Support for ARM, updated PDFAlto, Docker multi-architecture build

[lfoppiano]

Finally I nailed it!

This PR provide a multi-architecture build for amd64/arm64. I've switched to the eclipse-adoptium-17-jdk/jre but, for the moment, I'm using ubuntu 20.04 (focal) instead of ubuntu 22. For now Is limited to the manual build for the CRF only image.

This requires some tests, here here here the resulting image.

This PR should help to solve #1089, #928, #1014, #1119

Update: At the moment the ARM support for the Deep Learning image is not working natively because there is no linux/arm64 docker image of the base images (even of tensorflow/tensorflow:2.17.0)

[coveralls]

coverage: 40.395% (-0.03%) from 40.421%
when pulling a3c814d on multi-arch-docker-image
into f7596b3 on master.

[heijligers]

I'm getting

rosetta error: failed to open elf at /lib64/ld-linux-x86-64.so.2

when trying to run on macos m1

[lfoppiano]

@heijligers which image did you try?

This one was correctly built with multi arch and arm support: lfoppiano/grobid:latest-crf-multi-arch but I did not manage to get it to work in a stable way.

My next step was to try to solve this issue with spawning process by looking on a way to have a conditional decision in the Dockerfile

Any help is welcome for this issue 🙂

[AaronNGray]

I am currently working on this for the default CRF Docker (Docker.crf) .

This needs work :-

• ARM64 builds of a number of libraries
• generalizing for different architectures
• releases implementing for all the various sub project repos involved

There seems a number of libraries involved :-

• tini - https://github.com/krallin/tini
• pdfalto - https://github.com/kermitt2/pdfalto
• libwapiti - https://github.com/kermitt2/wapiti
• libcrfpp - https://github.com/taku910/crfpp/
• libjip - https://github.com/ninia/jep

tini

https://github.com/krallin/tini
There is a current release of tini for ARM64 :-
https://github.com/krallin/tini/releases/download/v0.19.0/tini-arm64
I have tried this and it works.

pdfalto

https://github.com/kermitt2/pdfalto
This builds fine on x64, and armhf, but has a compile overload call issue with ICU C++.
#1230 (comment)

libwapiti

https://github.com/kermitt2/wapiti

libcrfpp

https://github.com/taku910/crfpp/

libjep

https://github.com/ninia/jep
libjep is a library for Embed Python in Java it involved JNI. This potentially is more complex depending upon the ARM64 JNI support.

I will update this comment as a tracking comment as I make progress.

[AaronNGray]

@lfoppiano It seems the way to deal with multi platform and multi architecture OS/ARCH builds is with build and buildx :-

docker build --platform linux/amd64,linux/arm64 -t */*:* .

docker buildx build \
--push \
--platform linux/arm/v7,linux/arm64/v8,linux/amd64 \ --tag <username>/<repo>:<tag> .

https://github.com/docker/buildx?tab=readme-ov-file#building-multi-platform-images

This should work across multiple docker images :-

https://github.com/AaronNGray/docker-arch-test

[lfoppiano]

@AaronNGray yes, this is what I believe I did to build the image with github actions, however I did not succeed in having a running docker image for Apple M1.

[AaronNGray]

Good going. I posted some stuff to check regarding ABI's and compilers on Grobid-core Discord, it maybe obvious but may need checking.

[coveralls]

Pull Request Test Coverage Report for Build 21920119213

Details

• 0 of 0 changed or added relevant lines in 0 files are covered.
• 22 unchanged lines in 4 files lost coverage.
• Overall coverage increased (+0.06%) to 38.291%

---

Files with Coverage Reduction	New Missed Lines	%
org/grobid/core/engines/Segmentation.java	1	43.25%
org/grobid/core/document/Document.java	2	72.67%
org/grobid/core/layout/Block.java	4	68.18%
org/grobid/core/document/BasicStructureBuilder.java	15	29.98%

Totals
Change from base Build 21564556488:	0.06%
Covered Lines:	17274
Relevant Lines:	42503

---

💛 - Coveralls

[AaronNGray]

macos-13 is no longer supported by GitHub build system.

[lfoppiano]

I know, I've fixed in both pdfalto and wapiti, and here! :-)