 Compare November 4, 2021 16:03

aschwaighofer approved these changes Nov 4, 2021

View reviewed changes

etcwilde merged commit 28c6dd3 into swiftlang:main Nov 10, 2021

etcwilde deleted the ewilde/fix-weak-extendedFramePointerFlags branch November 10, 2021 18:04

Fixing the strong imported async frame pointer flags #40035

Fixing the strong imported async frame pointer flags #40035

Uh oh!

Conversation

etcwilde commented Nov 3, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

etcwilde commented Nov 3, 2021

Uh oh!

swift-ci commented Nov 3, 2021

Uh oh!

etcwilde commented Nov 3, 2021

Uh oh!

Uh oh!

etcwilde commented Nov 3, 2021

Uh oh!

swift-ci commented Nov 3, 2021

Uh oh!

etcwilde commented Nov 3, 2021

Uh oh!

etcwilde commented Nov 3, 2021

Uh oh!

swift-ci commented Nov 3, 2021

Uh oh!

DougGregor commented Nov 4, 2021

Uh oh!

swift-ci commented Nov 4, 2021

Uh oh!

etcwilde commented Nov 4, 2021

Uh oh!

etcwilde commented Nov 4, 2021

Uh oh!

aschwaighofer left a comment

Choose a reason for hiding this comment

Uh oh!

swift-ci commented Nov 4, 2021

Uh oh!

etcwilde commented Nov 4, 2021

Uh oh!

swift-ci commented Nov 4, 2021

Uh oh!

swift-ci commented Nov 4, 2021

Uh oh!

DougGregor commented Nov 5, 2021

Uh oh!

swift-ci commented Nov 5, 2021

Uh oh!

swift-ci commented Nov 5, 2021

Uh oh!

DougGregor commented Nov 5, 2021

Uh oh!

etcwilde commented Nov 5, 2021

Uh oh!

swift-ci commented Nov 5, 2021

Uh oh!

etcwilde commented Nov 9, 2021

Uh oh!

swift-ci commented Nov 9, 2021

Uh oh!

etcwilde commented Nov 10, 2021

Uh oh!

Uh oh!

etcwilde commented Nov 3, 2021 •

edited

Loading