patchci(windows): ensure that we do not pick up random executables

From: Johannes Schindelin <johannes.schindelin@gmx.de>

On the Windows build agents, a lot of programs are installed, and added to the PATH automatically.

One such program is Git for Windows, and due to the way it is set up, unfortunately its copy of `gpg.exe` is also reachable via the PATH.

This usually does not pose any problems. To the contrary, it even allows us to test the GPG parts of Git's test suite even if `gpg.exe` is not delivered as part of `git-sdk-64-minimal`, the minimal subset of Git for Windows' SDK that we use in the CI builds to compile Git.

However, every once in a while we build a new MSYS2 runtime, which means that there is a mismatch between the copy in `git-sdk-64-minimal` and the copy in C:\Program Files\Git\usr\bin. When that happens we hit the dreaded problem where only one `msys-2.0.dll` is expected to be in the PATH, and things start to fail.

Let's avoid all of this by restricting the PATH to the minimal set. This is actually done by `git-sdk-64-minimal`'s `/etc/profile`, and we just have to source this file manually (one would expect that it is sourced automatically, but the Bash steps in Azure Pipelines/GitHub workflows are explicitly run using `--noprofile`, hence the need for doing this explicitly).

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
---
    ci(windows): avoid using external gpg by mistake
    
    I stumbled over a CI failure this morning when I scrambled to tie up all
    the loose ends for an unexpected v2.33.1 release. Turns out that the CI
    picked up the installed Git for Windows' gpg.exe, and due to a (new, as
    of this morning) mismatch in the MSYS2 runtime between the installed Git
    for Windows and the subset of the Git for Windows SDK used for compiling
    in the CI runs, it worked just enough to pass the prereq, but then
    failed the tests.
    
    Seeing as essentially every single CI build will fail without this patch
    right up until the point in time when Git for Windows is upgraded in the
    build agents (to a version that does yet exist), it would be good to
    fast-track this to maint.
    
    Note: I based this on the earliest topic where it would apply without
    merge conflicts, js/ci-windows-update (which is unfortunately quite
    recent, it is not reachable from any version older than v2.33.0).

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1057%2Fdscho%2Fwork-around-windows-ci-failures-with-gpg-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1057/dscho/work-around-windows-ci-failures-with-gpg-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/1057

 .github/workflows/main.yml | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

So do you want this forked from v2.33.0 (and merged to 'maint' so that v2.33.2, if we need to issue it, would have it)?

https://github.com/git/git/actions/runs/1334961399 is the CI/PR run on v2.33.1 from yesterday. Our PATH wasn't contaminated and we didn't see the problem you fixed here by mere luck, and you were unlucky when you ran the same for generating your release material?

Thanks.

Hi Junio,

On Wed, 13 Oct 2021, Junio C Hamano wrote:

Well, I followed your custom and based it on the oldest applicable topic, js/ci-windows-update. My preference would be for this to be merged into `maint` rather sooner than later, so that GitGitGadget PRs could target `maint` and get a meaningful test coverage.

At the time this ran, the `msys2-runtime` package was already upgraded to a slightly newer version, and even deployed to Git for Windows' Pacman repository. However, that change was not yet reflected in the `git-sdk-64` repository (which is a Git mirror, tracking all the Pacman packages that make up Git for Windows' SDK), as this repository is updated by a nightly automated build.

As a consequence, the `git-sdk-64-minimal` build had not run, therefore your build picked up the previous one, with the earlier `msys2-runtime` version.

That means that the `msys-2.0.dll` file contained in `git-sdk-64-minimal` was identical to the one in Git for Windows as installed on GitHub's build agents.

So yes, the PATH was "contaminated". Meaning: the test suite run did pick up the `gpg.exe` from Git for Windows. And since there was no `msys-2.0.dll` mismatch, it did its job as intended.

The problem arose when I scrambled to get all the things I wanted to wrap up before the next Git for Windows release, this past morning, at which stage the `git-sdk-64-minimal` had already picked up that new `msys2-runtime` version. As a consequence, the MSYS2 Bash we use to run Git's test suite in the CI builds had a `msys-2.0.dll` that was different from the one in `C:\Program Files\Git\usr\bin` (which lives next to the `gpg.exe` and is therefore used when running that program). That, in turn, prevented certain functionality from working. Which made the GPG-related tests fail (maybe not even all of them).

Ciao, Dscho

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

Yes, will do. Tonight's pushout did not have it in 'next', but I do not think of a reason why we shouldn't fast-track the topic.

Thanks.