Allow user to specify branch of systemtests being triggered by trigger_systemtests.py #143
Conversation
|
Exemplary usage of the feature implemented here is shown in precice/fenics-adapter#60: Excerpt from .travis.yml |
BenjaminRodenberg
force-pushed
the
develop
branch
from
e936bdf
to
968fe69
Compare
| repo = adapters_info[ args.adapter ].repo | ||
| trigger_travis_build( generate_failure_callback(), args.owner,repo) | ||
| repo = adapters_info[args.adapter].repo | ||
| trigger_travis_build(generate_failure_callback(branch=args.branch), args.owner, repo) |
There was a problem hiding this comment.
This conditional branch that contains this line is only ever reached in the following case:
-
Some external repository, let's say fenics-adapter, commits something that triggers Travis to execute the
.travis.ymlfile. This file contains the crucial linepython trigger_systemtests.py --adapter fenics, which triggers a build on systemtests (and with this PR merged, on a specific branch of it). Note that the--waitargument is omitted, meaning that the fenics-adapter build on Travis will just send the request and instantly return as success. -
A systemtests build is launched to run the test that was requested. If this test now fails we reach the after_failure_action
systemtests/trigger_systemtests.py
Lines 105 to 106 in 55ffcc0
So, the systemtests build will send a request back to fenics-adapter which originally triggered the whole sequence.
At this point we have reached the highlighted line 282.
- On fenics-adapter, a mock build is launched. This build will just fail and do nothing else, so that the actual feedback for a failed test is negative on the original repository as well.
systemtests/trigger_systemtests.py
Line 282 in 55ffcc0
Because of this, we have to be careful with which branch we choose here, as this must not be a branch of systemtests. Currently, this mock build would always run on the master branch of the fenics-adapter. It would be nice to have this actually run on the specific branch that was originally commited to in point 1. Note that this is again a different branch than the systemtests branch that will be adjustable with this PR.
There was a problem hiding this comment.
Since the mock request is generated on systemtests, we would need it to have received information in some way about which branch of the outside repository triggered it. One way might be to have the original request contain the following as after failure action
if trigger_failure:
after_failure_action += " python trigger_systemtests.py --failure --owner {USER} --adapter {ADAPTER}" + "--branch {}".format(os.environ["TRAVIS_BRANCH"])
This way, when systemtests is executing trigger_systemtests.py as a result of a failed test, it will have received an args.branch value that correctly contains the branch of the outside repo that originally issued the test. Then we could actually call generate_failure_callback(branch=args.branch)
This way of formatting seems rather ugly, but it doesn't interfere with the formatting step done later on in
systemtests/trigger_systemtests.py
Lines 158 to 160 in 55ffcc0
We also don't require any further variables this way.
Suggestions?
See #143 (comment) and #143 (comment) for details Co-Authored-By: Konrad Eder <[email protected]>
closes #141