Repo.clone_from does not account for kill_after_timeout

[schanzel]

Hi there,

as stated in the docs, there is a kill_after_timeout parameter to the GitCmd.execute method which does exactly what it's supposed to be when, e.g., given to a repo.git.push call. But it does not seem to be considered when passed to Repo.clone_from (cf. example below). Is this intended or a bug? Or is there another way to freshly clone a repository and pass this argument?

repo = Repo.clone_from(some_remote, some_path, kill_after_timeout=1) # does not timeout
repo.git.push("origin", "master", kill_after_timeout=1) # this does timeout

[schanzel]

I ended up with this workaround for now but I think it would be good to address this problem as clone is a remote action that could time out.

repo = Repo.init(some_path)
repo.git.remote("add", "origin", some_remote)
repo.git.pull("origin", "master", kill_after_timeout=1)

[Byron]

In theory, it should pass through any kwarg to clone as well. Just recently there was the multi_options flag added, which might end up doing what you want.

Even though I didn't try to reproduce the issue, I think it's worth looking at it to see what's going on.

[Yikun]

The clone_from calls git clone function with as_process=True, it skips the kill monitor and call self.AutoInterrupt(proc, command) directly.

So, if we want to change clone_from to with kill monitor, we should set the as_process = False in here, but I'm not sure the detail backgroud of the original implentation.

@Byron Would you like take a look, and give some suggestion?

[Byron]

Thanks for looking into it, having the links to code is very helpful.

Setting as_process=False would immediately collect all output and wait for the process to exit, making the collection of progress impossible. However, the latter is only done if there is a progress handler which isn't necessarily the case for you.

That said, it's clearly an oversight in the implementation, one might want to try adding support for killing the process after timeout for the AutoInterrupt helper as well. Note that the latter might not even work as destructors are not reliable in Python (anymore), but that's another topic.

[Yikun]

Before adding support on it, you could also use the below wordaround to support git clone with timeout:

import git
import os
mygit = git.cmd.Git(os.getcwd())
mygit.clone(
    git.cmd.Git.polish_url(git_url), local_path,
    kill_after_timeout=timeout
)

from [1] https://opendev.org/zuul/zuul/commit/ba1c8c0e310bbe4a874aaa81042b8f1d3e8ac078

[RT4U]

Before adding support on it, you could also use the below wordaround to support git clone with timeout:

import git
import os
mygit = git.cmd.Git(os.getcwd())
mygit.clone(
    git.cmd.Git.polish_url(git_url), local_path,
    kill_after_timeout=timeout
)

from [1] https://opendev.org/zuul/zuul/commit/ba1c8c0e310bbe4a874aaa81042b8f1d3e8ac078

How can I get the repo object from this for iter_commits?

Also specifying as_process in clone_from function results in a duplicate keyword arg exeception.

[Byron]

How can I get the repo object from this for iter_commits?

git.Repo(local_path) might be worth a try.