ofborg Failure triage: an option to build on the base branch to compare failures

Consider https://github.com/NixOS/nixpkgs/pull/34676 — I see some failures, I would like to say build-base-branch diffpdf to see if these are new or inherited (OK, on Darwin it is just meta.platforms, but it would be nice to check aarch64-linux).

Feb 06 '18 19:02 7c6f434c

We are rapidly approaching a point where we simply can't contain everything in the github comments anymore :(

Feb 06 '18 21:02 grahamc

Unix way teaches us that every protocol can be embedded into text streams…

Feb 06 '18 22:02 7c6f434c

Perhaps you can help me understand what that means.

Feb 06 '18 22:02 grahamc

In this case, unfortunately, not as much as you might want…

I guess coalescing will be needed soon in any case.

What is the real limitation with comments? It's not like we have something that is impossible to make readable/navigable. Rate-limiting might be a problem even with just auto-building too much, regardless of any additional functionality..

Feb 06 '18 22:02 7c6f434c

In general the problem is with updating comments: updates don't load until the page is refreshed, plus all the caching problems of read-after-write consistency you want in data stores.

Feb 06 '18 22:02 grahamc

Read-after-write probably means that editing the original comment is too risky; editing own comments by replacing them with the new-correct text should be fine from consistency point of view as long as the comment editor process has monotonously increasing view of the builds.

After the last update a ping will create both a notification and a reminder to refresh the page… (alternative: it will contain all the data, and the original comment gets deleted, so refreshing is not needed)

Feb 06 '18 23:02 7c6f434c

What do you mean by a ping? I'm not sure this avenue is going to work :(

Feb 07 '18 23:02 grahamc

A separate comment primarily intended to create a new notification.

Feb 07 '18 23:02 7c6f434c

Then we may as well just post them all as individual comments, no?

Feb 07 '18 23:02 grahamc

Well, you could update the comment with the results as they appear — for those who like to refresh the page — then replace this with a new comment once you have all the results.

Feb 08 '18 07:02 7c6f434c

Re: original topic: maybe Borg could include Hydra links for comparison?

Feb 18 '18 12:02 7c6f434c

… in particular, a fact that ofborg failed to build a dependency with the same derivation as already failed on Hydra could also be useful.

Jan 21 '19 14:01 7c6f434c