Age | Commit message (Collapse) | Author |
|
Test: used clang from the app bundle on macOS Catalina
Bug: https://github.com/android/ndk/issues/1060
Change-Id: I995e69cb0a5825f1d79a39baf19bed269749be08
|
|
|
|
* changes:
Note the libc++ update in the changelog.
Mark abs.fail.cpp broken for pre-API 19.
Get libc++ from the new location.
|
|
Test: built and inspected on Linux
Bug: http://b/143105198
Change-Id: I624344a9de93f3a5e11a77eb3b51d7b4a2d13aa6
|
|
Test: None
Bug: None
Change-Id: Ib23dee97aaf9cfce7b6b7d05f73ff5fc62ac19ea
|
|
Switching to using Clang's `-verify` feature for testing the compile
failure tests means that we're now checking that all the *correct*
diagnostics are firing, rather than just compilation failing. The
-Wabsolute-value warning expected here doesn't fire for pre-API 19
(see the bug for more details).
Test: ./run_tests.py --rebuild --filter libc++
Bug: https://github.com/android/ndk/issues/1237
Change-Id: I4482823d28a7c38d3e29c040450b7741f697aa60
|
|
Test: treehugger
Bug: None
Change-Id: I7326949928a3a4ccc048862c55b8389446177af2
|
|
|
|
I'd originally built `_make_zip_package` to make it impossible to
avoid a single parent directory, but now we need multiple files in the
root of the zip because that's what the signer expects.
Test: unzip -l of the artifact, saw _codesign/volumename
Bug: http://b/143105198
Change-Id: Ib3834ed95154749136d8d6a36abfd47bb9a8f252
|
|
|
|
|
|
It seems codesign will not allow this file:
https://stackoverflow.com/a/25950545/632035
Test: Built framework bundle and inspected
Bug: http://b/143105198
Change-Id: I19ab69e2bfd22aeca818e58ae924861bd52a25ca
|
|
Test: removed darwin-only check and built framework bundle
Bug: http://b/143105198
Change-Id: I7c87a0e9e230105bc5193eaacdd5876afdf0afba
|
|
|
|
Apparently this step hasn't been needed for a while.
Test: treehugger
Bug: None
Change-Id: Ide70514ba135c24e60ccd52b60f4253bd3c20a6b
|
|
|
|
Test: saw correct error with broken LIT
Bug: None
Change-Id: I84e3252bf9e9091814c11f3dd4022a8b71dba684
|
|
|
|
Longer term we can remove the need for all of these.
1 can be removed today with a little work to migrate any modules still
using it.
2 can be removed once all our builds are off those ancient shell
scripts.
3 can be removed when we get properly integrated with CMake
(https://github.com/android/ndk/issues/463).
But for now documenting them means I'll stop forgetting step 3.
Test: None
Bug: None
Change-Id: I18625db76891ed9810bec29b92a61258063008d0
|
|
|
|
|
|
Test: ./checkbuild.py --package, without the darwin specific check
Bug: http://b/143105198
Bug: https://github.com/android/ndk/issues/1060
Change-Id: I070db662631886bcc449849a5b4582a0b4f9a8a8
|
|
The fix itself is in libc++abi.
Test: This is the test
Bug: https://github.com/android/ndk/issues/1200
Change-Id: If1c75e80b051e2a2e9316745e671a4e81ada6fbf
|
|
|
|
|
|
Bug: http://b/153310680
Test: suck it and see
Change-Id: Ie75aeffa5b9fabe5bfade4192a1eaf992fcd039e
|
|
Test: enh is going to use it
Bug: None
Change-Id: I55c90580606114914188e5cddb309732551e18a7
|
|
|
|
The meaning of `path` was accidentally changed from a file path to a
parent directory path.
Test: ls
Bug: https://github.com/android/ndk/issues/1229
Change-Id: Ie59051d8f20212f202e268bc727feef3aba26ad6
|
|
|
|
Test: N/A
Change-Id: Ic16114a9181340f33d8ac3b65c2d34a8c5313ed8
|
|
|
|
We never gained the ability to distribute the tarballs like we wanted
to do for Linux/Darwin, and the SDK requires zips.
I'm changing the default behavior here as well. It takes almost 10
minutes to zip -9 the NDK even with these beefy machines since zip
isn't parallel. I don't want that to be part of our regular workflow,
so `--package` must now be explicitly given to cause packaging to
happen. The build servers already pass this argument (long ago we
didn't package by default, and I never got around to cleaning up the
build configs when I flipped the default the last time).
Test: ./checkbuild.py
Test: ./checkbuild.py --package
Bug: None
Change-Id: Iad8e6579f54135881d87fc0473ea746262da648e
|
|
|
|
|
|
Removing *all* of these hinges on us being able to remove binutils
because a final part of the cleanup will be removing the
target-specific bin directories within the LLVM toolchain, which can
affect driver behavior. Since we don't know if we'll be able to get
that done for next year's LTS, don't be so explicit about the
timeline. We can be more specific when we know more.
Test: None, doc only
Bug: None
Change-Id: I0ddbd948c1c45b894d417b0ef3d1bfded0eccc7a
|
|
Test: None, doc only
Bug: https://github.com/android/ndk/issues/1176
Change-Id: I13a77f24e2c755126595d5d4da15d7e2211e12e7
|
|
|
|
This is the r21 toolchain with the cherry-picks for r21b. The
following Gerrit query shows the patches it contains:
https://android-review.googlesource.com/q/branch:llvm-r365631+status:merged+before:2020-03-30
Test: treehugger
Bug: https://github.com/android/ndk/issues/1178
Bug: https://github.com/android/ndk/issues/1184
Bug: https://github.com/android/ndk/issues/1198
Bug: https://github.com/android/ndk/issues/1205
Bug: https://github.com/android/ndk/issues/1206
Change-Id: I223b98e96c32e0c754d9c27486d4f0ccfad32851
|
|
|
|
This reverts commit fed0a4058d786dcc73ee4d625388f9f416b719cc.
Reason for revert: Broke Darwin again
Change-Id: I0e432102934a3fb2ab659924b461a23a0eb64c44
|
|
|
|
Rather than trying to replace `ld` in each location, this instead
relies on Clang defaulting to `ld.lld`, which it invokes from the
Clang bin directory.
Test: treehugger
Bug: None
Change-Id: I32b0655c0d89522848f044f1067b53f9214c5444
|
|
|
|
Because we look silly when our demo of IntelliSense in Visual Studio is
only offering completions that are header file include guard spam.
Bug: https://youtu.be/JXQBdMNDL7k?t=215
Test: treehugger
Change-Id: I608e1e000dec3f4b4919263dfb5688f42c02b1b1
|
|
|
|
This reverts commit 393761d2e07461366ea456cea7f67fade95e3b4b.
This broke Darwin. It seems that LLD assumes the target based on the
name of the executable, so it can't be reliably used from the
target-specific directory as I tried to do here.
Trying to use this with Darwin results in the following:
```
ld: warning: ignoring unknown argument: -z
ld: warning: ignoring unknown argument: --warn-shared-textrel
ld: warning: ignoring unknown argument: -z
ld: warning: ignoring unknown argument: -z
ld: warning: ignoring unknown argument: --hash-style=both
ld: warning: ignoring unknown argument: --enable-new-dtags
ld: warning: ignoring unknown argument: --eh-frame-hdr
ld: warning: ignoring unknown argument: -m
ld: warning: ignoring unknown argument: -dynamic-linker
ld: warning: ignoring unknown argument: -plugin
ld: warning: ignoring unknown argument: -plugin-opt=mcpu=x86-64
ld: warning: ignoring unknown argument: --gc-sections
ld: warning: ignoring unknown argument: -rpath-link=/Volumes/Android/danalbert/ndk/out/darwin/android-ndk-r22-canary/toolchains/llvm/prebuilt/darwin-x86_64/sysroot/usr/lib/x86_64-linux-android/21
ld: warning: ignoring unknown argument: -rpath-link=./obj/local/x86_64
ld: warning: ignoring unknown argument: --exclude-libs
ld: warning: ignoring unknown argument: --exclude-libs
ld: warning: ignoring unknown argument: --exclude-libs
ld: warning: ignoring unknown argument: --build-id
ld: warning: ignoring unknown argument: --no-undefined
ld: warning: ignoring unknown argument: --fatal-warnings
ld: error: -arch not specified and could not be inferred
clang++: error: linker command failed with exit code 1 (use -v to see invocation)
```
It's trivially fixed with `-fuse-ld=lld`, which causes Clang to pick
LLD from a different directory and passes the target explicitly. We
could do that in ndk-build and in the CMake toolchain to change the
default for those systems, but we really need to upstream a patch to
change the default in the Clang driver to affect other build systems.
Bug: None
Test: treehugger
|
|
|
|
Test: treehugger, some manual inspection of linker flags for sanity
Bug: None
Change-Id: Ib8b371cc932b46fddb623165c17d77757be9de2d
|
|
|