The toolchain was introduced in 710816d3e0
but was not used. Test cases now use fake resource dir to lookup
static-executable-args.lnk file, which is required by the toolchain but
is not present when not building stdlib for WASI.
In LLVM unified builds `%swift_obj_root` points to `<LLVM build dir>/tools/swift`,
and folders like `bin`, `lib` and `share` are not under `swift_obj_root`, which
makes some tests fail.
For the cases in which `%swift_obj_root/lib` was used, replace it by
using `%swift-lib-dir` instead. Replicate `%swift-lib-dir` to create
`%swift-bin-dir` and `%swift-share-dir`, and use those instead of
`%swift_obj_root/bin` and `%swift_obj_root/share`.
This alternates work both in Swift build-script builds and also in LLVM
unified builds.
CMake: fix missing `SWIFT_CONCURRENCY_GLOBAL_EXECUTOR` value
`SWIFT_CONCURRENCY_GLOBAL_EXECUTOR` is defined in `stdlib/cmake/modules/StdlibOptions.cmake`, which is not included during the first pass of evaluation of the root `CMakeLists.txt`. It is available on subsequent evaluations after the value is stored in CMake cache. This led to subtle bugs, where `usr/lib/swift_static/linux/static-stdlib-args.lnk` didn't contain certain flags on clean toolchain builds, but did contain them in incremental builds.
Not having these autolinking flags in toolchain builds leads to errors when statically linking executables on Linux.
Additionally, since are trivial tests previously didn't link Dispatch statically, the didn't expose a bug where `%import-static-libdispatch` substitution had a missing value. To fix that I had to update `lit.cfg` and clean up some of the related path computations to infer a correct substitution value.
Resolves some of the errors reported in https://github.com/apple/swift/issues/65097.
When the current toolchain is not a Xcode toolchain, derive
'-external-plugin-path' poinintng Xcode plugins paths, so we can use
plugins in Xcode.
rdar://108624128
Tool selection is primarily done by checking the executable (= symlink) name.
But sometimes (e.g. if the tool symlink is not there) it's useful to have an option for selecting the tool.
The selection option (e.g. -sil-opt) must be the first argument of swift-frontend.
* Factor out ASTContext plugin loading to newly introduced 'PluginLoader'
* Insert 'DependencyTracker' to 'PluginLoader'
* Add dependencies right before loading the plugins
rdar://104938481
This allows building the Swift standard library for targets which may
not have an actual OS running. This is identified by the OS field in
the target triple being set to `none`.
`%swiftc_driver` is a substitution already, which means
`%swiftc_driver_stdlib_target` was actually being substituted as
`%swiftc_driver` first with `-stdlib-target` tacked on the end.
Resolves rdar://106170241 and rdar://106170343.
Inject the necessary module maps and apinotes via the VFS. This cleans
up the developer build in preparation for a secondary change to remove
this need for deployed scenarios as well. Injecting the content via the
VFS will enable us restore the ability to work with a pristine
installation of Visual Studio, dropping the custom action for the Swift
installer, and open the pathway to per-user installation of Swift.
Thanks to @bnbarham for the help and discussion in resolving the test
issues.
* Remove support for linking arclite
Darwin no longer uses arclite and it's no longer distributed
in the macOS SDKs.
This leaves the options -link-objc-runtime and -no-link-objc-runtime
in place, but strips out all the logic that actually used them.
* Remove a dead function
* Warn if `-link-objc-runtime` is used
* Update tests to not look for arclite library
* Add an explicit test for the deprecation warning
* Move the macOS-only -link-objc-runtime test to a separate test file
Although nonescaping closures are representationally trivial pointers to their
on-stack context, it is useful to model them as borrowing their captures, which
allows for checking correct use of move-only values across the closure, and
lets us model the lifetime dependence between a closure and its captures without
an ad-hoc web of `mark_dependence` instructions.
During ownership elimination, We eliminate copy/destroy_value instructions and
end the partial_apply's lifetime with an explicit dealloc_stack as before,
for compatibility with existing IRGen and non-OSSA aware passes.
bdf60152f0 added some extra padding to
forFilelistCapture to avoid it being exactly 4096 bytes, triggering a
tail bug. Rather than do that, just run the sed's first so that the
input is quite small (definitely under 4096 bytes).
Resolves rdar://105395733.
Ensure that we link `swift_CompilerPluginSupport` into the compiler and
SourceKit, and set the rpath appropriately to find the library in its
installed location.
A number of driver tests copy the driver executable into a temporary
directory to isolate it from the build tree. Also copy the plugin
support library into its appropriate place near the driver executable
to ensure these tests keep working. To help with this, add a
`swift_swift_parser` lit feature, which we can use in tests that
involve the new parser's capabilities.
Various clients still use the old driver to convert a driver arguments to frontend arguments. This has to work when given new driver arguments, ie. not fail with an unknown argument error.
Move all "extra options" from the new driver into `Options.td` and mark them as new driver only.
Fixes#60406
Ran the linux x86_64 compiler validation suite on this pull with `-use-ld=lld -Xlinker --gc-sections`, then without stripping as in the linked issue, and only saw the 20 unrelated failures just like @drodriguez had.
This is important for performance diagnostics: it’s assumed that (non-generic) MemoryLayout constants do not need to create metadata at runtime. At Onone this is only guaranteed if the TargetConstantFolding pass runs.
rdar://94836837
`test/Driver/filelists.swift` writes out the swift driver jobs to a file
called `%t/forFilelistCapture`. We then `tail -2 | head -1` that file to
get the first line, run some seds, and expect that to the output file
list (eg. from the swift-frontend command). Unfortunately if that file
happens to end up being 4096 in length, `tail` skips the first line and
thus the seds run on *second* line. That produces `/usr/bin/ld` instead,
which obviously doesn't contain the output files!
Add a comment to the end of the file so that it's unlikely to be 4096
in length. Technically we could still hit it if we removed a bunch of
arguments, but this is fine until that bug is fixed.
