GitHub ci(windows): speed up initializing Git for Windows' minimal SDK again

From: Johannes Schindelin <[email protected]>

It used to be the case that initializing the minimal SDK (i.e. a radically slimmed-down subset of Git for Windows' development environment intended to perform the CI builds and little else) took a bit over one minute, would then be cached, and subsequent jobs would take at most half a dozen seconds to initialize said minimal SDK.

It is important that this step is fast because we have to run the test suite in parallel, in a set of matrix jobs, to offset the slowness of the shell-based test suite, and each and every job has to initialize the very same minimal SDK.

While it may sound as if parallelizing the jobs might only waste the generously-provided build minutes but at least the _wallclock_ time would pass quick, in reality it matters a lot: Frequently Git for Windows' or GitGitGadget PRs get stuck waiting for quite a while before CI builds start because other PRs' builds still spend substantial amounts of time to run, blocking due to the concurrency limit being reached.

Since 91839a88277 (ci: create script to set up Git for Windows SDK, 2024-10-09), the situation has worsened: every job that requires the minimal Git for Windows SDK spends roughly two-and-a-half minutes doing so.

With the switch away from the GitHub Action setup-git-for-windows-sdk, we incurred more downsides:

- It is no longer possible for said Action to fix problems independently from the Git repository, e.g. when new rules about GitHub Actions require changes in the way the minimal SDK is initialized.

- The minimal SDK was installed specifically outside of the worktree so as not to clutter it nor incur an additional cost to verify that the worktree is clean.

Therefore, even if it would be nice to have a shared process between GitHub and GitLab based CI builds, let's switch the GitHub-based CI back to the tried-and-tested setup-git-for-windows-sdk Action.

This commit partially reverts 91839a88277 (ci: create script to set up Git for Windows SDK, 2024-10-09).

---
Speed up the Git for Windows SDK initialization again

While waiting for way too many builds in
https://github.com/gitgitgadget/git/actions to finish, I noticed that the minimal Git for Windows SDK initialization now takes a whopping 2.5 minutes. That's way too much. It used to take a little over a minute when uncached, and 2-5 seconds when cached.

Let's fix this regression by reverting to using the setup-git-for-windows-sdk GitHub Action (also because that Action will soon see another dramatic speed-up, see
https://github.com/git-for-windows/setup-git-for-windows-sdk/pull/965).

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1841%2Fdscho%2Fci-windows-jobs-speedup-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1841/dscho/ci-windows-jobs-speedup-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/1841

 .github/workflows/main.yml | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

index 9301a1edd6d..916a64b6736 100644
@@ -113,15 +113,13 @@ jobs:
       cancel-in-progress: ${{ needs.ci-config.outputs.skip_concurrent == 'yes' }}
     steps:
     - uses: actions/checkout@v4
-    - name: setup SDK
-      shell: powershell
-      run: ci/install-sdk.ps1
+    - uses: git-for-windows/setup-git-for-windows-sdk@v1
     - name: build
-      shell: powershell
+      shell: bash
       env:
         HOME: ${{runner.workspace}}
         NO_PERL: 1
-      run: git-sdk/usr/bin/bash.exe -l -c 'ci/make-test-artifacts.sh artifacts'
+      run: . /etc/profile && ci/make-test-artifacts.sh artifacts
     - name: zip up tracked files
       run: git archive -o artifacts/tracked.tar.gz HEAD
     - name: upload tracked files and build artifacts

@@ -149,12 +147,10 @@ jobs:
     - name: extract tracked files and build artifacts
       shell: bash
       run: tar xf artifacts.tar.gz && tar xf tracked.tar.gz
-    - name: setup SDK
-      shell: powershell
-      run: ci/install-sdk.ps1
+    - uses: git-for-windows/setup-git-for-windows-sdk@v1
     - name: test
-      shell: powershell
-      run: git-sdk/usr/bin/bash.exe -l -c 'ci/run-test-slice.sh ${{matrix.nr}} 10'
+      shell: bash
+      run: . /etc/profile && ci/run-test-slice.sh ${{matrix.nr}} 10
     - name: print test failures
       if: failure() && env.FAILED_TEST_ARTIFACTS != ''
       shell: bash

base-commit: 631ddbbcbd912530e1b78e5d782e72879f7f1fb2

gitgitgadget

"Johannes Schindelin via GitGitGadget" <[email protected]>

Thanks, will queue.

Okay. Too bad that things regressed this badly with the change. I agree that reverting is the right thing to do for now. I may revisit this again in the next release cycle to investigate whether we can get it up to par with the GitHub Actions. It would be great if the build infra was shared between our CIs, so I think there's some value in it. But if the answer is "no" then I guess that's ultimately fine, as well.

Thanks!

Patrick

Hi Patrick,

The way I implemented it should be eminently possible in PowerShell, too.

Something like Start-Process, launching not an Invoke-WebRequest but instead C:\Windows\system32\curl.exe [*1*] to download the .tar.gz file (not the .zip file, more about that below). Another Start-Process should then be opened that executes tar.exe [*2*], and then the stdout of the former should be piped to the latter [*3*].

I do think that it makes a lot of sense to start extracting as soon as the first byte arrives instead of storing the archive as a temporary file and extracting it only once it has arrived completely.

Now, why not use the .zip file instead of the .tar.gz file? In my analysis [*4*], I pointed out that the .zip file is about 10MB smaller, after all, and BSD tar (at least the version in C:\Windows\system32) is able to handle those, too, right? Not so fast. In my experiments, when streaming the .zip file to the tar.exe -xf - process, the etc/ and .sparse/ directories were consistently dropped. A bug, I guess. I ran out of time to investigate this in more depth.

Since the artifacts are now hosted in a GitHub Release and updated regularly, and since those updates cannot be atomic (you can only upload a release asset if no asset of the same name exists already, read: the automation has to _delete_ that asset before uploading a new version), it would also be good to kind of expect that the file may be intermittently absent and add a back-off strategy [*5*].

It _could_ be done. But the advantages of having it versioned outside of the Git repository outweigh the benefits of that shared infrastructure, I believe.

Ciao,
Johannes

Footnote *1*:
https://github.com/git-for-windows/setup-git-for-windows-sdk/pull/965/commits/6db65223de699c4f75ab083f82f43947a53ad6ff#diff-6855ef61b94227f9264adab3ff9f2de95c2d7b4e451019cc0105896d32550eb0R58-R73

Footnote *2*:
https://github.com/git-for-windows/setup-git-for-windows-sdk/pull/965/commits/6db65223de699c4f75ab083f82f43947a53ad6ff#diff-6855ef61b94227f9264adab3ff9f2de95c2d7b4e451019cc0105896d32550eb0R77-R86

Footnote *3*:
https://github.com/git-for-windows/setup-git-for-windows-sdk/pull/965/commits/6db65223de699c4f75ab083f82f43947a53ad6ff#diff-6855ef61b94227f9264adab3ff9f2de95c2d7b4e451019cc0105896d32550eb0R88

Footnote *4*:
https://github.com/git-for-windows/git-sdk-64/pull/87/commits/fdb0cea373893ce7d40bcfcfbeb7fd091a3c4020

Footnote *5*:
https://github.com/git-for-windows/setup-git-for-windows-sdk/pull/965/commits/3d4ea07041d0740b21160a9d9a4181f569e706d8