ci: add support for macOS to GitLab CI

Hi,

this patch series extends GitLab CI to also support macOS. Besides extending test coverage for GitLab users, this change also has the added benefit that the macOS runners at GitLab are based on Apple silicon, which to the best of my knowledge is not something we're currently testing on.

This patch series builds on top of ps/gitlab-ci-static-analysis (currently at cd69c635a1 (ci: add job performing static analysis on GitLab CI, 2023-12-28)) to avoid a conflict.

Patrick

Patrick Steinhardt (3):
ci: make p4 setup on macOS more robust
Makefile: detect new Homebrew location for ARM-based Macs
ci: add macOS jobs to GitLab CI

 .gitlab-ci.yml             | 26 +++++++++++++++++++++++++-
 ci/install-dependencies.sh | 10 ++++------
 ci/lib.sh                  | 12 +++++++++++-
 config.mak.uname           | 13 +++++++++++++
 4 files changed, 53 insertions(+), 8 deletions(-)

base-commit: cd69c635a1a62b0c8bfdbf221778be8a512ad048

2.43.GIT

When setting up Perforce on macOS we put both p4 and p4d into "$HOME/bin". On GitHub CI this directory is indeed contained in the PATH environment variable and thus there is no need for additional setup than to put the binaries there. But GitLab CI does not do this, and thus our Perforce-based tests would be skipped there even though we download the binaries.

Refactor the setup code to become more robust by downloading binaries into a separate directory which we then manually append to our PATH. This matches what we do on Linux-based jobs.

---
 ci/install-dependencies.sh | 10 ++++------
 ci/lib.sh                  |  3 +++
 2 files changed, 7 insertions(+), 6 deletions(-)

index 4f407530d3..b4e22de3cb 100755
@@ -37,15 +37,13 @@ macos-*)
 	test -z "$BREW_INSTALL_PACKAGES" ||
 	brew install $BREW_INSTALL_PACKAGES
 	brew link --force gettext
-	mkdir -p $HOME/bin
-	(
-		cd $HOME/bin
+
+	mkdir -p "$P4_PATH"
+	pushd "$P4_PATH"
 		wget -q "$P4WHENCE/bin.macosx1015x86_64/helix-core-server.tgz" &&
 		tar -xf helix-core-server.tgz &&
 		sudo xattr -d com.apple.quarantine p4 p4d 2>/dev/null || true
-	)
-	PATH="$PATH:${HOME}/bin"
-	export PATH
+	popd
 
 	if test -n "$CC_PACKAGE"
 	then

index c749b21366..f631206a44 100755
@@ -344,6 +344,9 @@ macos-*)
 	then
 		MAKEFLAGS="$MAKEFLAGS APPLE_COMMON_CRYPTO_SHA1=Yes"
 	fi
+
+	P4_PATH="$HOME/custom/p4"
+	export PATH="$P4_PATH:$PATH"
 	;;
 esac
 

2.43.GIT

With the introduction of the ARM-based Macs the default location for Homebrew has changed from "/usr/local" to "/opt/homebrew". We only handle the former location though, which means that unless the user has manually configured required search paths we won't be able to locate it.

Improve upon this by adding relevant paths to our CFLAGS and LDFLAGS as well as detecting the location of msgfmt(1).

---
 config.mak.uname | 13 +++++++++++++
 1 file changed, 13 insertions(+)

index 3bb03f423a..dacc95172d 100644
@@ -158,6 +158,19 @@ ifeq ($(uname_S),Darwin)
 		ifeq ($(shell test -x /usr/local/opt/gettext/bin/msgfmt && echo y),y)
 			MSGFMT = /usr/local/opt/gettext/bin/msgfmt
 		endif
+	# On newer ARM-based machines the default installation path has changed to
+	# /opt/homebrew. Include it in our search paths so that the user does not
+	# have to configure this manually.
+	#
+	# Note that we do not employ the same workaround as above where we manually
+	# add gettext. The issue was fixed more than three years ago by now, and at
+	# that point there haven't been any ARM-based Macs yet.
+	else ifeq ($(shell test -d /opt/homebrew/ && echo y),y)
+		BASIC_CFLAGS += -I/opt/homebrew/include
+		BASIC_LDFLAGS += -L/opt/homebrew/lib
+		ifeq ($(shell test -x /opt/homebrew/bin/msgfmt && echo y),y)
+			MSGFMT = /opt/homebrew/bin/msgfmt
+		endif
 	endif
 
 	# The builtin FSMonitor on MacOS builds upon Simple-IPC.  Both require

2.43.GIT

Add two macOS-based jobs to GitLab CI, one for Clang and one for GCC. This matches equivalent jobs we have for GitHub Workflows, except that we use macOS 14 instead of macOS 13.

Note that one test marked as test_must_fail is surprisingly passing:

t7815-grep-binary.sh (Wstat: 0 Tests: 22 Failed: 0)
TODO passed: 12

This seems to boil down to an unexpected difference in how regcomp(1) works when matching NUL bytes. Cross-checking with the respective GitHub job shows though that this is not an issue unique to the GitLab CI job as it passes in the same way there.

Further note that we do not include the equivalent for the "osx-gcc" job that we use with GitHub Workflows. This is because the runner for macOS on GitLab is running on Apple M1 machines and thus uses the "arm64" architecture. GCC does not support this platform yet.

---
 .gitlab-ci.yml | 26 +++++++++++++++++++++++++-
 ci/lib.sh      |  9 ++++++++-
 2 files changed, 33 insertions(+), 2 deletions(-)

index 793243421c..9748970798 100644
@@ -7,7 +7,7 @@ workflow:
     - if: $CI_COMMIT_TAG
     - if: $CI_COMMIT_REF_PROTECTED == "true"
 
-test:
+test:linux:
   image: $image
   before_script:
     - ./ci/install-docker-dependencies.sh

@@ -52,6 +52,30 @@ test:
       - t/failed-test-artifacts
     when: on_failure
 
+test:osx:
+  image: $image
+  tags:
+    - saas-macos-medium-m1
+  before_script:
+    - ./ci/install-dependencies.sh
+  script:
+    - ./ci/run-build-and-tests.sh
+  after_script:
+    - |
+      if test "$CI_JOB_STATUS" != 'success'
+      then
+        ./ci/print-test-failures.sh
+      fi
+  parallel:
+    matrix:
+      - jobname: osx-clang
+        image: macos-13-xcode-14
+        CC: clang
+  artifacts:
+    paths:
+      - t/failed-test-artifacts
+    when: on_failure
+
 static-analysis:
   image: ubuntu:22.04
   variables:

index f631206a44..d5dd2f2697 100755
@@ -252,7 +252,14 @@ then
 	CI_COMMIT="$CI_COMMIT_SHA"
 	case "$CI_JOB_IMAGE" in
 	macos-*)
-		CI_OS_NAME=osx;;
+		# GitLab CI has Python installed via multiple package managers,
+		# most notably via asdf and Homebrew. Ensure that our builds
+		# pick up the Homebrew one by prepending it to our PATH as the
+		# asdf one breaks tests.
+		export PATH="$(brew --prefix)/bin:$PATH"
+
+		CI_OS_NAME=osx
+		;;
 	alpine:*|fedora:*|ubuntu:*)
 		CI_OS_NAME=linux;;
 	*)

2.43.GIT

Hi Patrick

This doesn't match whats in the rest of the commit message where you explain why there is no gcc job. The patch itself looks good to me and it is nice that we'll now be testing on arm64 with the GitLab runners.

nit: regcomp(3)?

Best Wishes

Phillip

Oops. In my first iterations I still had an osx-gcc job, but I could not get it passing due to the architectural difficulties explained in the message, so I ended up removing it. This sentence is thus a leftover from previous iterations.

Indeed, thanks!

Patrick

Yes and no. While these lines put it in PATH, this only works inside of "ci/install-dependencies.sh". When we call "ci/run-build-and-test.sh" we do not source this script though, which means that "$HOME/bin" will not be part of PATH during the actual test run unless it was already added by the CI.

I'll update the commit message to explain this better.

Patrick

Hi,

this is the second version of my patch series that adds a macOS job to GitLab CI. Changes compared to v1:

- Added a fix for a flaky test in t7527 that caused the pipeline to fail in ~50% of all runs.

- Improved some commit messages.

- Tests now write test data into a RAMDisk. This speeds up tests and fixes some hung pipelines I was seeing.

Thanks for your reviews so far!

Patrick

Patrick Steinhardt (5):
t7527: decrease likelihood of racing with fsmonitor daemon
Makefile: detect new Homebrew location for ARM-based Macs
ci: handle TEST_OUTPUT_DIRECTORY when printing test failures
ci: make p4 setup on macOS more robust
ci: add macOS jobs to GitLab CI

 .gitlab-ci.yml               | 34 +++++++++++++++++++++++++++++++++-
 ci/install-dependencies.sh   | 10 ++++------
 ci/lib.sh                    | 12 +++++++++++-
 ci/print-test-failures.sh    |  2 +-
 config.mak.uname             | 13 +++++++++++++
 t/t7527-builtin-fsmonitor.sh |  2 +-
 6 files changed, 63 insertions(+), 10 deletions(-)

Range-diff against v1:

-:  ---------- > 1:  554b1c8546 t7527: decrease likelihood of racing with fsmonitor daemon
2:  3adb0b7ae8 = 2:  32d8bd1d78 Makefile: detect new Homebrew location for ARM-based Macs
-:  ---------- > 3:  d55da77747 ci: handle TEST_OUTPUT_DIRECTORY when printing test failures
1:  a5d725bea7 ! 4:  1ed6e68650 ci: make p4 setup on macOS more robust

@@ Commit message
into a separate directory which we then manually append to our PATH. This matches what we do on Linux-based jobs.

+ Note that it may seem like we already did append "$HOME/bin" to PATH
+ because we're actually removing the lines that adapt PATH. But we only
+ ever adapted the PATH variable in "ci/install-dependencies.sh", and
+ didn't adapt it when running "ci/run-build-and-test.sh". Consequently,
+ the required binaries wouldn't be found during the test run unless the
+ CI platform already had the "$HOME/bin" in PATH right from the start.
+

## ci/install-dependencies.sh ##
3: d196cfd9d0 ! 5: c5ed38f0a6 ci: add macOS jobs to GitLab CI
@@ Metadata
## Commit message ##
ci: add macOS jobs to GitLab CI

- Add two macOS-based jobs to GitLab CI, one for Clang and one for GCC.
- This matches equivalent jobs we have for GitHub Workflows, except that
- we use macOS 14 instead of macOS 13.
+ Add a job to GitLab CI which runs tests on macOS, which matches the
+ equivalent "osx-clang" job that we have for GitHub Workflows. One
+ significant difference though is that this new job runs on Apple M1
+ machines and thus uses the "arm64" architecture. As GCC does not yet
+ support this comparatively new architecture we cannot easily include an
+ equivalent for the "osx-gcc" job that exists in GitHub Workflows.

Note that one test marked as test_must_fail is surprisingly passing:

t7815-grep-binary.sh (Wstat: 0 Tests: 22 Failed: 0)
TODO passed: 12

- This seems to boil down to an unexpected difference in how regcomp(1)
+ This seems to boil down to an unexpected difference in how regcomp(3P)
works when matching NUL bytes. Cross-checking with the respective GitHub
- job shows though that this is not an issue unique to the GitLab CI job
- as it passes in the same way there.
-
- Further note that we do not include the equivalent for the "osx-gcc" job
- that we use with GitHub Workflows. This is because the runner for macOS
- on GitLab is running on Apple M1 machines and thus uses the "arm64"
- architecture. GCC does not support this platform yet.
+ job shows that this is not an issue unique to the GitLab CI job as it
+ passes in the same way there.

@@ .gitlab-ci.yml: test:
+ image: $image
+ tags:
+ - saas-macos-medium-m1
++ variables:
++ TEST_OUTPUT_DIRECTORY: "/Volumes/RAMDisk"
+ before_script:
++ # Create a 4GB RAM disk that we use to store test output on. This small hack
++ # significantly speeds up tests by more than a factor of 2 because the
++ # macOS runners use network-attached storage as disks, which is _really_
++ # slow with the many small writes that our tests do.
++ - sudo diskutil apfs create $(hdiutil attach -nomount ram://8192000) RAMDisk
+ - ./ci/install-dependencies.sh
+ script:
+ - ./ci/run-build-and-tests.sh
@@ .gitlab-ci.yml: test:
+ if test "$CI_JOB_STATUS" != 'success'
+ then
+ ./ci/print-test-failures.sh
++ mv "$TEST_OUTPUT_DIRECTORY"/failed-test-artifacts t/
+ fi
+ parallel:
+ matrix:

base-commit: cd69c635a1a62b0c8bfdbf221778be8a512ad048

2.43.GIT

In t7527, we test that the builtin fsmonitor daemon works well in various edge cases. One of these tests is frequently failing because events reported by the fsmonitor--daemon are missing an expected event. This failure is essentially a race condition: we do not wait for the daemon to flush out all events before we ask it to quit. Consequently, it can happen that we miss some expected events.

In other testcases we counteract this race by sending a simple query to the daemon. Quoting a comment:

We run a simple query after modifying the filesystem just to introduce a bit of a delay so that the trace logging from the daemon has time to get flushed to disk.

Now this workaround is not a "proper" fix as we do not wait for all events to have been synchronized in a deterministic way. But this fix seems to be sufficient for all the other tests to pass, so it must not be all that bad.

Convert the failing test to do the same. While the test was previously failing in about 50% of the test runs, I couldn't reproduce the failure after the change anymore.

---
 t/t7527-builtin-fsmonitor.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index 78503158fd..363f9dc0e4 100755
@@ -978,7 +978,7 @@ test_expect_success !UNICODE_COMPOSITION_SENSITIVE 'Unicode nfc/nfd' '
 	mkdir test_unicode/nfd &&
 	mkdir test_unicode/nfd/d_${utf8_nfd} &&
 
-	git -C test_unicode fsmonitor--daemon stop &&
+	test-tool -C test_unicode fsmonitor-client query --token 0 &&
 
 	if test_have_prereq UNICODE_NFC_PRESERVED
 	then

2.43.GIT

With the introduction of the ARM-based Macs the default location for Homebrew has changed from "/usr/local" to "/opt/homebrew". We only handle the former location though, which means that unless the user has manually configured required search paths we won't be able to locate it.

Improve upon this by adding relevant paths to our CFLAGS and LDFLAGS as well as detecting the location of msgfmt(1).

---
 config.mak.uname | 13 +++++++++++++
 1 file changed, 13 insertions(+)

index 3bb03f423a..dacc95172d 100644
@@ -158,6 +158,19 @@ ifeq ($(uname_S),Darwin)
 		ifeq ($(shell test -x /usr/local/opt/gettext/bin/msgfmt && echo y),y)
 			MSGFMT = /usr/local/opt/gettext/bin/msgfmt
 		endif
+	# On newer ARM-based machines the default installation path has changed to
+	# /opt/homebrew. Include it in our search paths so that the user does not
+	# have to configure this manually.
+	#
+	# Note that we do not employ the same workaround as above where we manually
+	# add gettext. The issue was fixed more than three years ago by now, and at
+	# that point there haven't been any ARM-based Macs yet.
+	else ifeq ($(shell test -d /opt/homebrew/ && echo y),y)
+		BASIC_CFLAGS += -I/opt/homebrew/include
+		BASIC_LDFLAGS += -L/opt/homebrew/lib
+		ifeq ($(shell test -x /opt/homebrew/bin/msgfmt && echo y),y)
+			MSGFMT = /opt/homebrew/bin/msgfmt
+		endif
 	endif
 
 	# The builtin FSMonitor on MacOS builds upon Simple-IPC.  Both require

2.43.GIT

The TEST_OUTPUT_DIRECTORY environment variable can be used to instruct the test suite to write test data and test results into a different location than into "t/". The "ci/print-test-failures.sh" script does not know to handle this environment variable though, which means that it will search for test results in the wrong location if it was set.

Update the script to handle TEST_OUTPUT_DIRECTORY so that we can start to set it in our CI.

---
 ci/print-test-failures.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index c33ad4e3a2..b1f80aeac3 100755
@@ -8,7 +8,7 @@
 # Tracing executed commands would produce too much noise in the loop below.
 set +x
 
-cd t/
+cd "${TEST_OUTPUT_DIRECTORY:-t/}"
 
 if ! ls test-results/*.exit >/dev/null 2>/dev/null
 then

2.43.GIT

When setting up Perforce on macOS we put both p4 and p4d into "$HOME/bin". On GitHub CI this directory is indeed contained in the PATH environment variable and thus there is no need for additional setup than to put the binaries there. But GitLab CI does not do this, and thus our Perforce-based tests would be skipped there even though we download the binaries.

Refactor the setup code to become more robust by downloading binaries into a separate directory which we then manually append to our PATH. This matches what we do on Linux-based jobs.

Note that it may seem like we already did append "$HOME/bin" to PATH because we're actually removing the lines that adapt PATH. But we only ever adapted the PATH variable in "ci/install-dependencies.sh", and didn't adapt it when running "ci/run-build-and-test.sh". Consequently, the required binaries wouldn't be found during the test run unless the CI platform already had the "$HOME/bin" in PATH right from the start.

---
 ci/install-dependencies.sh | 10 ++++------
 ci/lib.sh                  |  3 +++
 2 files changed, 7 insertions(+), 6 deletions(-)

index 4f407530d3..b4e22de3cb 100755
@@ -37,15 +37,13 @@ macos-*)
 	test -z "$BREW_INSTALL_PACKAGES" ||
 	brew install $BREW_INSTALL_PACKAGES
 	brew link --force gettext
-	mkdir -p $HOME/bin
-	(
-		cd $HOME/bin
+
+	mkdir -p "$P4_PATH"
+	pushd "$P4_PATH"
 		wget -q "$P4WHENCE/bin.macosx1015x86_64/helix-core-server.tgz" &&
 		tar -xf helix-core-server.tgz &&
 		sudo xattr -d com.apple.quarantine p4 p4d 2>/dev/null || true
-	)
-	PATH="$PATH:${HOME}/bin"
-	export PATH
+	popd
 
 	if test -n "$CC_PACKAGE"
 	then

index c749b21366..f631206a44 100755
@@ -344,6 +344,9 @@ macos-*)
 	then
 		MAKEFLAGS="$MAKEFLAGS APPLE_COMMON_CRYPTO_SHA1=Yes"
 	fi
+
+	P4_PATH="$HOME/custom/p4"
+	export PATH="$P4_PATH:$PATH"
 	;;
 esac
 

2.43.GIT

Add a job to GitLab CI which runs tests on macOS, which matches the equivalent "osx-clang" job that we have for GitHub Workflows. One significant difference though is that this new job runs on Apple M1 machines and thus uses the "arm64" architecture. As GCC does not yet support this comparatively new architecture we cannot easily include an equivalent for the "osx-gcc" job that exists in GitHub Workflows.

Note that one test marked as test_must_fail is surprisingly passing:

t7815-grep-binary.sh (Wstat: 0 Tests: 22 Failed: 0)
TODO passed: 12

This seems to boil down to an unexpected difference in how regcomp(3P) works when matching NUL bytes. Cross-checking with the respective GitHub job shows that this is not an issue unique to the GitLab CI job as it passes in the same way there.

---
 .gitlab-ci.yml | 34 +++++++++++++++++++++++++++++++++-
 ci/lib.sh      |  9 ++++++++-
 2 files changed, 41 insertions(+), 2 deletions(-)

index 793243421c..43bfbd8834 100644
@@ -7,7 +7,7 @@ workflow:
     - if: $CI_COMMIT_TAG
     - if: $CI_COMMIT_REF_PROTECTED == "true"
 
-test:
+test:linux:
   image: $image
   before_script:
     - ./ci/install-docker-dependencies.sh

@@ -52,6 +52,38 @@ test:
       - t/failed-test-artifacts
     when: on_failure
 
+test:osx:
+  image: $image
+  tags:
+    - saas-macos-medium-m1
+  variables:
+    TEST_OUTPUT_DIRECTORY: "/Volumes/RAMDisk"
+  before_script:
+    # Create a 4GB RAM disk that we use to store test output on. This small hack
+    # significantly speeds up tests by more than a factor of 2 because the
+    # macOS runners use network-attached storage as disks, which is _really_
+    # slow with the many small writes that our tests do.
+    - sudo diskutil apfs create $(hdiutil attach -nomount ram://8192000) RAMDisk
+    - ./ci/install-dependencies.sh
+  script:
+    - ./ci/run-build-and-tests.sh
+  after_script:
+    - |
+      if test "$CI_JOB_STATUS" != 'success'
+      then
+        ./ci/print-test-failures.sh
+        mv "$TEST_OUTPUT_DIRECTORY"/failed-test-artifacts t/
+      fi
+  parallel:
+    matrix:
+      - jobname: osx-clang
+        image: macos-13-xcode-14
+        CC: clang
+  artifacts:
+    paths:
+      - t/failed-test-artifacts
+    when: on_failure
+
 static-analysis:
   image: ubuntu:22.04
   variables:

index f631206a44..d5dd2f2697 100755
@@ -252,7 +252,14 @@ then
 	CI_COMMIT="$CI_COMMIT_SHA"
 	case "$CI_JOB_IMAGE" in
 	macos-*)
-		CI_OS_NAME=osx;;
+		# GitLab CI has Python installed via multiple package managers,
+		# most notably via asdf and Homebrew. Ensure that our builds
+		# pick up the Homebrew one by prepending it to our PATH as the
+		# asdf one breaks tests.
+		export PATH="$(brew --prefix)/bin:$PATH"
+
+		CI_OS_NAME=osx
+		;;
 	alpine:*|fedora:*|ubuntu:*)
 		CI_OS_NAME=linux;;
 	*)

2.43.GIT

Hi Patrick

I've read though all the patches and they seem sensible to me though I'm hardly a macOS expert. I did wonder about the use of pushd/popd in the fourth patch as they are bashisms but that matches what we're doing on Ubuntu already. It's nice to see the GitLab CI running on macOS as well as Linux now.

Best Wishes

Phillip

Yeah, that part is a bit weird, agreed. As you say, I basically copied the code that we use on Ubuntu, and that is intentional because another follow-up patch series will rip out that part and move the shared code into a common "install-p4.sh" script. Like that, we can also easily use this script on the Docker-based Ubuntu jobs.

Thanks for your review!

Patrick

Thanks, both.