GitHub Actions on Windows doesn't stop on error

[justanotheranonymoususer]

Description

Here's a pipeline which demonstrates the bug, it should fail but it keeps on running, leading to unexpected consequences in the rest of the pipeline:

  build-windows-failure-bug:
    runs-on: windows-latest
    steps:
      - uses: actions/checkout@v3
      - name: Fail test
        run: |
          curl https://does-not-exist.this-will-fail/
          curl https://example.com/

Here's a more complete example:
https://github.com/justanotheranonymoususer/actions-test/blob/5c0935375762cfbcc8f6231aae60c50a6c2b284b/.github/workflows/blank.yml

Execution result:
https://github.com/justanotheranonymoususer/actions-test/actions/runs/3569092631

You can see that on ubuntu-latest, it fails properly, but in windows-latest, it doesn't fail even though it should.

Platforms affected

• Azure DevOps
• GitHub Actions - Standard Runners
• GitHub Actions - Larger Runners

Runner images affected

• Ubuntu 18.04
• Ubuntu 20.04
• Ubuntu 22.04
• macOS 10.15
• macOS 11
• macOS 12
• Windows Server 2019
• Windows Server 2022

Image version and build link

windows-latest

Is it regression?

Not/don't know

Expected behavior

Windows pipeline fails on first error

Actual behavior

Windows pipeline keeps running

Repro steps

See pipeline in the description

[ddobranic]

hello @justanotheranonymoususer, thank you! we will investigate.

[ddobranic]

@justanotheranonymoususer you could slightly modify you workflow and get the same results for Ubuntu and Windows.
Here is an example:

build-windows-failure-bug:
    runs-on: windows-latest
    steps:
      - uses: actions/checkout@v3
      - name: Fail test
        run: |
          curl https://dfgjndrgierngiergnregerg
          curl https://example.com/
        shell: bash

[ddobranic]

@justanotheranonymoususer, i am going to close the issue. if you have any questions or concerns, please feel free to reopen.

[justanotheranonymoususer]

You showed a workaround, but the default behavior is still undesired, is it?

…

On Wed, Dec 7, 2022, 12:07 Dusko Dobranic ***@***.***> wrote: Closed #6668 <#6668> as completed. — Reply to this email directly, view it on GitHub <#6668 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABMDRPCJLHPA5ZLRWA47K5TWMBOWJANCNFSM6AAAAAASQE7TYI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[ddobranic]

@justanotheranonymoususer it is about default behaviors of the two different shells. for bash, we set 'e' flag which gives exact behavior you are expecting - exit on error.

[justanotheranonymoususer]

I did a couple of other checks. I tried these shells: bash, sh, powershell, pwsh, cmd.

• These continue even if there's an error: pwsh, cmd
• These abort as expected: bash, sh, powershell

Here's what the docs say:

cmd

There doesn't seem to be a way to fully opt into fail-fast behavior other than writing your script to check each error code and respond accordingly. Because we can't actually provide that behavior by default, you need to write this behavior into your script.

pwsh (default on Windows)

Fail-fast behavior when possible. For pwsh and powershell built-in shell, we will prepend $ErrorActionPreference = 'stop' to script contents.

For cmd, it's known and documented that it continues after a failure. But for pwsh, it should abort after the first error, which seems like a bug as it actually continues execution.

I'm not sure what "when possible" means in the PowerShell documentation. I think it should be fixed, and, if not possible, documented properly.

[Gustl22]

@ddobranic how I can achieve the same behavior (to abort) with powershell or pwsh as I need some of their commands?

[coffeebe4code]

Same issue. Windows passes but linux fails. cargo test fails, but the e2e is successful. set -e should be the default

  build-win:
    runs-on: windows-latest
    steps:
      - uses: actions/checkout@v2
      - uses: actions-rs/toolchain@v1
        with:
          toolchain: stable
      - run: |
          cargo build --release
          cargo test
          ./target/release/e2e

  build-linux:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2
      - uses: actions-rs/toolchain@v1
        with:
          toolchain: stable
      - run: |
          cargo build --release
          cargo test
          ./target/release/e2e

[beamerblvd]

Should a new bug be filed, or is it possible to re-open this bug? Per this comment, shell pwsh is not working as documented, which is a bug (either with the workflow or the documentation). Worse, shell: cannot be parameterized with the matrix, so the choices are to use Bash with everything or have a completely different workflow script for Windows vs other OSes, neither of which are ideal choices.