I am not casting a vote.

These are observations. I am not treating them as a reason to reject
the release.

The source zip matches the tag. The staged jars were built from that
same commit, with JDK 21, from a personal fork
([email protected]:Bughue/seata.git). A JDK 21 build of the source zip
matches the class files. The clone recorded in the jar is not the
Apache repository named in the POM.

The binary tar and the staging jars were not compiled with the same
JDK. seata-core in the binary tar records Build-Jdk-Spec: 1.8. The
same jar in the staging repository records Build-Jdk-Spec: 21. The
release profile requires Java [21,) for the Maven deploy. The vote
mail says the build environment is JDK 8 or newer.

I am asking the release manager, or a mentor, to look at these
observations and say whether they warrant a deeper review of this
release.

Podling vote:

The podling vote on [email protected] passed on 2026-10-31, with 3
binding PPMC +1, 1 non-binding +1, and no -1. The artifacts named
there are the ones in this vote. I am not treating that tally as my
own review.

Result: https://lists.apache.org/thread/mb2jxh6pwh6rmok7gf4v4cwwzoqdk9z6
Vote: https://lists.apache.org/thread/p21v9qlbnp6lobhnpc3hb71m889561dd

Those threads are titled Round4. This call is titled Round3. The dist
URL, tag, hash, and staging URL are the same.

Source archives:

Checked each with its .asc and .sha512, staged at
https://dist.apache.org/repos/dist/dev/incubator/seata/2.7.0/

- apache-seata-2.7.0-incubating-src.zip
- apache-seata-2.7.0-incubating-bin.tar.gz

SHA-512 matched for both, and both have a good signature from the key
the vote names, id suffix 5D08040F. The source zip matches git tag
v2.7.0 at ddd37fffb92d34bdc12694771bb2ac6fb8d93c10, with no
AppleDouble files, no content differences, and no archive-only files.
incubating is in both filenames.

License and headers:

LICENSE, NOTICE, and DISCLAIMER are present. I am not marking LICENSE
accurate. license-eye 0.9.0 header check against .licenserc.yaml
exited 0: 4098 files, 3591 valid, 0 invalid, 507 ignored. That is the
project's header check, and it was clean.

No license violation was found, and the source tree has no compiled
binaries. The binary tar has 355 jars, which are the staged binary.
Java, Scala, Python, and Go have no port comments and no non-ASF
copyright lines. Of the JavaScript and CSS files, 74 have an Apache
header. LICENSE names the other three: bootstrap.css is Apache 2.0
with bundled normalize.css under MIT, PathMap.js is Apache 2.0 with an
Adobe copyright, and font-awesome.css is MIT. That CSS header also
says the font is SIL OFL 1.1, which LICENSE does not name, and no Font
Awesome font file is in the source zip or the binary tar.

distribution/licenses has 390 third-party license texts. LICENSE names
two of them by path, normalize.css-MIT and
alicloud-console-components-MIT, and does not list the other 388.

Build:

The JDK used for the staged jars is in the jar, not in the compiler
default. META-INF/MANIFEST.MF in the staged seata-core-2.7.0.jar
records Build-Jdk-Spec: 21. The release profile in build/pom.xml
requires Java [21,), so seata-threadpool-loom and json-common-jackson3
are included in Maven staging.

A JDK 21 build of the source zip matches the class files in the 43
staging jars compared. Differing class count: 0. The JDK 21 unit tests
returned BUILD SUCCESS: 13127 tests, 0 failures, 0 errors, 131
skipped, in 12:05. That is not the CI matrix.

./mvnw -B -ntp clean package -DskipTests=true
./mvnw -B -ntp test
JAVA_HOME=/usr/lib/jvm/java-21-openjdk
OpenJDK 21.0.13

Staged Maven files:

seata-git.properties is generated by Git-Commit-Id-Plugin and is
inside the staged jar. It records commit
ddd37fffb92d34bdc12694771bb2ac6fb8d93c10, the same as the tag, dirty
false, build time 2026-10-19T20:45:30+0800, and origin
[email protected]:Bughue/seata.git. The plugin records the local clone.
It does not read the POM.

That origin is the personal fork of Minghua Xie (GitHub Bughue), a
Seata PPMC member and committer. The incubator status page records the
committer vote on 2024-10-15 as Minghua Xie (Bughue). The public LDAP
roster lists uid minghua as both a member and an owner of the seata
podling. The vote mail is from Minghua Xie, and the signing key is
[email protected]. That identifies the release manager. It does not
make [email protected]:Bughue/seata.git the Apache remote.

The staged POM names a different remote.
https://repository.apache.org/content/repositories/orgapacheseata-1286/org/apache/seata/seata-compressor-bzip2/2.7.0/seata-compressor-bzip2-2.7.0.pom
has connection scm:[email protected]:apache/incubator-seata.git. That
block is copied from build/pom.xml. It is not valid Maven SCM syntax.
The form is scm:<provider>:<url>, so this string is missing the
provider. The same block is not in the source zip's root pom.xml.

The release guide does not say to build from a personal fork:
https://seata.apache.org/docs/developers/release-guide/release-guide_dev
The ten workflow files under .github/workflows do not name that fork.
CONTRIBUTING.md tells a contributor to clone a personal fork as origin
and add upstream as [email protected]:apache/incubator-seata.git. That is
the contributor setup, not the release workflow. The guide's Maven
step is a local mvn clean deploy -Prelease -DskipTests. The page it
cites, https://infra.apache.org/publishing-maven-artifacts.html,
describes mvn release:perform, which checks out the SCM URL in the POM
and deploys that checkout.

Of the 355 jars in the binary tar, 308 are third-party libraries and
are not in 
https://repository.apache.org/content/repositories/orgapacheseata-1286/.
Forty-three Seata jar names are in both. Each staging copy has the
extra seata-git.properties. seata-console-2.7.0.jar,
seata-namingserver.jar, and seata-server.jar are in the binary tar and
not at the expected path in the repository. Inside the binary tar, the
two copies of seata-common-2.7.0.jar differ: the copy under
seata-namingserver/lib and the copy under seata-server/lib differ in
101 classes.

The other open Seata staging repositories show the same pattern in
seata-core-2.7.0.jar. None record the Apache remote. Only 1286 is the
vote commit.

- 1282, 1284, 1285, 1286: [email protected]:Bughue/seata.git
- 1278, 1280: /Users/mengmengsmac/git/seata/incubator-seata

Build host:

OS: Rocky Linux 9.8 (Blue Onyx)
Architecture: x86_64
vCPU and class: 16, c7i.4xlarge
Root disk: 99G observed, recipe default 100 GB
Login user: rocky
Java: OpenJDK 21.0.13, JAVA_HOME=/usr/lib/jvm/java-21-openjdk
AMI recipe: 
https://github.com/edespino/cloudberry-image-factory/tree/main/vm-images/aws/cloudberry/build/rocky9
Installed after launch: OpenJDK 21.0.13, license-eye 0.9.0. The Seata
compile did not use Go.

Not checked:

The CI cells this host cannot run were not run: four macOS cells, and
the Ubuntu JDK 11 / Spring 2.5.15 cell that starts Redis and Nacos.

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to