Re: bisect: "Needed a single revision" message is confusing

Daniel Bonniot venit, vidit, dixit 02.11.2012 14:23:

Hi,

Suppose I'm doing a git bisect, say:

$ git bisect good 8c7a786b6c8eae8eac91083cdc9a6e337bc133b0

That works fine. The sha1 could also be a substring, as long as it's
unambiguous, e.g.:

$ git bisect good 8c7a786b6c

Now if it's ambiguous, I get an error message:

$ git bisect good 8
fatal: Needed a single revision
Bad rev input: 8

All fine and good. But what if I somehow input a non-existing sha1 (in
my case see [1]), e,g:

$ git bisect good 8c7a786b6c8eae8eac91083cdc9a6e337bc133b1
fatal: Needed a single revision
Bad rev input: 8c7a786b6c8eae8eac91083cdc9a6e337bc133b1

I understand that technically both "no revision" and "multiple
revisions" qualify as "not a single revision", but they correspond to
quite different situations. I think it would be more helpful to get
different error messages, something like:

Bad rev input: 8 refers to multiple revisions
Bad rev input: 8c7a786b6c8eae8eac91083cdc9a6e337bc133b1 does not refer
to a valid revision

(and avoid outputing the "fatal: Needed a single revision" message).

Is this a good idea? Anybody can think of better error messages?

I'm not familiar with the code base at all, but I could give a try at
implementing it, unless it's trivial enough that someone does it
earlier. After a quick look, it looks like either git-bisect itself or
rev-parse would need to be touched, any pointers and hints welcome.

Cheers,

Daniel

[1] if you want to know, I got a sha1 from one repository and used it
in another, which probably should work, except that when using git-svn
they don't. And the "single revision" error message lead me on a
tangent.

The error comes from rev-parse, which is called by bisect. The problem
is that

git rev-parse deadbeef
git rev-parse --verify deadbeef

give two very different error messages (if there's no dead beef there),
and that "git ref-parse --verify" gives the same error message for
non-existing as for ambiguous revs.

There are 3 places in builtins/rev-parse.c which call
die_no_single_rev(), and at least some of them should probably choose
the error message more carefully.

Cheers,
Michael