On Sun, Dec 4, 2016 at 1:34 PM, Bob Proulx <[email protected]> wrote: > Bob Proulx wrote: >> There has been some progress, regression, and progress again on the >> git server side of things. > > Plus one more time... > > One part of the service that I had not thought about testing was use > of git clone --depth 1 to build shallow clones. This results in two > separate problems in different places. Which again was discovered and > reported causing me to roll git back to the old server. > > Thank you Evelyn for the problem report! It was someone a couple of > miles from me that I know in real life. It is a small world! > > The first problem is with the git-daemon server. It is run as the > nobody user and has no permissions to write files in the file system. > However it is trying to create a shallow_XXXXXX file in the git > repository where XXXXXX is replaced by the mkstemp(3) file name guess. > This action is not permitted and should not be permitted. The git > daemon should not be writing files there. This prevents shallow > clones using anonymous git from functioning. However on other systems > the shallow_XXXXXX file is placed elsewhere that there is write > access. I am chasing this down. > > The second problem is that the http dumb transport also fails when > using the git clone --depth 1 option. However we should be using the > transport which should be working. I am chasing this down.
Thanks for all the work, Bob!
