Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [jgit-dev] [cross-project-issues-dev] EGit / line ending problems with simrel repo

2012/8/16 Ed Merks <ed.merks@xxxxxxxxx>

I doubt this conclusion is correct.  You can tell git which files (extensions) are binary via .gitattributes
I've not had problems with binary files (images for example) getting mangled.

JGit doesn't support gitattributes yet, that's why EGit doesn't know the difference between
ASCII and binary files. There is a pending draft [1] since a long time but it seems Marc didn't find
the time to continue on that. Tasktop recently said they intend to work on this feature later
this year [2].  


On 16/08/2012 12:32 PM, Ed Willink wrote:

After a quick Google it seems GIT does not normalize Windows line endings unless autocrlf is set true, which may have other bad effects like normalizing binary files too.

So if we use the default autocrlf=false we are left with the bad alternative; get all CR-LF producing tools fixed, which is difficult since Eclipse's default line-endings on Windows are CR-LF, so CR-LF production is correct.

I did a quick scan of my Workspace for CR-LFs; Eclipse locked up with over 65000 search matches.

After a kill and restart and a search in a single project, I had two rogue files; both manually edited Java files. It seems that once you get a CR-LF from somewhere, JDT's indentation preservation also preserves line termination and once there are some CR-LFs, JDT thinks you like them.

So until EGIT acquires CVS's binary flag our only solution is to regularly manually remove all CR-LFs that have leaked in somehow.


        Ed Willink

On 16/08/2012 10:44, Eike Stepper wrote:
Am 16.08.2012 11:25, schrieb Ed Willink:
Hi Eike


My suspicion is that the problem is in the comparison tooling.

The files in the repo seem to be normalized to LF line endings, but some Windows tooling creates CR-LF; some tools can
be fixed via Bugzillas but it's a losing battle.

I totally disagree. All these tools have been working fine with all other version control systems.
We agree. I was just elaborating the bad alternative.

Good ;-)

And the false positives in the staging view appear with no comparison tool being involved. And it's impossible to get
rid of them by means of the tool (EGit) that has created them.
There is a comparison. EGIT must do a file compare to determine whether the file is changed. If you edit a file and edit
it back again, the file disappears from the staging view, so EGIT must be using content rather than timestamp to detect
actually it's using both as also c git does, jgit tries to use meta data as much as possible
since lstat is a lot faster than computing SHA1 of the content. Though In some cases it has
to do that in order to provide correct results
Oh, of course I know that. I *guess* it's done with the SHA1 digests of the files' contents because they're needed anyway.
I'll try to implement what Gunnar suggested in [3] in order to improve
EGit's line break handling. Though full handling of autocrlf requires
jgit support for gitattributes.



Back to the top