Download Microsoft Git from GitHub releases #1853

mjcheetham · 2025-07-28T17:04:10Z

We've not been running the functional tests against a new version of Microsoft Git in a while.. today we're still pointing at v2.35.3.vfs.0.0 which was released in 2022!

The GVFS.Installers project (that creates the NuGet packages that the Functional Tests consume) is set to download the Git installer and portable archives from the ci/Dependencies feed in the GVFS Azure DevOps organisation. The package version is set centrally for the build here, in the Version.props file.

The build that publishes new versions of that package are woefully out of date and no longer run. The last run was June 20, 2022. This was around the time official microsoft/git builds were moved to GitHub workflows.

Here we instead source the Microsoft Git installers from GitHub releases in the build workflow. Updating the version of Git used in FTs can be updated by modifying the build.yaml workflow (or setting the variable at queue-time if doing a workflow_dispatch trigger).

Whilst we're at it, and since we now have ARM64-natively compiled versions of Git and Windows on ARM hosted runners, let's also run the functional tests on ARM64.

The functional tests have also been failing with the newer version of Microsoft Git since microsoft/git@5efe366 was introduced. Let's ignore these messages in the FTs.

~~Note: this is based upon this PR~~

dscho · 2025-07-28T18:50:26Z

I believe that the tests fail because of microsoft/git@5efe366, as @tyrielv had pointed out to us in private communication earlier.

So we'll need to adjust something in the tests to accommodate for the change of behavior.

I guess the easiest would be to patch

VFSForGit/GVFS/GVFS.FunctionalTests/Tools/GitHelpers.cs

Line 112 in 2ad18e8

    
           ProcessResult actualResult = GitHelpers.InvokeGitAgainstGVFSRepo(gvfsRepoRoot, command, environmentVariables);

to strip out the newly-added lines from actualOutput.

mjcheetham · 2025-07-28T18:53:50Z

I believe that the tests fail because of microsoft/git@5efe366, as @tyrielv had pointed out to us in private communication earlier.

So we'll need to adjust something in the tests to accommodate for the change of behavior.

I guess the easiest would be to patch

VFSForGit/GVFS/GVFS.FunctionalTests/Tools/GitHelpers.cs

Line 112 in 2ad18e8

ProcessResult actualResult = GitHelpers.InvokeGitAgainstGVFSRepo(gvfsRepoRoot, command, environmentVariables);

to strip out the newly-added lines from actualOutput.

That's what e6e4504 is (trying) to do.. but I'm debugging why it's not working at the moment.

mjcheetham · 2025-07-28T20:51:07Z

but I'm debugging why it's not working at the moment.

Of course status prints to standard output, and the existing filtering logic works on standard error. Should work now - we're filtering both.

dscho

Great!

I am only a bit concerned that the Functional Tests still fail:

GVFS.FunctionalTests.Tests.EnlistmentPerFixture.GitCorruptObjectTests.GitRequestsReplacementForTruncatedObject():
```
Failed at 10:54:50 PM taking 00:00:01.4814157
  Expected: 0
  But was:  128
```

GVFS.FunctionalTests.Tests.EnlistmentPerFixture.WindowsUpdatePlaceholderTests.LockToPreventUpdateAndDelete():

Test GVFS.FunctionalTests.Tests.EnlistmentPerFixture.WindowsUpdatePlaceholderTests.LockToPreventUpdateAndDelete()
Failed at 10:57:05 PM taking 00:00:01.4125659
  Expected substring 'Untracked files:
  (use "git add <file>..." to include in what will be committed)
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test_delete.txt
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test_delete2.txt
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test_delete3.txt' not found in 'HEAD detached at 5d7a7d4db1734fb468a4094469ec58d26301b59d
Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git restore <file>..." to discard changes in working directory)
	modified:   Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test.txt
	modified:   Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test2.txt
	modified:   Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test3.txt
Untracked files:
  (use "git add <file>..." to include in what will be committed)
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test_delete.txt
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test_delete2.txt
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test_delete3.txt
no changes added to commit (use "git add" and/or "git commit -a")'
  Expected: True
  But was:  False

GVFS.FunctionalTests.Tests.EnlistmentPerFixture.WindowsUpdatePlaceholderTests.LockMoreThanMaxReportedFileNames():

Test GVFS.FunctionalTests.Tests.EnlistmentPerFixture.WindowsUpdatePlaceholderTests.LockMoreThanMaxReportedFileNames()
Failed at 10:57:05 PM taking 00:00:00.2646733
  Expected substring 'HEAD is now at fec239ea12de1eda6ae5329d4f345784d5b61ff9' not found in 'error: Your local changes to the following files would be overwritten by checkout:
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test.txt
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test2.txt
	Test_EPF_UpdatePlaceholderTests/LockToPreventUpdateAndDelete/test3.txt
Please commit your changes or stash them before you switch branches.
Aborting'
  Expected: True
  But was:  False

As of 9e59b38, Git will loudly complain about corrupt objects. That is fine, as long as the idea isn't to re-download locally-corrupted objects. But that's exactly what we want to do in VFS for Git. This is even tested for in the functional tests of VFS for Git, which has been identified as a regression in microsoft/VFSForGit#1853.

Stop creating and uploading (for the functional tests) NuGet packages for the GVFS.Installers project. We no longer maintain the packages: * GitForWindows.GVFS.Installer * GitForWindows.GVFS.Portable Signed-off-by: Matthew John Cheetham <[email protected]>

Download the Microsoft Git installers from GitHub releases via the `gh` CLI tool. The installers (both portable and Inno Setup based) are included in the Installers_${{ matrix.configuration }} build artifact that is consumed by the functional test jobs. Signed-off-by: Matthew John Cheetham <[email protected]>

Install the correct version of Microsoft Git in the functional tests for the OS architecture, and print more OS/CPU information for good measure. Signed-off-by: Matthew John Cheetham <[email protected]>

Let's use the ARM64 Windows images (currently in public preview) on GitHub hosted runners to run the functional tests on ARM. VFS for Git is still only built for x86_64, but we do now include a natively compiled Git for ARM64 - so let's exercise that! Signed-off-by: Matthew John Cheetham <[email protected]>

In preparing to accounting for the "You are in a partially-hydrated checkout with %d% of tracked files present" message that was added in Microsoft Git v2.XX.Y, let's refactor the existing output filtering (which is so far only done for `stderr`) into its own method. Signed-off-by: Matthew John Cheetham <[email protected]> Signed-off-by: Johannes Schindelin <[email protected]>

There are tests that find multi-line needles in the Git output's haystack. So far, those tests work on `stdout` only. We are about to filter `stdout` using the new `FilterMessages()` method, therefore that method must stop its current practice where it splits by newlines and then joins using `Environment.Newline` (even if that is different from what the input had before). Signed-off-by: Johannes Schindelin <[email protected]>

Currently, we only filter `stderr`, but we're about to do that with `stdout`, too, where we do not want to skip empty lines. Signed-off-by: Johannes Schindelin <[email protected]>

Let the functional tests account for the "You are in a partially-hydrated checkout with %d% of tracked files present" message that was added in Microsoft Git v2.45.2.vfs.0.2. Do this by filtering out those messages in the same way we do for the old 'A new version of VFS for Git is available.' upgrader messages, and the 'Waiting for %s' lock waiting message. Note that the 'hydration' state message is printed to standard output and not standard error as the other messages, since this is the output from the `status` command. We filter both stdout and stderr. Signed-off-by: Matthew John Cheetham <[email protected]> Signed-off-by: Johannes Schindelin <[email protected]>

When running the functional tests locally, it is quite possible that the `core.fsmonitor` setting is enabled globally. Let's avoid chalking that up as a test failure. Signed-off-by: Johannes Schindelin <[email protected]>

dscho

Awesome!

mjcheetham requested review from dscho and tyrielv July 28, 2025 17:04

mjcheetham force-pushed the gh-msgit branch 4 times, most recently from a93cfa6 to 668408a Compare July 28, 2025 17:35

mjcheetham requested a review from mpysson July 28, 2025 18:21

mpysson approved these changes Jul 28, 2025

View reviewed changes

mjcheetham force-pushed the gh-msgit branch from e6e4504 to fe81203 Compare July 28, 2025 20:50

mjcheetham force-pushed the gh-msgit branch from fe81203 to f3db68e Compare July 28, 2025 22:48

dscho approved these changes Jul 29, 2025

View reviewed changes

mjcheetham mentioned this pull request Aug 8, 2025

Improved Git version parsing to support release candidates #1854

Merged

mjcheetham force-pushed the gh-msgit branch from f3db68e to 0228e0b Compare August 13, 2025 08:21

dscho mentioned this pull request Aug 13, 2025

Do re-download corrupt objects with GVFS microsoft/git#782

Merged

mjcheetham force-pushed the gh-msgit branch 2 times, most recently from 09ac6be to e3c33ca Compare August 14, 2025 10:57

mjcheetham force-pushed the gh-msgit branch 5 times, most recently from e85cafc to 3b663b4 Compare August 14, 2025 12:24

mjcheetham and others added 8 commits August 14, 2025 13:38

ft: install arch-specific msftgit and print OS version

3d6f13d

Install the correct version of Microsoft Git in the functional tests for the OS architecture, and print more OS/CPU information for good measure. Signed-off-by: Matthew John Cheetham <[email protected]>

Functional Tests: support retaining empty lines in FilterMessages()

23001d5

Currently, we only filter `stderr`, but we're about to do that with `stdout`, too, where we do not want to skip empty lines. Signed-off-by: Johannes Schindelin <[email protected]>

Functional Tests: heed Postel's law regarding core.FSMonitor

c69ba8e

When running the functional tests locally, it is quite possible that the `core.fsmonitor` setting is enabled globally. Let's avoid chalking that up as a test failure. Signed-off-by: Johannes Schindelin <[email protected]>

mjcheetham force-pushed the gh-msgit branch from 3b663b4 to c69ba8e Compare August 14, 2025 12:38

mjcheetham requested a review from dscho August 14, 2025 13:00

dscho approved these changes Aug 14, 2025

View reviewed changes

mjcheetham merged commit f2bc533 into microsoft:master Aug 14, 2025
8 checks passed

mjcheetham deleted the gh-msgit branch August 14, 2025 14:01

mjcheetham mentioned this pull request Aug 18, 2025

Release M262 #1857

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Download Microsoft Git from GitHub releases #1853

Download Microsoft Git from GitHub releases #1853

Uh oh!

mjcheetham commented Jul 28, 2025 •

edited

Loading

Uh oh!

dscho commented Jul 28, 2025 •

edited

Loading

Uh oh!

mjcheetham commented Jul 28, 2025

Uh oh!

mjcheetham commented Jul 28, 2025

Uh oh!

dscho left a comment

Uh oh!

dscho left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Download Microsoft Git from GitHub releases #1853

Download Microsoft Git from GitHub releases #1853

Uh oh!

Conversation

mjcheetham commented Jul 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dscho commented Jul 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mjcheetham commented Jul 28, 2025

Uh oh!

mjcheetham commented Jul 28, 2025

Uh oh!

dscho left a comment

Choose a reason for hiding this comment

Uh oh!

dscho left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

mjcheetham commented Jul 28, 2025 •

edited

Loading

dscho commented Jul 28, 2025 •

edited

Loading