Re: [jgit-dev] Missing tree during self-tests on NonStop

["Andrey Loskutov" <loskutov@xxxxxx>]

Hi,

I personally fully agree with Christian. Please do not underestimate costs for extra IO calls.
I fear that jgit performance on my 8 GB repository will decrease and this decrease can be measured in *real* money we will lost.

Kind regards,
Andrey Loskutov

http://google.com/+AndreyLoskutov


> Gesendet: Mittwoch, 04. März 2015 um 16:02 Uhr
> Von: "Randall S. Becker" <rsbecker@xxxxxxxxxxxxx>
> An: "'Christian Halstrick'" <christian.halstrick@xxxxxxxxx>
> Cc: 'jgit-dev' <jgit-dev@xxxxxxxxxxx>
> Betreff: Re: [jgit-dev] Missing tree during self-tests on NonStop
>
> > -----Original Message-----
> > From: Christian Halstrick [mailto:christian.halstrick@xxxxxxxxx]
> > Sent: March 4, 2015 7:40 AM
> > To: Randall S. Becker
> > Cc: Shawn Pearce; jgit-dev
> > Subject: Re: [jgit-dev] Missing tree during self-tests on NonStop
> > 
> > I like that JGit gains one more supported platform. But we should take
> care that
> > we are not slowing down JGit on standard JVMs. Doing extra file-system
> calls is
> > something I really avoid in JGit. I am not caring too much about
> JGitTestUtil but
> > changes to the run-time code should not decrease performance. Maybe one
> > could add a detection whether we run on a JVM/filesystem which creates
> > executable files on
> > File.createNewFile() and set a flag. That flag could be used to do avoid
> extra
> > filesystem calls to correct the executable bits. Or, alternatively, when
> > initializing repositories in the filesystem check the behaviour of
> > File.createNewFile() and set a configuration parameter (like
> core.filemode).
> 
> The situation at the moment is that java.io.File#createNewFile()
> documentation from Oracle is silent on the file mode used so it is
> implementation specific, not an issue of standards. Given that JGit and git
> are sensitive to this attribute, leaving it to chance is unwise even if it
> costs a file system call. For NonStop Java 1.6, the execute bit is set by
> this call. We are moving to 1.7 but I suspect that there is no change. Sure,
> a factory pattern could be used to pick off a NonStop personality, but that
> seems a bit extreme to me in this specific situation.
> 
> _______________________________________________
> jgit-dev mailing list
> jgit-dev@xxxxxxxxxxx
> To change your delivery options, retrieve your password, or unsubscribe from this list, visit
> https://dev.eclipse.org/mailman/listinfo/jgit-dev
>