FreeBSD ports frozen after someone commits the entire 150MB Linux Copilot binary

36 points by raymii

reidrac

FreeBSD ports frozen after someone commits the entire 150MB Linux Copilot binary

That's the title of the post I'm looking at right now.

I don't know if we can read too much into it (it could have been about any piece of software). But given that it's Copilot, an AI tool, it feels very fitting. Someone thought it was perfectly acceptable to commit a 150MB binary blob with murky copyright status, and it is AI related.

gamache

I read this completely differently. A human goofed. The team reversed the mistake, and is taking a moment to make the system more resilient to that. It is a boring, 1990's-class story and it warms my heart.

Did you see the commit? Adding the binary blob sure looks accidental to me.
- reidrac
  
  EDIT: You're right, looks like an honest mistake.
  
  It was me reading the commit wrong.
- david_chisnall
  I can't work out how to get to the commit message from that UI, but it looks like it was 'misc/github-copilot-cli: update 1.0.71 → 1.0.72', so there wasn't anything to suggest this was deliberate.
  
  If you're working on multiple ports, I'd imagine you'd do git commit *, but you might have muscle memory for doing git add * first to make sure you didn't accidentally forget to commit a file (which would break the build for anyone else). If you're testing with Poudriere (as you should be), each port is built in a jail and so it's quite common to leave files you're working with in the ports tree so they're easy to find poudriere testport -i or similar (build the port, drop into a shell for testing in the jail where it's built).
  
  There are some process failures that really need addressing:
  
  It doesn't look like there was a Phabricator review associated with it. Ports committers often just push to the tree directly and that's not ideal.
  
  There's no git receive hook thingy to check for accidentally committing large files. A port should probably not contain any files more than a few tens of KiBs in size, a 100+ MiB file is a big error that should have been rejected by automation.
  
  There's no pre-merge poudriere testport or similar in the ports tree.
  
  Postmortem should focus on those, not blaming copilot (much as I would love to blame copilot).
  - reidrac
    
    It was indeed:
    
    update 1.0.71 → 1.0.72
    
    The next commit was:
    
    Remove inadvertently added file
  - quasi_qua_quasi
    
    I've gotten bit by similar things when hopping around between branches in jj, which autostages files: if branch A produces adds something to gitignore, then you switch to B, it'll stage the A file because it's not ignored.
    
    (It does noisily let you know about files >1MB and refuse to stage them, but I've gotten hit by smaller artifacts a couple times.)
- nemin
  
  I'm somewhat surprised they haven't had a hard limit in the first place. One would assume a binary asset above the size of [some reasonably large amount of megabytes] could only be the result of a mistake.
  
  But it's good to see that instead of treating just the symptom, they are stopping the show and figuring out something more robust.