git-rev-list: --objects list inconsistency

git-rev-list: --objects list inconsistency

[Johannes Sixt]

I observed a puzzling behavior of git-rev-list:

$ git-rev-list master^^..master 
f3364ca9405e17772fecea1d06c40b9965b8e6e4
bb3bfda219a85d2d49e62c261b9c8f6795ebdc22
$ git-rev-list --max-count=2 master 
f3364ca9405e17772fecea1d06c40b9965b8e6e4
bb3bfda219a85d2d49e62c261b9c8f6795ebdc22
$ git-rev-list --objects master^^..master |wc -l
     10
$ git-rev-list --objects --max-count=2 master |wc -l
   2376

As you can see, without --objects, master^^..master and --max-count=2
list the same two commits.

But with --objects I get different object lists. I don't even know which
one is the one to expect, but I certainly would not have expected the
lists to be different. What's wrong here?

BTW, the objects listed with --objects --max-count=2 are basically the
entire tree at master plus the few objects that changed since master^^.

$ git --version
git version 1.4.4.rc2.gc8641

-- Hannes

Re: git-rev-list: --objects list inconsistency

[Junio C Hamano]

Johannes Sixt <J.Sixt@eudaptics.com> writes:

> I observed a puzzling behavior of git-rev-list:
>
> $ git-rev-list master^^..master 
> f3364ca9405e17772fecea1d06c40b9965b8e6e4
> bb3bfda219a85d2d49e62c261b9c8f6795ebdc22
> $ git-rev-list --max-count=2 master 
> f3364ca9405e17772fecea1d06c40b9965b8e6e4
> bb3bfda219a85d2d49e62c261b9c8f6795ebdc22
> $ git-rev-list --objects master^^..master |wc -l
>      10

You are asking to get objects reachable from master but exclude
objects reachable from master^^.

> $ git-rev-list --objects --max-count=2 master |wc -l
>    2376

You are asking to get objects reachable from master and master^.

You have 2300+ objects that are common in trees of master,
master^ and master^^.

Re: git-rev-list: --objects list inconsistency

[Linus Torvalds]

On Tue, 14 Nov 2006, Johannes Sixt wrote:
> 
> As you can see, without --objects, master^^..master and --max-count=2
> list the same two commits.

"--max-count=X" and "master^^..master" are two TOTALLY different things.

The fact that they happened to give the same results without "--objects" 
is a random thing, and not at all guaranteed. For example, if "master" (or 
its parent) had been a merge, it probably wouldn't have given the same 
result even _without_ "--objects".

> But with --objects I get different object lists. I don't even know which
> one is the one to expect, but I certainly would not have expected the
> lists to be different. What's wrong here?

Nothing is wrong. You're asking for two totally different things.

With "--max-count=2", you're saying "I want the first two commits", and 
then the "--objects" part extends that to all the other object types too.

So you get two commits _and_ every single object reachable from those two 
commits.

In contrast, with "master^^..master", you're saying "I want everything 
that is reachable from "master", but _not_ reachable from it's 
grand-parent. And the "--objects" flag extends that to all the other 
object types too.

So you get (in this case) two commits _and_ all the objects that are 
reachable from those two commits but that are _not_ reachable from the 
grandparent.

That's a totally different thing, obviously. There's likely to be a lot of 
objects in common between "master" and its grandparent, and you literally 
asked to not be shown those objects. In the first case, you didn't ask for 
the exclusion of the grandparent.

So you should expect a big difference in almost all cases. In one case, 
you ask for an exclusion, in the other case, you just say "just the first 
two commits, please". Not at all equivalent.