Win32: fix checkout problem with directories exceeding MAX_PATH

[kblees]

This fixes issue msysgit/msysgit#227
TODO: change symlinks.c upstream to use strbuf instead of fixed length buffers.

[t-b]

Is that bug also present in master?

[dscho]

TODO: change symlinks.c upstream to use strbuf instead of fixed length buffers.

@kblees do you want to address that before or after the merge?

[dscho]

Is that bug also present in master?

Yes. I cherry-picked it.

[dscho]

Is that bug also present in master?

Yes. I cherry-picked it.

Wooooow! I should not have done that. Not at all.

@kblees we cannot use MAX_LONG_PATH in symlinks.c, as you undoubtedly know, but I mistook your TODO for a "nice to have" rather than "this will break anywhere but Windows" ;-)

I am pretty confident that I unbroke it (BuildHive did not yet respond, but I am sure it will) with these three commits: d49a708...b31f626

@kblees could you have a look and adjust the PR accordingly? It really should apply very cleanly, and I'd suggest moving the strbuf patch to be the first of the two commits.

[kblees]

@dscho sorry for the mess, I probably should have marked this as preliminary more clearly.

This round compiles and passes the test suite on both Windows (on the maint-1.9 branch, master seems to be completely broken ATM) and Linux (Junio's master).

Also fixes a segfault in git-status with long paths.

[buildhive]

MSysGit - the development behind Git for Windows » git #240 SUCCESS
This pull request looks good
(what's this?)

[dscho]

master seems to be completely broken ATM

I only noticed when I started playing with my shiny new Windows VM (sponsored by Microsoft!) that there were test failures. However, last time I ran the test suite on master on my crappy old Windows VM (where it took all of 6 hours (!) to pass), there were no failures...

[kblees]

Updated the PR with the patches that are queued upstream.

[buildhive]

MSysGit - the development behind Git for Windows » git #242 SUCCESS
This pull request looks good
(what's this?)

[t-b]

I've seen that nd/path-max-must-go was merged upstream. The others also?

[kblees]

@t-b everything is in upstream/master except for the last patch, which depends on long path support

[t-b]

@kblees Thanks. So, just to be sure, we can merge the PR to maint-1.9?

[kblees]

@t-b Yes, ready to merge

[t-b]

The test suite runs trough with this PR. Only some svn tests have the usual problems.

[fetis]

When this will be released? I need it sooooo much

[dscho]

@fetis we made it soooooo easy to make your own installer from the current master. ;-)

Why don't you give it a try?

[fetis]

I did and it's unclear for me how can I build HEAD of maint-1.9 branch. It always reset to v1.9.4.msysgit.0 tag :((

[t-b]

@fetis See https://github.com/msysgit/msysgit/wiki/How-and-when-does-Git-for-Windows-get-released for some hints.

[fetis]

I read it twice. My set of command is

cd git
git remote add junio https://github.com/git/git.git
git remote update junio origin
git fetch junio --tags
git checkout maint-1.9
git pull origin maint-1.9
git checkout -b tentative-maint-1.9
git branch -u origin/maint-1.9
/share/msysgit/merging-rebase.sh maint-1.9

cd /
version=1.9.5
git_tag=v$version.msysgit.0
msysgit_tag=$version-preview`date +"%Y%m%d"`
# Tag /git
git tag -a -m "Git $version" $git_tag
# Commit the submodules
git commit -m "Commit submodule in preparation of $version" git doc/git/html src/git-cheetah

/share/WinGit/release.sh $msysgit_tag

but it stull use 1.9.4

[fetis]

any help on this?