git blame --reverse: it's all a bit mysterious

The --reverse option to git blame (which isn't listed in the option summary when you type just "git blame", but *is* listed in the cross-referenced git-rev-list(1), though that's not terribly helpful) doesn't seem to understand what I'm trying to ask, and the man page is not much help.

greenrd@cspcnh /var/db/paludis/repositories/arbor $ strace -f -e trace=file git blame --reverse 48d96ce0e486eedea9fda0f8e480ba3c6caffc90..HEAD exlibs/eutils.exlib execve("/usr/bin/git", ["git", "blame", "--reverse", "48d96ce0e486eedea9fda0f8e480ba3c"..., "exlibs/eutils.exlib"], [/* 41 vars */]) = 0 access("/etc/ld.so.preload", R_OK) = -1 ENOENT (No such file or directory) open("/etc/ld.so.cache", O_RDONLY) = 3 open("/usr/lib/libz.so.1", O_RDONLY) = 3 open("/usr/lib/libcrypto.so.0.9.8", O_RDONLY) = 3 open("/lib/libpthread.so.0", O_RDONLY) = 3 open("/lib/libc.so.6", O_RDONLY) = 3 open("/lib/libdl.so.2", O_RDONLY) = 3 getcwd("/var/db/paludis/repositories/arbor", 4096) = 35 stat64(".", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 stat64(".git", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 access(".git/objects", X_OK) = 0 access(".git/refs", X_OK) = 0 lstat64(".git/HEAD", {st_mode=S_IFREG|0644, st_size=23, ...}) = 0 open(".git/HEAD", O_RDONLY|O_LARGEFILE) = 3 access("/etc/gitconfig", R_OK) = 0 open("/etc/gitconfig", O_RDONLY|O_LARGEFILE) = 3 access("/home/greenrd/.gitconfig", R_OK) = 0 open("/home/greenrd/.gitconfig", O_RDONLY|O_LARGEFILE) = 3 stat64(".git", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 access(".git/config", R_OK) = 0 open(".git/config", O_RDONLY|O_LARGEFILE) = 3 access("/etc/gitconfig", R_OK) = 0 open("/etc/gitconfig", O_RDONLY|O_LARGEFILE) = 3 access("/home/greenrd/.gitconfig", R_OK) = 0 open("/home/greenrd/.gitconfig", O_RDONLY|O_LARGEFILE) = 3 access(".git/config", R_OK) = 0 open(".git/config", O_RDONLY|O_LARGEFILE) = 3 access("/etc/gitconfig", R_OK) = 0 open("/etc/gitconfig", O_RDONLY|O_LARGEFILE) = 3 access("/home/greenrd/.gitconfig", R_OK) = 0 open("/home/greenrd/.gitconfig", O_RDONLY|O_LARGEFILE) = 3 access(".git/config", R_OK) = 0 open(".git/config", O_RDONLY|O_LARGEFILE) = 3 lstat64("exlibs/eutils.exlib", 0xbf896768) = -1 ENOENT (No such file or directory) stat64(".git", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 getcwd("/var/db/paludis/repositories/arbor", 1024) = 35 chdir(".git") = 0 getcwd("/var/db/paludis/repositories/arbor/.git", 4096) = 40 lstat64("/var/db/paludis/repositories/arbor/.git", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0 chdir("/var/db/paludis/repositories/arbor") = 0 chdir("/var/db/paludis/repositories/arbor") = 0 lstat64("48d96ce0e486eedea9fda0f8e480ba3c6caffc90..HEAD", 0xbf896768) = -1 ENOENT (No such file or directory) fatal: cannot stat path '48d96ce0e486eedea9fda0f8e480ba3c6caffc90..HEAD': No such file or directory

Now this error message is very unhelpful. Why is git blame trying to interpret this as a filename, when the man page implies that it expects a revision range (or possibly a single revision, if you believe the option summary at the top of the man page)?

This isn't the first time I've noticed git misinterpreting a revision (range) as a filename, but in this case, I have no clue what is going on.

I notice from the strace output that it's looking for the specified filename first, which of course no longer exists in HEAD. The man page implies that blame --reverse only needs the file to exist in the first revision of the revision range; if it also needs to exist in the *working tree* or something like that, that should be documented in the man page.

I'm using git version 1.7.3.2.

Robin Green wrote:

[...]

This can be fixed by saying

  git blame --reverse 48d96ce0e486eedea9fda0f8e480ba3c6caffc90..HEAD -- exlibs/eutils.exlib

i.e., you need to supply the -- to explicitly tell it that the stuff after it is a filename.

It's probably a code (not documentation) bug since blame takes only a single revision range anyway, so everything after that should be assumed to be a filename, but maybe I'm missing something.