git --atomic fallback causes non-zero exit code

Expected Behavior

Commands triggering git push should return exit code zero hosts running git <2.4.0, ie Azure DevOps 😦

Current Behavior

Commands using git push without --atomic support always return the exit code of the atomic request attempt on success.

Possible Solution

This appears to be because childProcess sets process.exitCode on error. https://github.com/lerna/lerna/blob/6cb8ab2d4af7ce25c812e8fb05cd04650105705f/core/child-process/index.js#L138

When the non-atomic fallback is triggered the exit code from the first attempt remains.

Resetting the exit code in the catch on atomic errors does fix this issue… but I’m not familiar enough to gauge the risk of race condition when doing so.

https://github.com/lerna/lerna/blob/6cb8ab2d4af7ce25c812e8fb05cd04650105705f/commands/version/lib/git-push.js#L26

diff --git a/commands/version/lib/git-push.js b/commands/version/lib/git-push.js
index 35936398..a7f0de1e 100644
--- a/commands/version/lib/git-push.js
+++ b/commands/version/lib/git-push.js
@@ -25,6 +25,8 @@ function gitPush(remote, branch, opts) {
       ) {
         // --atomic is only supported in git >=2.4.0, which some crusty CI environments deem unnecessary to upgrade.
         // so let's try again without attempting to pass an option that is almost 5 years old as of this writing...
+        // reset process.exitCode set by childProcess on atomic failure
+        process.exitCode = 0;
         log.warn("gitPush", error.stderr);
         log.info("gitPush", "--atomic failed, attempting non-atomic push");
 

Steps to Reproduce (for bugs)

1. Remote host using git <2.4.0
2. lerna publish patch -y
3. echo $?

Exit code will be non-zero even on command success.

Your Environment

Executable	Version
lerna --version	4.0.0
npm --version	7.3.0
yarn --version	1.22.10
node --version	15.5.0

[kkazala]

@miluxmilux I had the same issue. I use lerna version --no-push and git push --follow-tags origin HEAD:$(Build.SourceBranch) instead. At least I don’t get false exitcode 1 😃

[MrStLouis]

Identical issue here, worrying how stale this issue is