From: Michael J Gruber <git@grubix.eu>
To: "Stefan Beller" <sbeller@google.com>,
"Kévin Le Gouguec" <kevin.legouguec@gmail.com>
Cc: "git@vger.kernel.org" <git@vger.kernel.org>
Subject: Re: git describe and "the smallest number of commits possible"
Date: Tue, 29 Aug 2017 15:34:09 +0200 [thread overview]
Message-ID: <f90ddec3-1bef-0b2f-b98b-e65247655b8a@grubix.eu> (raw)
In-Reply-To: <CAGZ79kZSmJBnfcU=WWmq3VEHJBJLQJzzri07hSU-uA-VgT0t3Q@mail.gmail.com>
Stefan Beller venit, vidit, dixit 28.08.2017 20:24:
> On Sat, Aug 26, 2017 at 7:47 AM, Kévin Le Gouguec
> <kevin.legouguec@gmail.com> wrote:
>> Hi,
>>
>> I've asked this question on the git-users Google Groups list[1], and
>> while the answers there were interesting, I still cannot figure
>> whether my problem comes from an actual bug, a misleading manpage, my
>> inability to understand the code and/or the manpage, or a combination
>> of the three.
>>
>> I noticed this problem on 2.1.4 (Debian oldstable); I can reproduce it
>> on next (7ef03bb281b2220d0f2414365e4949beb2337042). Quoting
>> git-describe(1)'s SEARCH STRATEGY section:
>>
>>> If multiple tags were found during the walk then the tag which has
>>> the fewest commits different from the input commit-ish will be
>>> selected and output. Here fewest commits different is defined as the
>>> number of commits which would be shown by `git log tag..input` will
>>> be the smallest number of commits possible.
>
> Maybe relevant
> https://public-inbox.org/git/20160421113004.GA3140@aepfle.de/
> (specifically the discussion after this patch, it sheds light on the
> heuristics used, though your case here doesn't use these heuristics)
git describe is driving my crazy sometimes, too - both the results and
the code ;)
[long read with lot's of debug output follows, sorry]
What is going on here is the following:
git describe --tags decides based on "depth".
git tag --merged dcc3ef3ee7 lists both emacs-25.1 and emac-25.2 (among
many others).
git rev-list --count emacs-25.1..dcc3ef3ee7
5126
git rev-list --count emacs-25.2..dcc3ef3ee7
4793
So, you would rightly expect 25.2 to be preferred over 25.1.
Now, "git describe" does not do a "rev-list --count" for each tag but,
instead, walks back from dcc3ef3ee7 and notes when it reaches a tagged
commit, and how many steps it takes to get there. The information
"reachable by the i-th tag that I encountered" is kept by a field of
width 27, which limits the --candidates values. Once the walk passes
more tags it is aborted.
Then the possible matches are sorted.
Then the depth calculation is finished.
Funny order of actions, isn't it? Smells like BUG.
At least it makes the debug output meaningless.
Here's the debug output before and after finish_depth_calculation():
searching to describe HEAD
lightweight 4069 emacs-25.1
lightweight 4086 emacs-25.1-rc2
lightweight 4126 emacs-25.1-rc1
annotated 4185 emacs-25.2
annotated 4220 emacs-25.2-rc2
lightweight 4226 emacs-25.0.95
annotated 4236 emacs-25.2-rc1
annotated 4280 emacs-25.1.91
lightweight 4305 emacs-25.0.94
lightweight 4329 emacs-25.1.90
traversed 4462 commits
more than 10 tags found; listed 10 most recent
gave up search at 5c587fdff164e8b90beb47f6da64b4884290e40a
[finish_depth_calculation()]
lightweight 129847 emacs-25.1
lightweight 4086 emacs-25.1-rc2
lightweight 4126 emacs-25.1-rc1
annotated 4185 emacs-25.2
annotated 4220 emacs-25.2-rc2
lightweight 4226 emacs-25.0.95
annotated 4236 emacs-25.2-rc1
annotated 4280 emacs-25.1.91
lightweight 4305 emacs-25.0.94
lightweight 4329 emacs-25.1.90
traversed 130257 commits
more than 10 tags found; listed 10 most recent
gave up search at 5c587fdff164e8b90beb47f6da64b4884290e40a
emacs-25.1-129847-gdcc3ef3ee7
Now, I can't claim that I can wrap my head around
finish_depth_calculation() and the way depth is calculated here, let
alone those huge numbers (probably per not-contained tag or so), but
that order of operations is clearly wrong.
Now, if we fix finish_depth_calculation() to actually do the work for
all possible candidates, and the sort afterwards, the result is:
LANG=C devgit describe --tags --debug --candidates=10
searching to describe HEAD
lightweight 129847 emacs-25.1
lightweight 129864 emacs-25.1-rc2
lightweight 129904 emacs-25.1-rc1
annotated 129981 emacs-25.2
lightweight 130004 emacs-25.0.95
annotated 130016 emacs-25.2-rc2
annotated 130032 emacs-25.2-rc1
annotated 130076 emacs-25.1.91
lightweight 130083 emacs-25.0.94
lightweight 130125 emacs-25.1.90
traversed 130257 commits
more than 10 tags found; listed 10 most recent
gave up search at 5c587fdff164e8b90beb47f6da64b4884290e40a
emacs-25.1-129847-gdcc3ef3ee7
LANG=C devgit describe --tags --debug --candidates=100
searching to describe HEAD
annotated 13470 emacs-24.5-rc3-fixed
lightweight 13471 emacs-24.5-rc3
lightweight 13476 emacs-24.5-rc2
lightweight 13482 emacs-24.5-rc1
lightweight 13511 emacs-24.4.91
lightweight 13554 emacs-24.4.90
lightweight 13899 emacs-24.4
lightweight 13902 emacs-24.4-rc1
lightweight 13959 emacs-24.3.94
annotated 13972 mh-e-8.6
lightweight 14049 emacs-24.3.93
lightweight 14176 emacs-24.3.92
lightweight 129847 emacs-25.1
lightweight 129864 emacs-25.1-rc2
lightweight 129904 emacs-25.1-rc1
lightweight 129971 emacs-25.0.92
annotated 129981 emacs-25.2
lightweight 130004 emacs-25.0.95
annotated 130016 emacs-25.2-rc2
annotated 130032 emacs-25.2-rc1
annotated 130076 emacs-25.1.91
lightweight 130083 emacs-25.0.94
lightweight 130099 emacs-25.0.91
lightweight 130125 emacs-25.1.90
lightweight 130239 emacs-25.0.93
lightweight 130255 emacs-25.0.90
traversed 130257 commits
more than 26 tags found; listed 26 most recent
gave up search at 96b894717caa773aa6d98ff57385f1c7537e8972
emacs-24.5-rc3-fixed-13470-gdcc3ef3ee7
At least the latter kind of makes sense:
git rev-list --count emacs-24.5-rc3-fixed..dcc3ef3ee7
13470
In short, I think that depth calculation is heavily problematic. First,
commit_list_insert_by_date() influences which tags are considered at all.
Second, and worse, the code for "describe cmt" clearly tries to
calculate the number of commits in cmt that are not in t (for each
candidate t), which ought to be "rev-list --count t..cmt", but fails
spectacularly.
I kind of suspect the priming of the tag depth on first encounter
("t->depth = seen_commits - 1;") to be the culprit because that depends
on the way we walk, not the topology, but I wouldn't know how to do better.
Michael
prev parent reply other threads:[~2017-08-29 13:34 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-08-26 14:47 git describe and "the smallest number of commits possible" Kévin Le Gouguec
2017-08-28 18:24 ` Stefan Beller
2017-08-29 13:34 ` Michael J Gruber [this message]
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=f90ddec3-1bef-0b2f-b98b-e65247655b8a@grubix.eu \
--to=git@grubix.eu \
--cc=git@vger.kernel.org \
--cc=kevin.legouguec@gmail.com \
--cc=sbeller@google.com \
/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).