-
Notifications
You must be signed in to change notification settings - Fork 1k
Conversation
oh, awesome! i think i've run into similar issues in the past in gps. there isn't the same kind of testing harness in gps as dep has (though maybe there should be), so it's not as central of a fix as this. still, though, it would be great to adapt this to over there. i'm going to wait on this until we can get those travis failures fixed, though. hopefully #156 will do that for us. |
This should treat parallel tests. So please wait. |
still bad |
ya this was super racey before it would fail like 40% of the time but i'd just rebuild, thanks for putting the time into fixing this :) |
i don't know why travis can't get its act together 😡 |
terminate all git.exe and git-remote-https.exe aaa
Whatever, it works on appveyor - that's the only place where a change actually occurred. |
Thank you. I just figure out what process hold the handle of the directory. https://technet.microsoft.com/en-us/sysinternals
|
Read the destination of named symbolic link
As far as I can see, It seeks git.exe call fork(2) to do something about ls-remote or git-remote-http on Windows. But it often quit without waiting the child process. So the git.exe hold the handle of directory which should be removed. This occured while removing temporary directory in tg.cleanup. I don't know how to fix this git process. For example:
Waiting something (maybe git.exe) fixes this issue. But no reason to wait the bad child. Then, terminate all git.exe in tg.This patch should work.