Re: [jgit-dev] Modification status

[Christian Halstrick <christian.halstrick@xxxxxxxxx>]

Wow, you can share your repo. I didn't expected that (I often have
support requests from people working on closed source repos). Then
it's maybe easy: Can you sent me a zip of repo (including working tree
and .git). Then I can run it here against a jgit version with more
tracing and can debug.

You're last experiments make sense. If you have really modified the
working tree file then it's exactly the call to
WorkingTreeIterator.isModified() which should ultimately compute sha-1
of the file content and compare it to what we have in index. The
method will not do this if lastModification date and file length have
not changed. And there are some special cases regarding "racily-clean"
situations (if you happen to modify, stage and modify again a file
before the filesystem timer has advanced). But I can find out about
this with the zip file.
Ciao
  Chris


On Thu, Oct 30, 2014 at 5:28 AM, Aleś Bułojčyk <alex73mail@xxxxxxxxx> wrote:
> Hi Chris.
>
> See my repo at https://github.com/alex73/trans. I made tests on the
> master==cb073c84ff70f690e2811ffc748df4d3dc4a1588.
>
> alex@d:~/om3/t1$ ./org.eclipse.jgit.pgm-3.5.1.201410131835-r.sh
> debug-show-dir-cache
> 100644     77 2014-10-28,06:35:47.000
> 381fd17ed31490ca7603a2849f9e969dd29cfb3d 0    compiled/1.txt
> 100644  68868 2014-10-28,06:35:47.000
> 27fbd2d93d47fed96b00982072fa2cc6b265ec91 0    compiled/Bundle_be.properties
> 100644  69790 2014-10-28,06:35:47.000
> 2ee9a811ccd038f37763581113b9b3ac67966fb8 0
> compiled/org/omegat/Bundle_be.properties
> 100644    631 2014-10-30,06:51:31.000
> df3a2a85b3f311e0ae7958880319c2ee90a2c367 0    omegat.project
> 100644      0 2014-10-29,07:52:38.203
> e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 0    omegat/ignored_words.txt
> 100644      0 2014-10-29,07:52:38.203
> e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 0    omegat/learned_words.txt
> 100644    861 2014-10-30,06:51:31.000
> 5b9fa1d37fbeda44c3515db9700ab03f7ed2d023 0    omegat/project_save.tmx
> 100644   9992 2014-10-28,06:35:47.000
> f415a6ad1180a7934294026f3ab26f8f434cb098 0    reki.tmx
> 100644     25 2014-10-28,06:35:47.000
> b7e50185defb84dfdbc7f87fccc772b70436ca3b 0    source/1.properties
> 100644     77 2014-10-28,06:35:47.000
> 381fd17ed31490ca7603a2849f9e969dd29cfb3d 0    t2/1.txt
> 100644     25 2014-10-28,06:35:47.000
> b7e50185defb84dfdbc7f87fccc772b70436ca3b 0    t2/1_be.properties
> 100644  68889 2014-10-28,06:35:47.000
> 966a7b4c7bd86b99393ee425b508ac10fd068038 0    t2/Bundle_be.properties
> 100644  69811 2014-10-28,06:35:47.000
> 7af3559dc4d2e4ea6fbce4c8abbdfd95fd70dd20 0
> t2/org/omegat/Bundle_be.properties
> 100644     77 2014-10-28,06:35:47.000
> 381fd17ed31490ca7603a2849f9e969dd29cfb3d 0    target/1.txt
> 100644     25 2014-10-28,06:35:47.000
> b7e50185defb84dfdbc7f87fccc772b70436ca3b 0    target/1_be.properties
> 100644  68889 2014-10-28,06:35:47.000
> 966a7b4c7bd86b99393ee425b508ac10fd068038 0    target/Bundle_be.properties
> 100644  69811 2014-10-28,06:35:47.000
> 7af3559dc4d2e4ea6fbce4c8abbdfd95fd70dd20 0
> target/org/omegat/Bundle_be.properties
>
>
> alex@d:~/om3/t1$ git config -l
> gui.spellingdictionary=none
> push.default=simple
> user.email=alex73mail@xxxxxxxxx
> user.name=Aleś Bułojčyk
> color.ui=auto
> core.repositoryformatversion=0
> core.filemode=false
> core.bare=false
> core.logallrefupdates=true
> core.autocrlf=false
> remote.origin.url=git@xxxxxxxxxx:alex73/trans.git
> remote.origin.fetch=+refs/heads/*:refs/remotes/origin/*
> gui.wmstate=normal
> gui.geometry=1920x1001+0+25 507 692
>
> There is no .gitattributes.
>
> Could it be side effects of the my using of working tree ? My software
> usually creates file omegat/project_save.tmx.new(untracked), then call some
> checkouts, delete branches, etc, then renames omegat/project_save.tmx.new
> into omegat/project_save.tmx(tracked). So, jgit can already saw file with
> the same length/date/sha1, but for other filename. Or even for
> omegat/project_save.tmx filename before some operations. And working tree
> contains untracked file with committed version of omegat/project_save.tmx
> but with other name(.bak).
>
> My working copy diff:
>        </tuv>
> -      <tuv lang="be" changeid="Aleś Bułojčyk" changedate="20141029T032910Z"
> creationid="Aleś Bułojčyk" creationdate="20141028T041213Z">
> -        <seg>1111</seg>
> +      <tuv lang="be" changeid="Aleś Bułojčyk" changedate="20141030T035118Z"
> creationid="Aleś Bułojčyk" creationdate="20141028T041213Z">
> +        <seg>5555</seg>
>        </tuv>
>
> I investigated a little bit: treeWalk.next() doesn't return
> omegat/project_save.tmx because IndexDiffFilter.include() returns false
> after :
>         // Only one chance left to detect a diff: between index and working
>         // tree. Make use of the WorkingTreeIterator#isModified() method to
>         // avoid computing SHA1 on filesystem content if not really needed.
>         return wi.isModified(di.getDirCacheEntry(), true,
> tw.getObjectReader());
>
> So, looks like my software need to force SHA1 comparison.
>
> WBR, Alex.
>
> On 29 October 2014 09:04, Christian Halstrick
> <christian.halstrick@xxxxxxxxx> wrote:
>>
>> That's strange. I never saw this.
>>
>> Regarding cache: Yes, there is a cache. It called "the git index" :-).
>> But there is nothing what you expect, a in memory cache of file
>> content/metadata. IndexDiff will compare the files modification date
>> and length to what is "cached" in git's index. If both are the same
>> (and if the file is not "racily clean / smudged") then git will think
>> the file does not differ to what we have in index.
>>
>> Can you share the repo?
>> Can you attach the output of the following command executed in your
>> repo: "org.eclipse.jgit.pgm-3.5.1.201410131835-r.sh
>> debug-show-dir-cache"
>> Can you attach the output of the following command executed in your
>> repo: "git config -l" (Be careful not to expose secret data here and
>> delete passwords etc)
>> Does your repo contain .gitattributes?
>
>