On 2014-06-13 01.30, David Turner wrote:
> On Thu, 2014-06-12 at 12:47 -0700, Junio C Hamano wrote:
>> David Turner <dtur...@twopensource.com> writes:
>>> This issue bit us again recently.
>>> In talking with some colleagues, I realized that the previous version
>>> of this patch, in addition to being potentially slow, was incomplete.
>>> Specifically, it didn't handle the case of refs/heads/case/one vs
>>> refs/heads/CASE/two; these are case clones even though they strcasecmp
>>> different.
>> Good catch to realize that two refs that share leading paths that
>> are the same except for cases are also problematic, but that makes
>> this feature even less about "case clones", doesn't it?
> I agree: word "clone" is less good now.  Maybe "case conflicts"?
>> Also it somehow feels that the patch attempts to solve the issue at
>> a wrong level.  On a platform that cannot represent two refs like
>> these (e.g. trying to create "FOO" when "foo" already exists, or
>> trying to create "a/c" when "A/b" already exists---ending up with
>> "A/c" instead, which is not what the user wanted to create), would
>> it be more sensible to fail the ref creation without touching the
>> users of ref API such as receive-pack?  That way, you would also
>> catch other uses of refs that are not supported on your system,
>> e.g. "git branch a/c" when there already is a branch called "A/b",
>> no?
> So we would change is_refname_available?  And to do this, we would
> change the ref_dir functions to take case into account?
> This might be somewhat complicated because we could be starting from a
> repo which already has case conflicts.  What should we do about that?  I
> think this is even possible on a case-insensitive filesystem with
> packed-refs, although I have not checked. 
It is:

user@mac:~/test> git init
Initialized empty Git repository in /Users/tb/test/.git/
user@mac:~/test> git checkout -b Branch
Switched to a new branch 'Branch'

user@mac:~/test> echo a>a && git add a && git commit -m "Add a"
[Branch (root-commit) e27e99c] Add a
 1 file changed, 1 insertion(+)
 create mode 100644 a

user@mac:~/test> ls .git/refs/heads/

user@mac:~/test> git pack-refs --all

user@mac:~/test> ls .git/refs/heads/ 

user@mac:~/test> git checkout -b BRANCH

Switched to a new branch 'BRANCH'

user@mac:~/test> ls .git/refs/heads/
user@mac:~/test> git branch

> The simplest idea is probably
> to pretend that the first conflicting refname component we find is the
> one true one, and reject new refs containing versions which are case
> conflicting with this one until the user cleans up their repo.  In other
> words, if the user has A/b and a/c already, and we find A/b first, then
> we reject a/d but allow A/d.   This is arbitrary, but workable.  We
> could warn about this situation when we load up the refs, too.  
> Does this match what you are suggesting? 
> If so, I think it is possible, and if I don't hear anything back from
> the other ref folks, I'll see if I have time to implement it.

I can see 2 ways "forward": Either to try to avoid case insensitve refs at all.
If you do that carefully in a project, it may work with the patches you suggest.

Or try to have a functionality to always use packed refs, and have a 
for it:

The advantage can be that branch names like "Branch" and "BRANCH" can live 
in a project, regardless if you have a case sensitive or insensitve file system.
This assumes that e.g. "core.packedrefs" is default true whenever 
is true, which we can easily do in init_db.c

Another advantage with the "always packed refs" is that you can have branches
bugfix and bugfix/next-bug side by side.
Today when you create a branch bugfix theh you have a file bugfix,
and can not create a directory called bugfix/ to store the file bugfix/next-bug.

I don't know if this is scaling well for many refs, some reports say it does 
But I can imagine to mmap .git/packed-refs into memory, throw every line
into a hash table and do a lazy check if a ref is "good" whenever it is needed.
Or do it in git fsck ?

The disadvantege with packed refs is that you need to carefully lock one central
file, and that may be a bottleneck on a server.

I don't have that many refs in my projects, so I only can test on dummy repos,
more or less close to reality.

To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to