Windows runner: ssh cannot resolve submodule hostname

[audricschiltknecht]

Hello,

I've setup an action for a private repository with (private) submodules. I'm using the deploy keys with comments as per the docs.

Everything works fine on the ubuntu-latest runner but when I switch to windows-latest it seems that ssh is not using the proper hostname and instead use the one including the keyname, resulting in an obvious "Name not known" error:

   17:58:32.064475 run-command.c:655       trace: run_command: unset GIT_CONFIG_PARAMETERS GIT_DIR; GIT_PROTOCOL=version=2 ssh -o SendEnv=GIT_PROTOCOL git@key-<SHA>.github.com 'git-upload-pack '\''<ORG>/<SUBMODULE>'\'''
  Cloning into 'D:/a/<REPO>/<REPO>/<SUBMODULE>'...
  ssh: Could not resolve hostname key-<SHA>.github.com: Name or service not known
  Error: fatal: Could not read from remote repository.

This seems a bit weird as I haven't seen anyone mentioning that problem in the issues, so I'm wondering if it could be a config problem instead?

Thanks!

[audricschiltknecht]

I've dug a bit deeper and it seems there is something going on with the home path.
The ssh config is created in C:\Users\runneradmin/.ssh/ as seen in the logs:

2022-11-03T19:53:35.8735847Z ##[group]Run webfactory/[email protected]
2022-11-03T19:53:35.8736307Z with:
2022-11-03T19:53:35.8738032Z   ssh-private-key: ***

2022-11-03T19:53:35.8738308Z env:
2022-11-03T19:53:35.8738526Z   GIT_TRACE: 1
2022-11-03T19:53:35.8738759Z   GIT_CURL_VERBOSE: 1
2022-11-03T19:53:35.8739029Z   GIT_SSH_COMMAND: ssh -vvv
2022-11-03T19:53:35.8739277Z ##[endgroup]
2022-11-03T19:53:39.2552847Z Adding GitHub.com keys to C:\Users\runneradmin/.ssh/known_hosts

but it appears that SSH is expecting it somewhere else when we clone the submodule:

2022-11-03T19:53:52.8591485Z OpenSSH_9.0p1, OpenSSL 1.1.1q  5 Jul 2022
2022-11-03T19:53:52.8591959Z debug1: Reading configuration data /etc/ssh/ssh_config
2022-11-03T19:53:52.8592696Z debug3: expanded UserKnownHostsFile '~/.ssh/known_hosts' -> '/d/a/_temp/<some uuid>/.ssh/known_hosts'
2022-11-03T19:53:52.8593507Z debug3: expanded UserKnownHostsFile '~/.ssh/known_hosts2' -> '/d/a/_temp/<some uuid>/.ssh/known_hosts2'
2022-11-03T19:53:52.8594366Z debug2: resolving "key-<SHA>.github.com" port 22
2022-11-03T19:53:52.8595246Z debug3: resolve_host: lookup key-<SHA>.github.com:22
2022-11-03T19:53:52.8596185Z ssh: Could not resolve hostname key-<SHA>.github.com: Name or service not known
2022-11-03T19:53:52.8620858Z ##[error]fatal: Could not read from remote repository.
2022-11-03T19:53:52.8628627Z 
2022-11-03T19:53:52.8628905Z Please make sure you have the correct access rights

[audricschiltknecht]

Finally found the culprit: https://github.com/actions/checkout/blob/v3.1.0/src/git-auth-helper.ts#L122

checkout is overriding $HOME before doing the checkout. It's copying .gitconfig but not the .ssh folder :|

[audricschiltknecht]

For reference, I've quickly hacked the action to change the system-wide config on windows: https://github.com/audricschiltknecht/ssh-agent/tree/fix-path-windows

This is not ideal as I don't know what happens with the config put there at the end of the action or even if for self-hosted runners this will work.

[DilumAluthge]

I have encountered a bug on Windows that seems similar to this issue, but my workaround is different. Specifically, I'm able to resolve the bug by setting GIT_SSH_COMMAND as described here: #230 (comment)