Refactor integration test core into a new 'obo-tests' crate #60
Conversation
refi64
force-pushed
the
wip/refi64/commander-tests
branch
from
2fa48b7 to
4ced142
Compare
|
@refi64 as a first pass can you fix CI and the conflicts? |
refi64
force-pushed
the
wip/refi64/commander-tests
branch
from
4ced142 to
cb034e3
Compare
|
@sjoerdsimons conflicts fixed, I don't think |
refi64
force-pushed
the
wip/refi64/commander-tests
branch
2 times, most recently
from
fe73a54 to
fb9b128
Compare
refi64
force-pushed
the
wip/refi64/commander-tests
branch
from
fb9b128 to
44c397f
Compare
refi64
force-pushed
the
wip/refi64/commander-tests
branch
3 times, most recently
from
0a098a8 to
0f81984
Compare
| let srcmd5_prefix = format!( | ||
| "srcmd5 '{}' ", | ||
| if log_test == MonitorLogTest::Unavailable { | ||
| ZERO_REV_SRCMD5.to_owned() |
There was a problem hiding this comment.
nitpick; why converting to an owned value if it's being formatting anyway?
There was a problem hiding this comment.
The dir.srcmd5 in the other branch has its lifetime tied to the dir in the same block, so trying to borrow out of there messes with the lifetimes.
| let prune = context.run().command("prune").go().await; | ||
| assert!(!prune.ok()); |
There was a problem hiding this comment.
if you use unwrap_err() instead when it fails it will dump what Ok value got returned. If that's not useful at least assert on is_err()`
There was a problem hiding this comment.
(same pattern in other places)
There was a problem hiding this comment.
I added docs to the traits as mentioned in the other comment, but rough TL;DR prune isn't actually a Result, it's a custom type representing the result of a command. If it failed, it would have already logged that failure too.
refi64
force-pushed
the
wip/refi64/commander-tests
branch
2 times, most recently
from
10135e3 to
4661b3f
Compare
refi64
changed the title
This is largely just a rename, as well as the slightly-silly `obo-test-support` crate that exists to be depended on by `obo-core` *and* `obs-gitlab-runner`.
Most of this logic is rather intricate and shareable between multiple implementations, so it makes sense for them to be in a separate crate. This necessitates quite a bit of refactoring of the test structure so that the shared test code can actually run commands and inspect the logs + artifacts. Since there are various somewhat strange requirements around command execution (i.e. needing to use timeouts but still get the result), that API was redone around a builder trait, which also makes it easy for implementation-specific tests to have their own methods on it.
refi64
force-pushed
the
wip/refi64/commander-tests
branch
from
4661b3f to
9d78d16
Compare
As usual, only the last commit is new, depends on #58.