[service] Reduce the number of days until stale

[prince-chrismc]

This will only affect 22 PRs which have not been touched. Only 4 are less than 20 days.

Motivation: This will hopefully make it easier to spot PRs that could use help where the OP got stuck.

There is currently ~60 PRs that are 'inactive', or at least, with have no Human events.

• Recipes like boost that received lots of PRs are being kept alive by the bots.
• There's a section of PRs that had comments hidden which also refreshed their life span.

Labeling them sooner will make it easier to filter PRs.

[conan-center-bot]

Changes not allowed in build 1:

[.github/stale.yml]

Split them in several pull requests if you are making changes in more than one folder.

[Croydon]

I don't like this. I even considered in the past to suggest increasing the number of days. There were several cases, where the bot was just annoying and that should not be the case.

Recipes like boost that received lots of PRs are being kept alive by the bots.

Do comment edits count as a new activity or only new comments?

[prince-chrismc]

Yes any change to a comment (edit, hide, delete) would cause an update.

Those PRs could be labeled appropriately and not be targeted by the bot? If that's an item to comprise on?

I've only seen 2-3 PRs wake up the a thread but never annoy. I am curious how many it's closed 🤔

Two comments in 60 days is far my definition of annoying =) I think your different perspective is noteworthy and drives home the importance of discussion.

[Croydon]

I think we rather have to try to implement a custom logic into @ericLemanissier's bot. Something like:

when was the last activity by a human? If > then X days, stop updating the potential conflict messages -> the stale bot does not see any activity anymore and actually closes the PR

@prince-chrismc Would you be fine with the current number of days, if PRs would not be kept alive by bots?

[prince-chrismc]

Eric's bot already has that logic 🙃I had asked for it a while back and it made a significant difference. ericLemanissier/conan-center-conflicting-prs#7 We are just starting to get close to that x=15 for some of the PRs.
➡️ I do not think this is the majority.
There are some PRs who've been update, but left no trace in the UI, I suspect the Conan team runs tests which also trip up the stale bot. I also might be guilty of that!

I personal feel 30/60 days is far to huge of a time period to wait for nothing to occur. I feel at 30 days the OP long moved on or there's a legitimate reason and it should be labeled.
Roughly 70% of PRs merged are less then a week old. My suspicion is the "No Beta User" is lowering that stat but I've not found a reliably way to measure that.

I think there would be a ⏲️ benefit from reducing the number of PRs we check when far to often they have gone stale.
Out of the 5 pages, usually from 3 and 4 there's maybe 2 or 3 PRs that need attention... and they are not easy to find 🔍 .
Most have comments and are waiting for the Op to apply them.

I think a good case study is uilianries' profile with the "reviewed prs" for this repo at 53, the mast majority is 47XX range and 2-3 in 45XX, the oldest is 4556 opened 21 days ago, but that was updated a few hours ago.

That leaves 2 pages over 50 PRs (excluding the last page that are blocked) in some kind of limbo state. This PR propose to tag half of those are stale so we can focus on the more important ones.

---

To get an actual number I found this https://github.com/UnsafePointer/nisekoi... which gave

$ ~/go/bin/nisekoi calc conan-io/conan-center-index --access-token ***********
Average landing PR time is: 232.66 hours, for a total of 3473 landed PRs

But that's every PR ever not the last 3 months I've been tracking with my bot. I think we've gotten faster.

Compared to the current average age of 74d 3h 27m noted by my bot... Oddly enough the different is 60 days 🤣

[prince-chrismc]

There was a huge burst of activity 03/07-03/08... 8 PRs were refreshed with more than 30 days

[jgsogo]

We have asked for a feature that would help with this... if we were able to ignore comments from users, then... probot/stale#329

[prince-chrismc]

The challenge is that would be a change to the GitHub service...

If you check the query they are using https://github.com/probot/stale/blob/7510fac863900a98057436d308e9fe21bbe314ec/lib/stale.js#L111

There's nothing to filter the commenter