1.7.4 Patch Release.

[trivialfis]

This is a bug-fix release for 1.7.

• [R] fix OpenMP detection on macOS #8684
• Make sure input numpy array is aligned. #8690
• Fix different number of features in gpu_hist evaluator. #8754
• Fix GPU L1 error. #8749
• Fix feature types param #8772
• Fix ranking with quantile dmatrix and group weight. #8762
• Fix CPU bin compression with categorical data. #8809

[trivialfis]

R winbuilder result. It's failing on i386 but otherwise looks good.

Your package xgboost_1.7.4.1.tar.gz has been built (if working) and checked for Windows.
Please check the log files and (if working) the binary package at:
https://win-builder.r-project.org/GyaSK35w21NB
The files will be removed after roughly 72 hours.
Installation time in seconds: 703
Check time in seconds: 432
Status: OK
R version 4.2.2 (2022-10-31 ucrt)

Your package xgboost_1.7.4.1.tar.gz has been built (if working) and checked for Windows.
Please check the log files and (if working) the binary package at:
https://win-builder.r-project.org/TmDkKJPbkQ9G
The files will be removed after roughly 72 hours.
Installation time in seconds: 1412
Check time in seconds: 557
Status: 2 ERRORs, 1 WARNING, 1 NOTE
R version 4.1.3 (2022-03-10)

[trivialfis]

Uploaded all python packages, and submitted CRAN package.

[trivialfis]

Failed CRAN pre-test:

* checking C++ specification ... NOTE
  Specified C++14: please drop specification unless essential

[hcho3]

Let me upload the JVM packages

[trivialfis]

@hcho3 If you haven't, please don't upload anything yet.

[hcho3]

@trivialfis Is there any more modifications to make?

[trivialfis]

We might need a fix for #8824 , @RAMitchell might be able to share some insights when he's available.

I'm looking into #8816 , which is not reproducible on my machines (tried two different devices and docker). Based on the log from this debug PR, the thread local cache might be problematic even though I don't have any clue about it. The regression test is accessing a DMatrix constructed from a classification test ...

[trivialfis]

Let's skip #8816 for now. It's caused by scala using a gc thread for booster destruction, resulting in different threads for construction and destruction.

[marcphilipp]

Any updates on getting the JVM packages to Maven Central?

[trivialfis]

@hcho3 Please continue the 1.7.4 release. I will wrap up all needed changes for 2.0 as soon as possible. The 1.7.4 release branch is quite difficult to work with now with the CI upgraded on the master branch.

[hcho3]

I went ahead and uploaded 1.7.3 to Maven Central. https://repo1.maven.org/maven2/ml/dmlc/xgboost4j_2.12/1.7.4/