Re: Unexpected git merge exit code when killing merge driver during ancestor merge

["brian m. carlson" <sandals@crustytoothpaste.net>]

[-- Attachment #1: Type: text/plain, Size: 3897 bytes --]

On 2024-04-04 at 16:16:05, Matt Cree wrote:
> Hello all. I have observed some strange behaviour when exiting a custom merge driver that I was wondering if there’s any reason for — I think it may be a bug but I’ll leave that to you to decide.
> 
> I’m configuring that merge driver to exit during a merge at the first sign of conflicts — the exact nature of the rules for the decision to exit early isn’t too important I think though so given it’s ‘work stuff’ I’ll leave some details out.
> 
> Here is my current understanding of how the ort strategy will deal with this.
> 
> - Ort runs the merge driver with the parameters for the current file to be merged
> - When the driver returns exit code 0 is returned it is treated as having no conflicts
> - When the driver returns exit code 1-128 is returned it is treated as having conflicts
> - When the driver returns exit code 129+ is returned it is treated as some kind of error scenario
> 
> 
> Then subsequently
> - If all files returned exit code 0 during the merge git will return exit code 0 i.e. no conflicts
> - If any file returned exit code 1-128 during the merge git will return exit code 1 i.e. conflicts
> - At any time if the driver returns 129+, git will stop merging and return exit code 2 i.e. error?
> 
> However, when setting up a criss-cross merge scenario and ‘short circuiting’ the merge during an ancestor merge, I get exit code 134
> 
> Here’s a couple of quick scripts that help recreate the situation https://gist.github.com/mattcree/c6d8cc95f41e30b5d7467e9d2b01cd3d

Thanks for the repro steps.  I'm on Debian, which uses dash as /bin/sh,
and I also use a different default branch (dev), so I was able to
reproduce with the following patch applied:

----
diff --git a/init-repo.sh b/init-repo.sh
old mode 100644
new mode 100755
index e0f42a4..25d7f25
--- a/init-repo.sh
+++ b/init-repo.sh
@@ -1,5 +1,5 @@
 rm -rf merge-driver-test
 mkdir merge-driver-test
 cd merge-driver-test
-git init .
+git init -b master .
 git commit --allow-empty -m "Initial"
\ No newline at end of file
diff --git a/run-merge.sh b/run-merge.sh
old mode 100644
new mode 100755
diff --git a/run-recursive-merge.sh b/run-recursive-merge.sh
old mode 100644
new mode 100755
index 6920720..c63d652
--- a/run-recursive-merge.sh
+++ b/run-recursive-merge.sh
@@ -1,3 +1,5 @@
+#!/bin/sh
+
 cd merge-driver-test
 
 current_time=$(date "+%Y%m%d-%H%M%S");
@@ -12,7 +14,7 @@ featureA="$current_time-feature-a";
 featureB="$current_time-feature-b";
 featureC="$current_time-feature-c";
 
-function writeFiles() {
+writeFiles() {
 cat > $xmlFileName << EOM
 <?xml version="1.0" encoding="UTF-8"?>
 <CustomLabels xmlns="http://soap.sforce.com/2006/04/metadata">
----

I take it from the "Abort trap" message below, you're on macOS, but I
don't think that's relevant to reproduction.

> The logs also show 
> 
> ```
> Assertion failed: (opt->priv == NULL), function merge_switch_to_result, file merge-ort.c, line 4661. ./run-recursive-merge.sh: line 162: 78797 Abort trap: 6 git merge $featureC --no-ff --no-commit
> ```

This is definitely a bug because we triggered an assertion.  The
assertion asserts that that case will never happen, so if it does, we've
made a mistake in our code.

This also explains the 134 exit status, because on most Unix systems,
`SIGABRT` is signal 6, and when a program exits with a signal, the shell
returns an exit status of 128 plus the signal number.  Because a failed
assertion calls `abort`, which raises `SIGABRT`, that would lead to an
exit status in the shell of 134.

I've CC'd Elijah Newren, who's the author of merge-ort and who wrote the
code.  I'm not familiar at all with merge-ort, so I can't speak to what
might be going wrong here.
-- 
brian m. carlson (they/them or he/him)
Toronto, Ontario, CA

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 262 bytes --]