andygrove opened a new issue, #3292:
URL: https://github.com/apache/datafusion-comet/issues/3292
### Describe the bug
## Description
The `build-release-comet.sh` script has a bug where it builds native
binaries from the specified branch (via `-b` flag) but builds JVM components
from whatever branch is currently checked out on the host machine. This can
result in mismatched releases where native code and JVM code come from
different branches.
## Steps to Reproduce
1. Check out `main` branch on host machine
2. Run the release build script with a different branch specified:
```bash
./build-release-comet.sh -b branch-0.13
```
3. Observe the resulting artifacts
## Expected Behavior
The entire build (both native binaries and JVM components) should be built
from the branch specified via the `-b` flag (`branch-0.13` in this example).
## Actual Behavior
- Native binaries are correctly built from `branch-0.13` (inside Docker
containers)
- JVM components and git hash are built/captured from `main` (on the host
machine)
This creates a mismatched artifact that combines code from two different
branches.
## Root Cause
The script has two separate build phases:
1. **Native binaries** (lines 132-160): Built inside Docker containers that
properly clone the repo and checkout the specified branch
```bash
docker run ... $BUILDER_IMAGE_AMD64 "${REPO}" "${BRANCH}" amd64
```
2. **JVM components** (lines 196-209): Built on the host machine from
whatever is currently checked out
```bash
pushd $COMET_HOME_DIR
GIT_HASH=$(git rev-parse --short HEAD) # Gets hash from current checkout!
./mvnw install # Builds from current checkout!
```
The script never switches branches on the host machine before building the
JVM components.
## Proposed Solutions
### Option 1: Check out the branch on host (Recommended)
Add a git checkout before building JVM components:
```bash
pushd $COMET_HOME_DIR
git checkout "${BRANCH}" # Add this line
GIT_HASH=$(git rev-parse --short HEAD)
./mvnw install
popd
```
### Option 2: Document the requirement
If the script is intended to build JVM components from the current branch,
document this clearly:
- Update the usage message to indicate the user must be on the correct branch
- Add a validation check that fails if the current branch doesn't match the
`-b` parameter
### Option 3: Build everything in Docker
Build JVM components inside the Docker container alongside the native
binaries to ensure consistency.
## Impact
This bug can lead to:
- Release artifacts that combine code from multiple branches
- Version mismatches between native and JVM components
- Difficult-to-debug issues when the release doesn't behave as expected
- Incorrect git hashes in release metadata
## Environment
- Script: `dev/release/build-release-comet.sh`
- Affected versions: Current main branch
### Steps to reproduce
_No response_
### Expected behavior
_No response_
### Additional context
_No response_
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
