Add support for diff-informed queries #2559
Conversation
src/actions-util.ts
Outdated
| /** | ||
| * Deepen the git history of the given ref by one level. Errors are ignored. | ||
| */ | ||
| export const deepenGitHistory = async function () { |
There was a problem hiding this comment.
Two comments about the additions to this file in general:
- Why completely ignore errors? Does it make sense to at least debug log them?
- Nit, optional: This PR adds a bunch of functionality that interacts with git — now might be a good time to pull this functionality out into its own file.
There was a problem hiding this comment.
The error is indeed logged centrally within runGitCommand(). By "Errors are ignored", I was trying to convey that errors are not reported back to the caller. I have update the function documentation to clarify the behavior.
Moving the git interactions to its own file sounds like a good idea, though trying to incorporate that into this PR is a bit messy. Can I defer that to a dedicated PR after this one merges?
| const results = new Array<[string, number, number]>(); | ||
|
|
||
| let changedFile = ""; | ||
| for (const line of diffHunkHeaders) { |
There was a problem hiding this comment.
In general this seems to keep going when it sees certain kinds of unexpected items in the diff. It seems that there's a risk of missing changed files — should we be more robust here, or fail more eagerly, or is that not important for now?
There was a problem hiding this comment.
Can you tell me more about the risk you see, and how the code can be made more robust?
There was a problem hiding this comment.
Here are some examples where we seem to find unexpected items in the diff:
// Not an actual file path; probably /dev/nullFailed to parse diff hunk header line: ${line}. Skipping.
In these cases we seem to be encountering something unexpected, but we log or ignore the error and carry on. My question is could these cases where the diff isn't quite as we expect cause us to miss some files that have been changed by the PR, and therefore cause us to miss alerts? If so, we'd probably want to return undefined and fall back to a full analysis.
There was a problem hiding this comment.
That is a great observation. Thank you! I have updated the PR accordingly.
cklin
force-pushed
the
cklin/diff-range-filtering
branch
from
d757098
to
38487cd
Compare
Thanks for the detailed comments and helpful suggestions! PTAL. |
cklin
force-pushed
the
cklin/diff-range-filtering
branch
from
38487cd
to
94b5d39
Compare
I updated the PR, which implements your suggestion and also fixes a minor bug in the handling of deletion hunks (where git reports the insertion line range as 0). See diff for the latest force-push for details. |
Merge / deployment checklist