Merge darwin/libdispatch-1121 to master #435
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
CMake: carry over some nuances from autotools test config Signed-off-by: Daniel A. Steffen <[email protected]>
1. add a CMake module to define additional warning flags to enable when compiling dispatch. 2. enable the additional warning flags by default for CMake 3. match autotools behavior of not including the BSD_OVERLAY when compiling the dispatch_c99 test case. This avoids a warning about __printflike being redefined when compiling this test case with the expanded set of warning flags. Signed-off-by: Daniel A. Steffen <[email protected]>
Enable additional compiler warnings for CMake build Signed-off-by: Daniel A. Steffen <[email protected]>
Signed-off-by: Daniel A. Steffen <[email protected]> Signed-off-by: Daniel A. Steffen <[email protected]>
Signed-off-by: Daniel A. Steffen <[email protected]> Signed-off-by: Daniel A. Steffen <[email protected]>
…ge-master Merge darwin/libdispatch-913.1.4 to master Signed-off-by: Daniel A. Steffen <[email protected]>
Add support to configure the build to use sanitizers to aid in finding issues. The `DISPATCH_USE_SANIZITERS` is a cmake option that defaults to none, but can be one of `Address, Memory, MemoryWithOrigins, Undefined, Leaks, Address;Undefined`. This allows for easily building libdispatch and the tests with the sanitizers. Signed-off-by: Daniel A. Steffen <[email protected]>
build: add support to build with sanitization Signed-off-by: Daniel A. Steffen <[email protected]>
Newer versions of glibc indicate that they intend to move the major macro from sys/types.h to sys/sysmacros.h. Add a check for the header and include that earlier to ensure that the macro is provided by the newer header when available/possible. This avoids an unnecessary warning from the system headers. Because `config_ac.h` is not available at the right location, we cannot include the header to check whether the header is available. Rely on the compiler provided `__has_include` instead of the configure check. Adjust the inclusion of `sys/cdefs.h` accordingly. Signed-off-by: Daniel A. Steffen <[email protected]>
linux: update header used for `major` macro Signed-off-by: Daniel A. Steffen <[email protected]>
Signed-off-by: Daniel A. Steffen <[email protected]>
Signed-off-by: Daniel A. Steffen <[email protected]>
Fix building on armv6 Signed-off-by: Daniel A. Steffen <[email protected]>
This fixes issues around resuming or re-registering file-descriptors with a previously triggered EPOLLONESHOT event. Signed-off-by: Pierre Habouzit <[email protected]> Fixes: SR-5759 Signed-off-by: Daniel A. Steffen <[email protected]>
Track armed events in muxnotes closely Signed-off-by: Daniel A. Steffen <[email protected]>
It is possible for _dispatch_muxnote_create to return NULL, this error condition was not being handled in dispatch_unote_register leading to an immediate SIGSEGV when it happened. Signed-off-by: Daniel A. Steffen <[email protected]>
SR-4677: avoid SIGSEGV if _dispatch_muxnote_create returns NULL Signed-off-by: Daniel A. Steffen <[email protected]>
This adjusts the symlink creation so that it always occurs. This also allows us to ensure that we wire up the dependency for the swift module to the symlink creation. The CMake based build would fail without this when trying to build dispatch for SourceKit on Linux. Signed-off-by: Daniel A. Steffen <[email protected]>
build: fix symlink creation Signed-off-by: Daniel A. Steffen <[email protected]>
rdar://problem/34132958 Signed-off-by: Daniel A. Steffen <[email protected]>
Synchronize the Linux and Darwin overlay versions of Time.swift Signed-off-by: Daniel A. Steffen <[email protected]>
…val. rdar://problem/34440944 Signed-off-by: Daniel A. Steffen <[email protected]>
Fix overflow traps in DispatchTime/DispatchWallTime/DispatchTimeInterval Signed-off-by: Daniel A. Steffen <[email protected]>
Changes for SR-5759 introduced a bug where disarmed events got dropped when a muxnote is reused, due to erroneously setting dmn->dmn_events to the combination of new events and currently armed events. Combining existing events with the combination of new events and armed events implements the intended behavior. Signed-off-by: Daniel A. Steffen <[email protected]>
Properly pick up new events when reusing a muxnote Signed-off-by: Daniel A. Steffen <[email protected]>
Update the documentation to reference the CMake build infrastructure rather than autotools for the non-Darwin targets. Once the Darwin build is switched over, the entire build process can be described in a single block. Signed-off-by: Daniel A. Steffen <[email protected]>
docs: update documentation to show how to use CMake Signed-off-by: Daniel A. Steffen <[email protected]>
Most targets use unsigned chars. Explicitly change the static tables to `signed char`. An alternative would be to force signed chars during the compilation of libdispatch. When building for Linux AArch64, the use of `-1` causes signed conversion warnings when initializing the table. Signed-off-by: Daniel A. Steffen <[email protected]>
Attempt to be better with signed conversions Signed-off-by: Daniel A. Steffen <[email protected]>
Signed-off-by: Daniel A. Steffen <[email protected]>
There's a quirk on Windows where exiting the process will still run fiber-local storage destructors for the thread that called ExitProcess() (but only for that thread). pthreads-based platforms don't call any key destructors on exit, and _libdispatch_tsd_cleanup() assumes this. So, on Windows, calling exit() from a queue thread will trigger the fatal error in _dispatch_frame_cleanup(). Implement a reliable method of detecting whether the process is exiting, and if it is, ignore calls to _libdispatch_tsd_cleanup(). Signed-off-by: Kim Topley <[email protected]>
Ignore Windows FLS destructors during process exit Signed-off-by: Kim Topley <[email protected]>
…of Data.enumerateBytes() for Swift 4.2 and obsolete the old version. rdar://problem/40775762 Signed-off-by: Kim Topley <[email protected]>
Change the signature for DispatchData.enumerateBytes() to match that … Signed-off-by: Kim Topley <[email protected]>
rdar://problem/40951342 Signed-off-by: Kim Topley <[email protected]>
Add documentation comments. Signed-off-by: Kim Topley <[email protected]>
CMakeLists.txt checks for librt's presence but the result is never used. This causes a link failure on systems where clock_gettime() is provided by librt. Signed-off-by: Kim Topley <[email protected]>
build: link against librt if it's available Signed-off-by: Kim Topley <[email protected]>
generic_unix_stubs.c seems to no longer be necessary. generic_unix_stubs.h still had some required definitions in it, so move those into shims.h and a new shims/mach.h. This also lets us remove some duplicated definitions in the Windows shims. Signed-off-by: Kim Topley <[email protected]>
Remove generic_unix_stubs Signed-off-by: Kim Topley <[email protected]>
When building debug, swift requires a supplementary library. The swift driver is aware of this and will make the appropriate arrangements. However, since the swift component is compacted into the C library, the linker driver being used does not understand this requirement. Explicitly link against the support library. This allows building the debug configuration of libdispatch with a release configuration of the swift standard library. Signed-off-by: Kim Topley <[email protected]>
build: link against swiftOnoneSupport in debug configuration Signed-off-by: Kim Topley <[email protected]>
Allow libdispatch to be embedded in other CMake projects using add_subdirectory(). This essentially means that `CMAKE_SOURCE_DIR` and `CMAKE_BINARY_DIR` cannot be used directly. Use the versions of these variables which are scoped to the current project or directory. Signed-off-by: Kim Topley <[email protected]>
build: support embedding in other CMake projects Signed-off-by: Kim Topley <[email protected]>
This hack was only needed for Swift 3 mode in a narrow case. Now that we're dropping Swift 3 from master, the source break this was working around is no longer relevant. Signed-off-by: Kim Topley <[email protected]>
Drop @_downgrade_exhaustivity_check attribute Signed-off-by: Kim Topley <[email protected]>
swiftlang/swift#18895. Signed-off-by: Kim Topley <[email protected]>
Signed-off-by: Kim Topley <[email protected]>
Remove version number used in 'available' attribute for non-specific platform '*'. Signed-off-by: Kim Topley <[email protected]>
…d to swiftlang/swift#18623. Signed-off-by: Kim Topley <[email protected]>
Fix redundant access-level modifiers. Signed-off-by: Kim Topley <[email protected]>
Signed-off-by: Kim Topley <[email protected]>
Signed-off-by: Kim Topley <[email protected]>
|
@swift-ci please test |
compnerd
requested changes
Dec 7, 2018
| BlocksRuntime_STATIC) | ||
| endif() | ||
| endif() | ||
| STATIC |
There was a problem hiding this comment.
This is bad. I think that we are missing a merge from master. This will remove the shared BlocksRuntime build. We currently are building block static and shared.
There was a problem hiding this comment.
Thanks for looking. I will be working on completing this source drop today.
compnerd
reviewed
Dec 7, 2018
| SHARED IMPORTED GLOBAL) | ||
| set_target_properties(swiftCore | ||
| PROPERTIES | ||
| IMPORTED_LOCATION |
There was a problem hiding this comment.
This was moved to using the exported targets from Swift to avoid computing the path. This is much more fragile and I believe may break the build currently.
compnerd
reviewed
Dec 7, 2018
| @@ -112,8 +152,7 @@ if(ENABLE_INTERNAL_PTHREAD_WORKQUEUES) | |||
| set(HAVE_PTHREAD_WORKQUEUES 0) | |||
| else() | |||
| check_include_files(pthread/workqueue_private.h HAVE_PTHREAD_WORKQUEUE_PRIVATE_H) | |||
| check_include_files(pthread_workqueue.h HAVE_PTHREAD_WORKQUEUE_H) | |||
| if(HAVE_PTHREAD_WORKQUEUE_PRIVATE_H AND HAVE_PTHREAD_WORKQUEUE_H) | |||
| if(HAVE_PTHREAD_WORKQUEUE_PRIVATE_H) | |||
| COMMAND | ||
| ${CMAKE_COMMAND} -E copy_if_different "${PROJECT_SOURCE_DIR}/private/generic/module.modulemap" "${PROJECT_SOURCE_DIR}/private/module.modulemap") | ||
| ${CMAKE_COMMAND} -E create_symlink "${PROJECT_SOURCE_DIR}/private/generic/module.modulemap" "${PROJECT_SOURCE_DIR}/private/module.modulemap") |
There was a problem hiding this comment.
I think that we should keep the new behaviour for the module map handling as it exposed some issues with the different ports.
| endif() | ||
| add_custom_target(module-maps ALL | ||
| add_custom_target(module-map-symlinks |
There was a problem hiding this comment.
This is going to break dependencies as the target name changed.
| list(APPEND compile_flags -target;${AST_TARGET}) | ||
| list(APPEND link_flags -target;${AST_TARGET}) | ||
| if(ASL_TARGET) | ||
| list(APPEND FLAGS -target;${ASL_TARGET}) |
There was a problem hiding this comment.
This change breaks the compilation/cross-linking.
| COMMAND | ||
| ${CMAKE_COMMAND} -E make_directory ${module_directory} | ||
| COMMAND | ||
| ${CMAKE_SWIFT_COMPILER} ${flags} -c ${sources} -o ${ASL_OUTPUT}) |
There was a problem hiding this comment.
This increases the compile time significantly and forces a full recompile on every change.
| PRIVATE | ||
| swift/DispatchStubs.cc | ||
| ${CMAKE_CURRENT_BINARY_DIR}/swiftDispatch.o) | ||
| if(CMAKE_BUILD_TYPE MATCHES Debug) |
There was a problem hiding this comment.
The change to use the swift driver for linking obviated this
| if("${CMAKE_C_SIMULATE_ID}" STREQUAL "MSVC") | ||
| target_compile_options(dispatch PRIVATE /EHs-c-) | ||
| else() | ||
| if(NOT "${CMAKE_C_SIMULATE_ID}" STREQUAL "MSVC") |
There was a problem hiding this comment.
This breaks builds with clang-cl.
| add_custom_command(TARGET dispatch POST_BUILD | ||
| COMMAND ${CMAKE_COMMAND} -E make_directory .libs | ||
| COMMAND ${CMAKE_COMMAND} -E copy $<TARGET_FILE:dispatch> .libs | ||
| COMMENT "Copying libdispatch to .libs") |
There was a problem hiding this comment.
Everything has been migrated, this is no longer needed
| RUNTIME DESTINATION bin) | ||
|
|
||
| DESTINATION | ||
| "${INSTALL_TARGET_DIR}") |
| @@ -205,14 +199,16 @@ enum { | |||
| DISPATCH_IOCNTL_MAX_PENDING_IO_REQS, | |||
| }; | |||
|
|
|||
| static struct dispatch_io_defaults_s { | |||
| extern struct dispatch_io_defaults_s { | |||
There was a problem hiding this comment.
Hmm, don't think that the extern is needed here as the statement is a definition.
|
This PR is not ready. Closing. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Merge tag darwin/libdispatch-1121 on darwin/trunk. This includes the latest Apple source drop libdispatch-1121 (up from 913.1.4), on top of tag darwin/darwin-012 which is the result of pulling into the Apple repository the PRs since the merge of darwin/libdispatch-913.1.4 (PR #289).