On Wed, Aug 5, 2020 at 6:54 PM Kyle Meyer <kyle@xxxxxxxxxx> wrote: > > christian w writes: > > > Sorry for the insufficient reproduction steps. > > > > I managed to figure out it has something to do with git repos within > > git repos. It happened to us because of go packages that are inside > > our repo. These are the reproduction steps that work for me > > consistently using Go version go1.14.4 darwin/amd64. This is the repo > > that gets cloned into src/golang.org/x/debug by the go get command: > > https://github.com/golang/debug/tree/66ec140f2f72d15dc6133502edd2bb7238b1740c > > > > $ mkdir testdir > > $ cd testdir > > $ git init . > > $ GOPATH=$(pwd) go get -u golang.org/x/debug/ || true > > $ git ls-files --others '*.txt'# > > src/golang.org/x/debug/ > > Okay, removing Go from the equation: > > $ # in fresh repo > $ mkdir a > $ git init a/d0 > $ touch a/f0.txt > $ git ls-files --others 'a/*.txt' > a/d0/ > a/f0.txt > > It looks like the spurious a/d0/ entry doesn't show up until v2.27.0, in > particular 95c11ecc73 (Fix error-prone fill_directory() API; make it > only return matches, 2020-04-01). Adding the author to the Cc. (Sorry, > Elijah, I recall your frustration with dir.c from a previous ls-files > issue that I reported [1]). Thanks for digging in and coming up with a smaller list of steps to reproduce. You are right that I dread reports that touch dir.c -- if I would have just ignored the first report back in March of 2018, there's a really, really long list of problems I could have avoided... I'll try to take a look in the next week. > > [1] https://lore.kernel.org/git/CABPp-BFG3FkTkC=L1v97LUksndkOmCN8ZhNJh5eoNdquE7v9DA@xxxxxxxxxxxxxx/
