On Thu, Sep 12, 2024 at 11:45:47AM +0200, Patrick Steinhardt wrote:
> > It turns out to be an "ar" archive full of *.o files. Perhaps
> > creation of libgit.a was interrupted, then later somebody said
> > "git add .", and the file somehow ended up as a part of Git.
> >
> > I've applied the series after removing the cruft.
>
> I've noticed that something leaves behind this garbage from time to
> time. It's a shame that its name is just random gibberish, because that
> makes it impossible to set up a gitignore rule for it.
I've seen it once or twice, too. It's always st[A-Za-z0-9]{6}, at least
on my system. Doing this:
diff --git a/Makefile b/Makefile
index 91f65d7dc5..b58312edab 100644
--- a/Makefile
+++ b/Makefile
@@ -2860,7 +2860,7 @@ scalar$X: scalar.o GIT-LDFLAGS $(GITLIBS)
$(filter %.o,$^) $(LIBS)
$(LIB_FILE): $(LIB_OBJS)
- $(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^
+ $(QUIET_AR)$(RM) $@ && ($(AR) $(ARFLAGS) $@ $^ & sleep 0.01; kill $$!) && false
$(XDIFF_LIB): $(XDIFF_OBJS)
$(QUIET_AR)$(RM) $@ && $(AR) $(ARFLAGS) $@ $^
and then running "make" several times got me:
Untracked files:
st5AOnzl
st8gTTvD
stJABM4u
stKA7riS
stLZ4yDr
stMWJCVY
stjYXpN0
stwcX3kZ
I don't think we'd want to put anything in the tracked .gitignore, since
it involves system-specific details. But I think that:
/st[0-9a-zA-Z][0-9a-zA-Z][0-9a-zA-Z][0-9a-zA-Z][0-9a-zA-Z][0-9a-zA-Z]
in your .git/info/excludes (or user-level gitignore) would work if you
want, and is probably not too likely to hit false positives (since it
lacks an extension).
I do feel like I have only started seeing it in the last month or so. I
wonder if a new version of binutils changed behavior or something.
-Peff
