Tickets should not close just because they've been around for two weeks

[lee-b]

This issue was closed because it has been inactive for 14 days since being marked as stale.

Originally posted by @github-actions in #7391

The way that these tickets are currently being handled is essentially ignoring issues. It's a logically invalid way to handle tickets.

Tickets should close when they've been reviewed and been found to be no longer relevant (logically, due to code changes), or invalid for some reason, such as insufficient information provided. AI can help with such reviews, but an automated timer that just closes stuff after 14 days is... commonplace, but wild.

Consider as well, that even simply allowing these to build up, when there are no resources to deal with them, preserves knowledge for a time when AI can deal with them easily (which may be right now, in many cases).

[am17an]

There is a github job which closes issues,
https://github.com/ggml-org/llama.cpp/blob/945501f5ea4b8ca56b181ecb035e9ee3fb31f432/.github/workflows/close-issue.yml

As you can see there are plenty of exceptions to this rule, and you can always re-open if you think it's worthwhile for the developers to see. There are many low-effort issues which this rule aims to close.

[0cc4m]

It's also not 14 days, but 30 days until it is marked stale, and then another 14 days until it is closed, unless someone intervenes. It's not difficult to keep an issue active if it is important enough.

[IMbackK]

This issue was closed because it has been inactive

[IMbackK]

just kidding, but i also think that the current policy is fine.

[henk717]

Current policy has an issue for sure, using close bots isn't the biggest issue. The big issue is the lack of warnings and the lack of reopening. People patiently wait for a contributor to pick up the issue, it gets closed and then they reply that its still relevant only for it to be never reopened. This is especially an issue for feature requests.

Until this threat I wasnt aware of the stale grace period, does the bot comment about this? Because when an issue / feature request gets submitted its reasonable to assume the person submitting isn't going to constantly open the request and may just rely on email comment notifications. If there is no email triggering them to reply its already to late when the bot comments that it has been closed.

[henk717]

Here is an example of where it went wrong : #9086

The lack of this makes llamacpp loose to backends that have this on multi GPU setups. Its a year old and still relevant. It has seen multiple comments including one of my own that this is still a relevant issue for llamacpp that makes it uninteresting for some.use cases.

But due to it being closed nobody looks at the issue and due to the bot not warning in advance it was never bumped in time. The bot provided us no way to get it reopened.

[am17an]

@henk717 I re-opened that issue and added a research label so that it doesn't get closed. Also, llama.cpp is an open-source project, you are welcome to PR any feature you would like to see in mainline

[0cc4m]

For feature requests, it's less important. This is an open source project, the only way to ensure a feature gets implemented is to implement it yourself. Most of us don't go through issues and look for ideas, we work on what interests us.

Whether an issue disappears out of lack of interest on pages 2 and onwards of the issues list, or they get eventually closed automatically, isn't a big difference. They disappear either way.

Going through issues and marking relevant ones with a tag that prevents them from going stale would be good, but it's a chore.

[JohannesGaessler]

The lack of this makes llamacpp loose to backends that have this on multi GPU setups. Its a year old and still relevant. It has seen multiple comments including one of my own that this is still a relevant issue for llamacpp that makes it uninteresting for some.use cases.

Let me be frank: I don't decide what to work on based on feature requests, especially old ones. The reason that feature is not implemented is not a lack of awareness but lack of developers working on it. Originally I was going to work on this in August but then multiple people told me that they wanted to work on it without bringing it to completion. As it is I'm not going to delegate it to someone else again and my goal is to have something usable by the end of the year.