Temporary git files for the gitdir created on a separate drive in workdir

>
> Hultqvist <hultqvist@silentorbit.com> writes:
>
> > Considering that the gitdir could be located on a different drive than
> > the workdir wouldn't it make more sense to create the temporary files
> > in a subdirectory inside the gitdir rather tan in the workdir?
>
> I do not think we intend to create temporary files, whose final
> destination is somewhere under $GIT_DIR/, in any working tree;
> rather, I think we try to create them inside $GIT_DIR (or possibly
> if the destination is a file in a subdirectory of $GIT_DIR, then in
> the same subdirectory).  What you are seeing definitely smells like
> a bug in the worktree code, perhaps getting confused by the fact
> that the full path to these places look "unusual" by starting with a
> single alphabet followed by a colon (IOW, this may manifest only in
> Windows port).

>
> On Fri, Sep 7, 2018 at 6:48 PM Junio C Hamano <gitster@pobox.com> wrote:
> >
> > Hultqvist <hultqvist@silentorbit.com> writes:
> >
> > > Considering that the gitdir could be located on a different drive than
> > > the workdir wouldn't it make more sense to create the temporary files
> > > in a subdirectory inside the gitdir rather tan in the workdir?
> >
> > I do not think we intend to create temporary files, whose final
> > destination is somewhere under $GIT_DIR/, in any working tree;
> > rather, I think we try to create them inside $GIT_DIR (or possibly
> > if the destination is a file in a subdirectory of $GIT_DIR, then in
> > the same subdirectory).  What you are seeing definitely smells like
> > a bug in the worktree code, perhaps getting confused by the fact
> > that the full path to these places look "unusual" by starting with a
> > single alphabet followed by a colon (IOW, this may manifest only in
> > Windows port).
>
> I agree. Auditing the setup code did not reveal anything though. Our
> code should recognize these unusual Windows paths as absolute and
> while I spotted an incorrect use of '/' (instead of is_dir_sep) it
> does not explain the problem here.
>
> Hultqvist, if you set environment variable GIT_TRACE_SETUP to 1 and
> run "git status" in G:\Test1, what does it say?
> --
> Duy

>
> The bash commands are using a git and bash bundle that was installed
> in parallel with gitextensions(a gui for git)
>
> G:\Min enhet> set GIT_TRACE_SETUP=1
> G:\Min enhet> git st
> 10:40:28.881927 trace.c:318             setup: git_dir:
> C:/Users/hultqvist/Drive.git
> 10:40:28.881927 trace.c:319             setup: git_common_dir:
> C:/Users/hultqvist/Drive.git
> 10:40:28.881927 trace.c:320             setup: worktree: G:/Min enhet
> 10:40:28.881927 trace.c:321             setup: cwd: G:/Min enhet
> 10:40:28.881927 trace.c:322             setup: prefix: (null)
> 10:40:28.882930 chdir-notify.c:67       setup: chdir from 'G:/Min
> enhet' to 'G:/Min enhet'

>
> On Sat, Sep 8, 2018 at 11:28 AM Hultqvist <hultqvist@silentorbit.com> wrote:
> >
> > The bash commands are using a git and bash bundle that was installed
> > in parallel with gitextensions(a gui for git)
> >
> > G:\Min enhet> set GIT_TRACE_SETUP=1
> > G:\Min enhet> git st
> > 10:40:28.881927 trace.c:318             setup: git_dir:
> > C:/Users/hultqvist/Drive.git
> > 10:40:28.881927 trace.c:319             setup: git_common_dir:
> > C:/Users/hultqvist/Drive.git
> > 10:40:28.881927 trace.c:320             setup: worktree: G:/Min enhet
> > 10:40:28.881927 trace.c:321             setup: cwd: G:/Min enhet
> > 10:40:28.881927 trace.c:322             setup: prefix: (null)
> > 10:40:28.882930 chdir-notify.c:67       setup: chdir from 'G:/Min
> > enhet' to 'G:/Min enhet'
>
> Unfortunately this looks good. Whenever those files 'index',
> 'config'... are created, they should always be created in
> C:\Users\hultqvist\Drive.git, not G:\Min enhet, including their
> temporary versions. I don't know if there are any more changes on the
> windows fork that could affect this though, I only checked git.git.

>
> On Sat, Sep 8, 2018 at 3:09 PM Duy Nguyen <pclouds@gmail.com> wrote:
> >
> > On Sat, Sep 8, 2018 at 11:28 AM Hultqvist <hultqvist@silentorbit.com> wrote:
> > >
> > > The bash commands are using a git and bash bundle that was installed
> > > in parallel with gitextensions(a gui for git)
> > >
> > > G:\Min enhet> set GIT_TRACE_SETUP=1
> > > G:\Min enhet> git st
> > > 10:40:28.881927 trace.c:318             setup: git_dir:
> > > C:/Users/hultqvist/Drive.git
> > > 10:40:28.881927 trace.c:319             setup: git_common_dir:
> > > C:/Users/hultqvist/Drive.git
> > > 10:40:28.881927 trace.c:320             setup: worktree: G:/Min enhet
> > > 10:40:28.881927 trace.c:321             setup: cwd: G:/Min enhet
> > > 10:40:28.881927 trace.c:322             setup: prefix: (null)
> > > 10:40:28.882930 chdir-notify.c:67       setup: chdir from 'G:/Min
> > > enhet' to 'G:/Min enhet'
> >
> > Unfortunately this looks good. Whenever those files 'index',
> > 'config'... are created, they should always be created in
> > C:\Users\hultqvist\Drive.git, not G:\Min enhet, including their
> > temporary versions. I don't know if there are any more changes on the
> > windows fork that could affect this though, I only checked git.git.
>
> BTW do you notice these files showing up after any particular command
> or they're always there after cloning? Perhaps some command got the
> ".git" directory discovery wrong and assumed $GIT_DIR=$GIT_WORK_TREE.
> We have a much bigger problem then.
> --
> Duy

>
> First I need to correct my previous observations.
>
> Today there appeared new set of config files in the root.
> I looked into a few of them and found that their content doesn't match that of the repo at "G:/Min enhet".
> Instead separate files had content from separate git repos within the G drive.
> These repos are not like the one we're discussed previously, they are completely within G: using a classical .git directory.
>
> I guess git is creating the temporary files as close as possible to the root, since "G:\" can't be written to, only "G:\Min enhet". and then copy them to the final destination which in this case is the same drive.
> If so then we can't get away from the duplicate files, the duplicates themselves are the fault of Google Drive File Stream.
> If git would create the config/master/index files within the .git dir to start with that would hide the clutter from the root but not prevent the duplicates.
> Is this observation correct, that git creates temporary files closer to the root than inside the .git directory or a sub directory thereof?
>
> I haven't experiences any issues working with these repos themselves.
> One risk could be that the final "index" is an older version than "index (45)".
> I'm certain this haven't been the case since I push every commit to a remote repo that doesn't accept changes in history.
>
> I understand that master and index needs to be updated with regular use.
> Does config need the same amount of updates?
> I've considered that file to only change when I make explicit changes, not during regular use(push, commit, status).
>
> Thanks for the help looking into this for me.
>
>
> Den sön 9 sep. 2018 kl 17:30 skrev Hultqvist <hultqvist@silentorbit.com>:
>>
>> Since this thread started I haven't seen a single file mentioned being created,
>> Usually they appear during work days when there is more activity.
>> I've never seen the files created directly, only a larger amount of
>> them once in a while.
>>
>> I will run process monitor and get back once I find out more.
>> Den lör 8 sep. 2018 kl 15:44 skrev Duy Nguyen <pclouds@gmail.com>:
>> >
>> > On Sat, Sep 8, 2018 at 3:09 PM Duy Nguyen <pclouds@gmail.com> wrote:
>> > >
>> > > On Sat, Sep 8, 2018 at 11:28 AM Hultqvist <hultqvist@silentorbit.com> wrote:
>> > > >
>> > > > The bash commands are using a git and bash bundle that was installed
>> > > > in parallel with gitextensions(a gui for git)
>> > > >
>> > > > G:\Min enhet> set GIT_TRACE_SETUP=1
>> > > > G:\Min enhet> git st
>> > > > 10:40:28.881927 trace.c:318             setup: git_dir:
>> > > > C:/Users/hultqvist/Drive.git
>> > > > 10:40:28.881927 trace.c:319             setup: git_common_dir:
>> > > > C:/Users/hultqvist/Drive.git
>> > > > 10:40:28.881927 trace.c:320             setup: worktree: G:/Min enhet
>> > > > 10:40:28.881927 trace.c:321             setup: cwd: G:/Min enhet
>> > > > 10:40:28.881927 trace.c:322             setup: prefix: (null)
>> > > > 10:40:28.882930 chdir-notify.c:67       setup: chdir from 'G:/Min
>> > > > enhet' to 'G:/Min enhet'
>> > >
>> > > Unfortunately this looks good. Whenever those files 'index',
>> > > 'config'... are created, they should always be created in
>> > > C:\Users\hultqvist\Drive.git, not G:\Min enhet, including their
>> > > temporary versions. I don't know if there are any more changes on the
>> > > windows fork that could affect this though, I only checked git.git.
>> >
>> > BTW do you notice these files showing up after any particular command
>> > or they're always there after cloning? Perhaps some command got the
>> > ".git" directory discovery wrong and assumed $GIT_DIR=$GIT_WORK_TREE.
>> > We have a much bigger problem then.
>> > --
>> > Duy

>
> Sending again without HTML
>
> Den mån 10 sep. 2018 kl 12:28 skrev Hultqvist <hultqvist@silentorbit.com>:
> >
> > First I need to correct my previous observations.

> > Today there appeared new set of config files in the root.
> > I looked into a few of them and found that their content doesn't match that of the repo at "G:/Min enhet".
> > Instead separate files had content from separate git repos within the G drive.
> > These repos are not like the one we're discussed previously, they are completely within G: using a classical .git directory.
> >
> > I guess git is creating the temporary files as close as possible to the root, since "G:\" can't be written to, only "G:\Min enhet". and then copy them to the final destination which in this case is the same drive.