after_failure script not run if script errors

[lwaldron]

@craigcitro @hadley @jimhester I have the following in my .travis.yml file for an R package at https://github.com/waldronlab/curatedMetagenomicDataCuration. The .push_gh_pages.sh script runs after a successful package check, but not after a package check where the unit tests produce any error. I've left the package intentionally in an erroring state because I want ./push_gh_pages.sh to run even if the unit tests fail (actually, especially if the unit tests fail, because the vignette is designed to help identify the reason for failure).

after_script:
  - ./.push_gh_pages.sh
after_failure:
  - ./.push_gh_pages.sh

[BanzaiMan]

Looks like the build script is bailing out early, without a chance for travis-build to clean up. https://github.com/travis-ci/travis-build/blob/97c7efd0cd1102bbcaaf896536f2ff236687c9df/lib/travis/build/script/r.rb#L239

[lwaldron]

I'm now working around this issue by using a before_script script to build and push to gh-pages before the travis-CI R script is run. FYI this previous version shows how I was originally trying to run a script after success or failure.

[chris-ratcliffe]

@BanzaiMan I am seeing this issue as well. In my script section I include a call to set -e to force processing to stop when any particular command fails, however, that seems to cause the "bail out" to happen, in which case, neither after_script nor after_failure run. If I remove the set -e, both of them run when there is a failure, however, I then have all of my unnecessary steps after the failure continue to run. Is there now way to cause the script section to stop on failure, and make use of the after_* sections?

[ericchenshine]

Is there any update on this issue?

[BanzaiMan]

Sorry, but I have no updates.

[jimhester]

@BanzaiMan Is there a different way to signal failure that would still execute the after_failure block? I can change the script if so.

[BanzaiMan]

At the moment there is not, I'm afraid.

[dotmpe]

I was a bit confounded about an issue with the same behaviour. Until I read the comment by @chris-ratcliffe.

Forgot about some scripts sourced into the build shell, which set -e at the beginning. Problems went away by reversing exit on error again set +e.

It would be useful to mention in the docs, not only about error-on-exit but also advising against using exit and prefer to return. There is an issue on that, maybe if Ihave some time later. travis-ci/docs-travis-ci-com#1672

https://travis-ci.org/bvberkum/script-mpe/builds/354613175