Homebrew
Formula fails to build but workflow succeeds #2929
-
|
I setup CI for the coin-or-tools/coinor tap using the yml files that Homebrew creates when initializing a new tap. Upon my initial PR where I iteratively fix the formulae to keep test-bot happy, I noticed that the workflow can be marked as a success even though certain formulae fail to build. Here is an example: https://github.com/coin-or-tools/homebrew-coinor/runs/5015187989?check_suite_focus=true#step:8:292 In the Action summary, the build error only appears as a warning: https://github.com/coin-or-tools/homebrew-coinor/actions/runs/1775923851 The I find that that somewhat defeats the purpose of CI and requires manual examination of the logs, even in case of success. Is there an easy way to make the job and workflow fail when a formula fails to build? Many thanks! |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
|
The “warning” behaviour only kicks in when a formula has no bottle on the given OS. If it only warns on build failure when a formula has a pre-existing bottle, that’s a bug. Still, it’s not the first I’ve heard of a request to change this behaviour. (Hi @ipatch.) A |
Beta Was this translation helpful? Give feedback.
-
|
Touché! :-) Many thanks for your help. I haven't been able to verify that there is indeed an error if a bottle already exists. If so, that will help greatly. I'll keep the PR in mind but the Homebrew code is too sophisticated for my babbling Ruby. Do you have a pointer to get started? |
Beta Was this translation helpful? Give feedback.
-
|
All decisions to ignore a failure flow through here: https://github.com/Homebrew/homebrew-test-bot/blob/e486c73e3067672689ab9d0c1c50fe2a2d7e6c62/lib/step.rb#L157-L163 It should be straightforward to add a command-line switch to override ignoring and just fail. It might take a little bit of work getting |
Beta Was this translation helpful? Give feedback.
The “warning” behaviour only kicks in when a formula has no bottle on the given OS. If it only warns on build failure when a formula has a pre-existing bottle, that’s a bug.
Still, it’s not the first I’ve heard of a request to change this behaviour. (Hi @ipatch.) A
-Werror-type flag fortest-botshould do this and be relatively easy to implement. (We probably won’t want to be selective about which warnings we promote to errors, though, as that would make things too complicated.) Happy to review a PR that does this.