Potential usability improvements

[pflanze]

• Working directory pool:

  • do not rename dirs on failure, save a context file,
  • offer 'enter' subcommand (that sets env vars (and a run run subcommand for running there?)),
  • and perhaps a command to mark it for use again?

• Save context information how a job got inserted (by branch name, poll, manually), to show that information in the listing? Bare commit ids aren't informative.

• More consistent logging and error reporting

• Signalling when jobs are done or errors happened via Slack?

[pflanze]

"Save context information how a job got inserted (by branch name, poll, manually), to show that information in the listing? Bare commit ids aren't informative.": implemented in fceb474 and 1f961df, shown in "list" and saved as "reason.ron" files

[pflanze]

Also see benchmarking: prevent long-running jobs from blocking high-priority jobs

[pflanze]

"More consistent logging and error reporting" is done for a good part; maybe redirecting the output from evobench-evaluator runs into the log file would still be good. Proper dynamic error scopes (nesting) would be nice but not required.

[pflanze]

" Working directory pool: do not rename dirs on failure, save a context file, offer 'enter' subcommand (that sets env vars (and a run run subcommand for running there?)), and perhaps a command to mark it for use again?":

The "do not rename" and "save a context file" instead parts are done. Still no command to do the latter, no urgent need.

[pflanze]

Signalling to Slack doesn't actually appear useful, we just look at the list watch display instead, and we start and have automatically started jobs in too high numbers to deal with Slack messaging.