Linus Torvalds [off-list ref] writes:
If you do a
git pull linus v3.4-rc5
in order to just update to the state of my latest tag, then git will
assume you want to do a new commit (and thus a non-fast-forward) just
so that git can record the tag signature in the commit.
The sad part is that I don't think you can even override the second
case.
...
That's inconvenient, and an unintended consequence of the behavior I
wanted as a top-level maintainer. But I really do think it's wrong for
normal developers who might validly just want to update to some
particular tagged release.
Junio? Any ideas?
"Ideas" meaning "recipe to do with deployed binaries"?
When a normal developer wants to _reset to_ a particular tagged release,
in order to _start_ new work, she wouldn't be doing even the above "git
pull linus v3.4-rc5". That will contaminate the result with whatever
random stuff she happened to have on the current branch. A more natural
sequence would be "git fetch --tags linus" followed by either
git checkout v3.4-rc5 ;# to detach
or
git checkout -b mywork v3.4-rc5 ;# to start
So the case to "reset to" is not very interesting.
But when a normal developer wants to _sync to_ a particular tagged
release, in order to _continue_ working on her topic, she would need to
have a merge (unless she does not have _anything_ herself), and at that
point, merging v3.4-rc5 vs v3.4-rc5^0 would not make that much of a
difference. If she absolutely detests the "mergetag" header, she could do
a "git fetch --tags linus" followed by
git merge v3.4-rc5^0
which admittedly is two more letters than she used to type.
If you mean by "Ideas" for additional features, obviously the last step
could be enhanced to use a more intuitive command line that requires the
user to type even more, i.e.
git merge --ff v3.4-rc5
Once that is done, "git pull --ff linus v3.4-rc5" would fall out as a
logical consequence.
But obviously these two would need new code ;-)
On Tue, 2012-05-01 at 20:49 -0700, Junio C Hamano wrote:
Linus Torvalds [off-list ref] writes:
When a normal developer wants to _reset to_ a particular tagged release,
in order to _start_ new work, she wouldn't be doing even the above "git
pull linus v3.4-rc5". That will contaminate the result with whatever
random stuff she happened to have on the current branch. A more natural
sequence would be "git fetch --tags linus" followed by either
git checkout v3.4-rc5 ;# to detach
The problem is, I like to know what has been pulled into mainline. I
have patches in quilt (for ktest only, not for my other work) and will
start adding them on a "clean" release. By doing a git pull (or fetch
and merge), I like to see the fast forward to know if everything that's
in my current branch has been pulled. If it hasn't, then something may
have been missed.
or
git checkout -b mywork v3.4-rc5 ;# to start
But then I would end up with several branches that would require
deleting. One way I could see myself in handling this case would be to
delete the current branch and start again (thinking that everything was
already pulled). But by doing that, if something wasn't pulled in, then
I would have lost those changes without ever knowing.
So the case to "reset to" is not very interesting.
But when a normal developer wants to _sync to_ a particular tagged
release, in order to _continue_ working on her topic, she would need to
have a merge (unless she does not have _anything_ herself), and at that
point, merging v3.4-rc5 vs v3.4-rc5^0 would not make that much of a
difference. If she absolutely detests the "mergetag" header, she could do
a "git fetch --tags linus" followed by
git merge v3.4-rc5^0
which admittedly is two more letters than she used to type.
This would fit into my workflow. Thus I could use this.
If you mean by "Ideas" for additional features, obviously the last step
could be enhanced to use a more intuitive command line that requires the
user to type even more, i.e.
git merge --ff v3.4-rc5
Once that is done, "git pull --ff linus v3.4-rc5" would fall out as a
logical consequence.
But obviously these two would need new code ;-)
The -ff would make sense as it seems to be the logical thing a user
would want. If they specify the fast-forward flag, then the user would
expect the merge to be a fast forward if possible.
BTW, is there a git compare type option. That is, I like to compare two
separate branches with the result that one currently gets with git when
a branch is following another branch. When you check out that branch, it
gives you an update on how those two branches are related (is one a fast
forward of the other, are they off by different commits?). It would be
nice if git could do this with any two branches. I wrote a script to do
this for me (attached) but it would be nice if git had it natively.
$ git-branch-status v3.0.4 v3.0.5
Branch v3.0.4 can be fast forward to v3.0.5 in 240 commits
$ git-branch-status v3.0.4 v3.1
Branch v3.0.4 and v3.1
differ by 257 and 9380 commit(s) respectively
-- Steve