Add reminder to use ready shortcut when using author shortcut

[Urgau]

This PR adds a small reminder to use @bot ready when using @bot author, similar to what was proposed in in #1865 (comment).

This is done to help newcomers, which might not know or have forgot what the command is.

@rustbot author

Reminder, use @{bot} ready to indicate that the PR is ready for review.

To avoid spam the bot adds the reminder to the comment using the command instead of creating a new comment.

To avoid spam the bot only post the reminder once per PR.

cc @RalfJung
r? @ehuss

[RalfJung]

To avoid spam the bot adds to the comment using the comment instead of creating a new comment.

I assume one of these "comment" should be "command" :D

[jackh726]

Not quite sure I like the fact that this edits comments. Would rather post a new comment once (and then no more).

[Urgau]

Would rather post a new comment once (and then no more).

We do already post such comment, but from my experience it doesn't seem to work very much, in particular for new-comers.

I think that is mainly due to due to the time between the comment and the usage of @rustbot author by reviewers.

[jackh726]

Well...then maybe we should just post the comment every time. (We could e.g. even filter if the user is not in the team repo.)

Edit: just looked at the link, that's not quite what I was saying, it could still be valuable to have another comment create in response to the author command

[jackh726]

Also, this doesn't handle the case were the label is manually changed, right? I imagine that happens plenty often.

[Urgau]

Edit: just looked at the link, that's not quite what I was saying, it could still be valuable to have another comment create in response to the author command

I fear that posting a comment instead of editing the comment will quickly be annoying/spam.

I personally wouldn't want the bot to add a comment every-time I (or someone else) uses the author command. I would probably find it too distracting and annoying, as it would mean that I would be getting a new notification, that I would then need to triage (mark the email as read or mark it as read on GitHub UI).

@RalfJung, as the issue author, would you happen to have an opinion/preference on the new comment vs edit comment?

Also, this doesn't handle the case were the label is manually changed, right? I imagine that happens plenty often.

Yes it wouldn't, and that's a conscience choice from me. I think that changing labels and using a command are two distinct way of doing an action and that we shouldn't see them having the same intent.

I use the command with newcomers but often switch the labels manually with regular contributors (as it's faster and they know the process).

[RalfJung]

We do already post such comment, but from my experience it doesn't seem to work very much, in particular for new-comers.

We also post that only on their first PR.

I fear that posting a comment instead of editing the comment will quickly be annoying/spam.

Yeah that's why my original proposal involved the reviewer using a slightly different command to express their intent of posting a hint for the PR author.

How easy would it be to only post a comment the first time someone uses @rustbot author in a PR?

[Urgau]

How easy would it be to only post a comment the first time someone uses @rustbot author in a PR?

Not too difficult, we have the infra to have data per issue/pr inside triagebot.

---

So if I understand, the consensus seems to be to post the reminder once in a PR in a new comment at the first use of @rustbot author, is that right?

I could get behind that, I don't think it will be as effective but it also won't be annoying and spam; and as Jack alluded to earlier we could even omit the reminder when the PR author is in the team database.

[ehuss]

A minor concern I have is that I think editing comments could be flaky or dangerous. For example, any handlers that run after this one will not see the edited comment. If they do other strange things like editing, then they will conflict with one another. Another concern is that editing the comment will trigger another webhook, and triagebot will process the comment again. Although triagebot is pretty good about not re-running commands on edit, I don't have 100% confidence in it.

[Urgau]

That's a very good argument that by editing a comment we could render some state inconsistent within triagebot.

Given this and what seems to be the consensus, I have switch the implementation to post a unique new comment at the first use of the author shortcut, instead of editing the comment.