Tonight I accidentally did something stupid and committed binaries to my git repository, despite knowing better in the first place. After the commit I tried to push.
I hadn’t even noticed my mistake until I came back to my terminal after 10 minutes and saw
Writing objects: 42% (16/38), 10.02 MiB | 21 KiB/s
Unfortunately simply git rm-ing the files is no good, because they’re
in the object store at this point, which is essentially git’s historical
record— its database of versions. I need to go back through the history
and remove all traces of the binaries in order to shrink the object
store back down to a reasonable size.
Git’s filter-branch command is designed to do this. It’s pretty complicated and not particularly well documented, but if I do this…
git filter-branch --index-filter 'git rm -r --cached --ignore-unmatch <big-binary-mess/>'
it obligingly hoovers up my big binary mess for me. Then I can either add
an exclusion in .gitignore, or git rm the offending blobs.
Done.