From: Johannes Schindelin <Johannes.Schindelin@gmx.de>
To: Junio C Hamano <gitster@pobox.com>
Cc: Luciano Joublanc <ljoublanc@dinogroup.eu>, git@vger.kernel.org
Subject: Re: Bad refspec messes up bundle.
Date: Fri, 30 Mar 2018 12:20:21 +0200 (DST) [thread overview]
Message-ID: <nycvar.QRO.7.76.6.1803301102430.5026@qfpub.tvgsbejvaqbjf.bet> (raw)
In-Reply-To: <xmqq370wvugk.fsf@gitster-ct.c.googlers.com>
Hi Luciano,
On Mon, 19 Mar 2018, Junio C Hamano wrote:
> Luciano Joublanc <ljoublanc@dinogroup.eu> writes:
>
> > Yesterday I created a git bundle as best as I can remember like this
> >
> > git bundle save chunk chunk.bundle --all master
> >
> > Note the 'master' I added accidentally at the end - this was a user
> > error but still the bundle was created.
> >
> > When I tried to clone this, I get
> >
> > ~\local\src> git clone 'G:\My Drive\chunk.bundle' fs2-columns
> > Cloning into 'fs2-columns'...
> > Receiving objects: 100% (31/31), done.
> > Resolving deltas: 100% (5/5), done.
> > fatal: multiple updates for ref 'refs/remotes/origin/master' not allowed.
> > ~\local\src> git bundle verify chunk.bundle
> > The bundle contains these 3 refs:
> > 3c804437a5f8537db1bfb5d09b7bff4f9950605e refs/heads/master
> > 3c804437a5f8537db1bfb5d09b7bff4f9950605e HEAD
> > 3c804437a5f8537db1bfb5d09b7bff4f9950605e refs/heads/master
> > The bundle records a complete history.
> > chunk.bundle is okay
> >
> > After trying a couple of things, I finally managed to clone it using
> >
> > ~\local\src> git clone -b master --single-branch .\chunk.bundle fs2-columns
> >
> > i.e. the '--single-branch' option saved me.
> >
> > Is this a bug? Should bundle allow providing multiple refspecs when
> > `--all` is provided? I admit this was clearly a case of 'caveat
> > emptor', but shouldn't this be disallowed (i.e. is there any situation
> > when this is useful?)
>
> Thanks for a report.
>
> Just like a remote repository that reports the same ref more than
> once in its initial advertisement (i.e. "git ls-remote $remote"
> gives duplicate entries), a bundle file that records the same ref
> more than once *is* a bug, I would think.
>
> A "git bundle create" command that creates such a bundle file
> shouldn't. It is not very useful to diagnose it as an error; it
> probably makes more sense to dedup the refs instead when writing the
> bundle file. Of course, we should abort with an error *if* the code
> ever tries to store the same ref twice with different object name
> (i.e. attempt to dedup, in vain).
>
> Also, "git clone" from such a bundle file (or for that matter, a
> remote repository that advertises the same ref twice) probably
> should do a similar deduping, with a warning message.
I agree that it is a bug if a bundle file records a ref multiple times.
Luciano, here are some pointers so you can fix it:
- probably the best way to start would be to add a new test case to
t/t5607-clone-bundle.sh. The script *should* be relatively easy to
understand and imitate. The new test case would probably look somewhat
like this:
test_expect_failure 'bundles must not contain multiple refs' '
git bundle create all.bundle --all master &&
grep master all.bundle >master.lines &&
test_line_count = 1 master.lines
'
- then, ensure that this test passes and reports this correctly as a known
breakage.
- now it is time to fix the actual bug. The code in question is the
write_bundle_refs() function in bundle.c (careful, there are two files
of that name in Git's source code, one in the builtin/ subdirectory, the
other one in the toplevel directory, you will want to look at the
latter one).
- notice that there is already a construct in the loop over the pending
refs where some are skipped:
if (e->item->flags & UNINTERESTING)
continue;
The "uninteresting" refs would be those on the excluded end of a commit
range, e.g. if you called `git bundle create x.bundle master..next`, the
`master` ref would be such an "uninteresting" ref.
- one might be tempted to introduce another flag and set it once a ref was
shown and skip a ref for which that flag is set. And indeed, this was my
own first thought!
However, this would be incorrect, as the flags are stored with the
*commit*, not with the ref. So if two refs point to the same commit,
that new code would skip the second one by mistake!
By the way, this makes me think that there is another very real bug in
the bundle code, in the part I showed above. Suppose you have a `master`
and a `next` ref, and both point at the same commit, then you would want
`git bundle create next.bundle master..next` to list `next`, don't you
think? This one would be a lot harder to fix, though, as the
setup_revisions() function really only sets the flag on the respective
commits because it wants to set things up for revision walking, not
necessarily to list "interesting" refs. A possible solution might be to
abuse the `mode` field in the object_array (which is the type of the
`pending` field in the `rev_info` struct where the parsed command-line
parameters are held, when that `mode` makes only sense for blobs and
we store refs pointing to commits or tags).
That's a side-track, though.
- most likely, the best way to avoid duplicate refs entries is to use the
actual ref name and put it into a hash set. Luckily, we do have code
for this, and examples how to use it, too. See e.g. fc65b00da7e
(merge-recursive: change current file dir string_lists to hashmap,
2017-09-07). So you would define something like
struct ref_name_entry {
struct hashmap_entry e; /* must be first field */
char ref_name[FLEX_ARRAY];
}
and implement the cmp function similar to the path_hashmap_cmp() in
merge-recursive.c, then initialize a hashset in write_bundle_refs(),
add handled ref names via
FLEX_ALLOC_STR(entry, ref_name, e->name);
hashmap_entry_init(entry, strhash(e->name));
hashmap_add(&shown_refs, entry);
and look up whether a ref should be skipped via
if (hashmap_get_from_hash(&shown_refs, strhash(e->name), e->name))
continue;
Oh, and before returning from write_bundle_refs(), the hashset
(including its entries) should be released via
hashmap_free(&shown_refs, 1);
These hints should get you started on this project.
Looking forward to your contribution,
Johannes
next prev parent reply other threads:[~2018-03-30 10:20 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-03-19 8:39 Bad refspec messes up bundle Luciano Joublanc
2018-03-19 17:36 ` Junio C Hamano
2018-03-30 10:20 ` Johannes Schindelin [this message]
2018-03-30 17:18 ` Junio C Hamano
2018-03-30 18:58 ` Johannes Schindelin
2018-03-31 8:50 ` Luciano Joublanc
2018-04-03 14:38 ` Johannes Schindelin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
List information: http://vger.kernel.org/majordomo-info.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=nycvar.QRO.7.76.6.1803301102430.5026@qfpub.tvgsbejvaqbjf.bet \
--to=johannes.schindelin@gmx.de \
--cc=git@vger.kernel.org \
--cc=gitster@pobox.com \
--cc=ljoublanc@dinogroup.eu \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this public inbox
https://80x24.org/mirrors/git.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).