mozlint job started by reviewbot should run with outgoing option (-o)
Categories
(Developer Infrastructure :: Source Code Analysis, task, P2)
Tracking
(Not tracked)
People
(Reporter: Sylvestre, Unassigned)
References
(Blocks 1 open bug)
Details
Currently, if we have a one line change in a js file, we are still going to run eslint on the whole tree at review phase.
For example, for this job, https://treeherder.mozilla.org/#/jobs?repo=try&revision=a6ba46c4bdd8816ffa2956efbc1c0256508380a8
We are updating two js tests:
https://hg.mozilla.org/try/rev/9ecdfe00125402e77937b5c0d1514c7afcecc80d
This triggers the whole eslint analysis (18m!)
https://treeherder.mozilla.org/logviewer.html#?job_id=301525276&repo=try
while -o would have been enough.
AFAIK, all checkers called by mozlint (besides clippy - but mozlint clippy is going to rebuild the whole crate) are not aware of the context. This means that we should not have any issue like "I touched this API and it triggered an issue in the consumer of this API".
Even if "-o" decreases the quality of the analysis, we could run the full job either in autoland or in m-c for the specific checker.
Configuration is done here:
https://searchfox.org/mozilla-central/source/taskcluster/ci/source-test/mozlint.yml
Reporter | ||
Updated•5 years ago
|
Reporter | ||
Comment 1•5 years ago
|
||
In term of cost, it isn't a priority, reviewbot is pretty cheap!
Reporter | ||
Comment 2•5 years ago
|
||
I discussed with Andrew about this.
if we didn't do earlier, here is a risk that, if we miss a patch for any reason, we might have an incorrect state (ie an error in a tree and the tool would not catch it).
To mitigate this, we should create a mozlint task which will run all the checkers into a single take (maybe in m-c only)
Comment 3•5 years ago
|
||
Yeah, I think we should make the tasks configurable such that on try they run with --outgoing
and on autoland/m-c they run the full set.
Updated•5 years ago
|
Updated•3 years ago
|
Description
•