Don't ping everyone on rebase mistakes #1672
Conversation
src/handlers/mentions.rs
Outdated
| .into_iter() | ||
| .any(|commit| commit.commit.author.name == "bors") | ||
| { | ||
| return Ok(Some(MentionsInput::HasBorsCommit)); |
There was a problem hiding this comment.
I think this will cause us to warn on clippy and Miri subtree bumps since those also use bors.
There was a problem hiding this comment.
Yeah, seems like this isn't the most unreasonable behavior for that case though. The PR author could of course ping manually, but I also don't think that it's even necessary in case of a subtree bump
There was a problem hiding this comment.
It does seem just as relevant to ping on a subtree bump if it touches code outside the subtree (as is not uncommon, I believe, though maybe not particularly desirable).
There was a problem hiding this comment.
Oh, I did not know that was possible. How does that happen?
I'll update in that case
There was a problem hiding this comment.
rust-lang/rust#101907 (comment) would also help here -- it would mean we could distinguish between Miri-bors commits and rustc-bors commits.
src/handlers/mentions.rs
Outdated
| None => write!(result, "Some changes occurred in {to_mention}").unwrap(), | ||
| match input { | ||
| MentionsInput::HasBorsCommit => { | ||
| result = config.bors_commit_message.to_string(); |
There was a problem hiding this comment.
I think this doesn't provide a way to opt-out and get the real message (e.g., if you are pulling in miri or clippy, and making changes as part of the merge to other files).
I think one option is to escape all of the mentions and still post the message - that gives the author the ability to re-ping the right folks. Though only if they're a member of the org for teams.
I think a better option is to detect the bors merge commits and then set the PR as draft and bail out. (Probably leaving a comment is still reasonable). That means that the author can just close the PR if they don't want to ping folks or fix it and then mark as ready (which should already retrigger the logic if I'm remembering right).
There was a problem hiding this comment.
That seems quite annoying for the cases in which this is intentional - the current version will warn but at least it doesn't require the PR author to take any additional steps if that's what they intended to do
There was a problem hiding this comment.
The additional step is clicking in the UI on ready for review - that feels pretty lightweight to me? It's also not strictly required - e.g., bors will listen to you in draft state iirc.
There was a problem hiding this comment.
Hmm, looking at the github API, I don't think we can mark someone else's pull request as being a draft. Escaping the pings is possible though, I'll update
There was a problem hiding this comment.
It seems like the conversion is possible, just needs graphql rather than the rest API.
https://docs.github.com/en/graphql/reference/mutations#convertpullrequesttodraft
JakobDegen
force-pushed
the
rebase-mistake
branch
from
9f8c61b to
24668d2
Compare
JakobDegen
force-pushed
the
rebase-mistake
branch
from
24668d2 to
379832d
Compare
| @@ -109,6 +132,11 @@ pub(super) async fn handle_input( | |||
| None => write!(result, "Some changes occurred in {to_mention}").unwrap(), | |||
| } | |||
| if !cc.is_empty() { | |||
| let cc: Vec<String> = if input.has_bors_commit { | |||
| cc.iter().map(|s| format!("`{s}`")).collect() | |||
There was a problem hiding this comment.
On Zulip there was a proposal to also mark the PR as draft, to prevent it from being merge.
Ideally we'd want a scheme where the pings definitely happen when the PR gets approved or lands, but I am not sure if that is possible.
EDIT: Ah I saw that was already discussed and doesn't seem possible. Bummer.
| let has_bors_commit = event.action == IssuesAction::Opened | ||
| && config.bors_commit_message.is_some() | ||
| && event | ||
| .issue | ||
| .commits(&ctx.github) | ||
| .await | ||
| .map_err(|e| log::error!("failed to fetch commits: {:?}", e)) | ||
| .unwrap_or_default() | ||
| .into_iter() | ||
| .any(|commit| commit.commit.author.name == "bors"); |
There was a problem hiding this comment.
Does this even work for problematic rebases? As someone pointed out on Zulip, https://github.com/rust-lang/rust/pull/103918/commits does not have any bors-authored commits.
| assert!(config.paths.len() == 1); | ||
| assert_eq!( | ||
| config.bors_commit_message.as_deref(), | ||
| Some("has bors commit") |
There was a problem hiding this comment.
How could this middle test possibly fail?
This test should be over the overall output, not the config system
See Zulip discussion here. Instead of pinging everyone, we'll send a helpful message with advice on how to correctly rebase.