:construction_worker: (feat): Add ARM64 Builds by KrisJohnstone · Pull Request #1527 · apache/maven-mvnd

KrisJohnstone · 2026-02-05T02:22:12Z

Added support for ARM64 builds using glibc 2.17.

KrisJohnstone · 2026-02-05T02:28:32Z

Had a quick look, but this should resolve #921 and #888.

cstamas · 2026-03-04T17:53:10Z

@gzm55 can you take a peek please?

gzm55 · 2026-03-04T23:58:24Z

I chose 2.28 as RHEL8 and Ubuntu 18.04 LTS both still use this version afaik and are the oldest operating systems I know of that are still under support of some sort.

Could use 2.17 as that's when support for ARM64 was added from what I can see, but there were also test failures in versions 2.18 etc so figured best case is to choose a later version, which is when I settled on using the newest version supported by the oldest operating system.

I can't test the github actions from a fork but it should work, I've verified that the content of the steps works, its the conditionals on the steps I'm not 100% on.

for aarch64, I think it is ok to begin from glibc 2.28, but when someone need, we should accept the pr to lower the version to 2.17. @KrisJohnstone @cstamas
for x64, we should stick on glibc 2.12, not bump the dependency to 2.28. @KrisJohnstone

cstamas · 2026-03-06T12:55:02Z

Release is imminent (3.9.13 is out today). Any way to apply proposed changes to PR?

KrisJohnstone · 2026-03-06T12:57:42Z

It’s 11PM here will try to update and get it to you before I zzzz.

cstamas · 2026-03-06T12:58:43Z

No need to rush; I can wait 24h or more. I was just interested in "is it gonna happen" or not. 😄

KrisJohnstone · 2026-03-06T13:26:43Z

No need to rush; I can wait 24h or more. I was just interested in "is it gonna happen" or not. 😄

I've updated the PR, is that more in line with what you were thinking? I don't have access to work laptop atm so no easy access to arm64 boxes.

cstamas · 2026-03-06T13:29:57Z

Yes, this looks more in line with what summoned @gzm55 proposed:

leave linux/amd64 "as is" (as it was)
make linux/arm64 use glibc 2.28

cstamas · 2026-03-06T15:23:26Z

Will leave it here. Seems patching is not correct? Also, keep in mind to keep in sync the two workflows (early preview and release).

KrisJohnstone · 2026-03-07T03:57:42Z

Will leave it here. Seems patching is not correct? Also, keep in mind to keep in sync the two workflows (early preview and release).

It worked prior to me making the suggested changes. It could be due to referencing 2.2.5 in dynamic-libc-start.c and glibc.redef which is why I made the change to start with.

gzm55 · 2026-03-07T04:04:40Z

dynamic-libc-start.c

in this .c code, we can select different required GLIBC_ versions by checking arch macros, like

#if defined(__x86_64__)
...
#elif defined(__aarch64__)
...
#else
#  error("")
#endif

KrisJohnstone · 2026-03-07T17:18:19Z

This will need to wait till Monday, QEMU just isn't playing ball on my desktop. My work laptop died last week and pick new one up Monday when I will have access to arm64 based arch to run this without emulation.

cstamas · 2026-03-07T17:21:44Z

No problem, just ping here (or on #1500) if you are done or just want to punt on it for 1.0.4

cstamas · 2026-03-11T11:27:12Z

Maven 3.9.14 goes out tomorrow, and from that point nothing stops the mvnd 1.0.4 release. Please ping should we wait for this or not.

cstamas · 2026-03-13T11:44:39Z

Punting this from 1.0.4, that is about to be released.

cstamas · 2026-03-13T12:41:14Z

This PR is for Linux ARM but having Windows ARM would be nice thing. But, I don't think this "patching of glibc" applies to Windows ARM.... Origin and root issue #823

KrisJohnstone · 2026-03-16T04:10:22Z

First laptop was DOA. Back on this now :)

…m for arm64

KrisJohnstone · 2026-03-16T12:32:07Z

Had some other stuff that needed doing so mainly got solid time after work. This should be it, tested compiling on a amd64 and arm64 box.

Fun fact <family>linux</family> isn't actually a family, drove me nuts and went around in circles today a bit, it works due to a package maven depends on having an else clause that causes a fallback. (Ref: https://github.com/codehaus-plexus/plexus-utils/blob/4d25130f1df0540676ccc17edc2d18e01f0dbed8/src/main/java/org/codehaus/plexus/util/Os.java#L309)

cstamas · 2026-03-16T12:35:32Z

FYI, master uses Nisse:
https://github.com/apache/maven-mvnd/blob/master/.mvn/extensions.xml

It has OS detector like properties, we may backport Nisse usaget to 1.x build too....

Example of emitted (and hence, available) properties (take a peek at nisse.os. prefixed ones)
maveniverse/nisse#89

cstamas · 2026-03-16T13:06:53Z

@gzm55 can you take a peek? I am fine to make 1.0.5 with this single change to cover linux-arm as well.

cstamas · 2026-03-16T13:08:44Z

@KrisJohnstone that said, these are not the only changes needed... as this adds new release variant..... for example https://github.com/apache/maven-mvnd/blob/mvnd-1.x/.github/workflows/release.yaml

KrisJohnstone · 2026-03-16T13:23:53Z

@KrisJohnstone that said, these are not the only changes needed... as this adds new release variant..... for example https://github.com/apache/maven-mvnd/blob/mvnd-1.x/.github/workflows/release.yaml

Added.

gzm55 · 2026-03-17T00:26:26Z

@gzm55 can you take a peek? I am fine to make 1.0.5 with this single change to cover linux-arm as well.

@cstamas the pr LGTM, with only a typo

KrisJohnstone · 2026-03-17T03:52:11Z

@gzm55 can you take a peek? I am fine to make 1.0.5 with this single change to cover linux-arm as well.

@cstamas the pr LGTM, with only a typo

Fixed the only typo I saw, which was in client/pom.xml.

gnodet

Review

Overall, this PR is well-structured and follows the established patterns for glibc version pinning. The approach of targeting glibc 2.17 for aarch64 is sound (it's the earliest possible version for that architecture). Nice work!

Must Fix

BUG: Verification regex in `release.yaml` allows glibc up to 2.28, not 2.17

In release.yaml, the ARM64 verification step has:

objdump -T client/target/mvnd | grep GLIBC_ | grep -v 'GLIBC_\([01]\|2\.[0-9]\|2\.2[012345678]\)[^0-9]' || err=$?

This allows GLIBC versions 2.20 through 2.28 — far more permissive than the stated goal of glibc 2.17. The regex 2\.2[012345678] matches 2.20 to 2.28, which has nothing to do with 2.17.

Compare with the early-access.yaml version which is correct:

objdump -T client/target/mvnd | grep GLIBC_ | grep -v 'GLIBC_\([01]\|2\.[0-9]\|2\.1[017]\)[^0-9]' || err=$?

The release.yaml regex should match the early-access.yaml one. A release build could silently ship with symbols requiring glibc 2.18–2.28, defeating the purpose of the glibc-lowering work.

Suggestions (non-blocking)

Consider using 2\.1[0-7] instead of 2\.1[017] in the early-access regex to allow the full range 2.10–2.17 for robustness (any symbol versioned at 2.12–2.16 would also be acceptable).
A brief comment in glibc.redef.aarch64 noting that 2.17 is the minimum because it's the first glibc with aarch64 support would help future maintainers.

KrisJohnstone · 2026-03-17T04:17:03Z

Fixed.

gnodet

The release.yaml regex fix in the latest commit addresses the concern. LGTM.

Port ARM64 build support from mvnd-1.x PR #1527 to master: - Add ubuntu-24.04-arm runner for Linux ARM64 builds - Fix broken matrix exclude rules (macos-latest→macos-15, windows-latest→windows-2025) - Add glibc.redef.aarch64 pinning all symbols to GLIBC_2.17 - Add #if defined(__aarch64__) conditional in dynamic-libc-start.c - Add uname -m arch detection in gcc wrapper script - Add linux-aarch64-image-only-require-glibc-2.17 profile in client/pom.xml - Add ARM64 glibc patch/verify steps in CI workflows - Add linux-aarch64 release asset upload steps Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

KrisJohnstone force-pushed the arm64-builds branch from 392ee48 to 8790d37 Compare February 5, 2026 02:24

cstamas added this to the 1.0.4 milestone Mar 6, 2026

cstamas added the enhancement New feature or request label Mar 6, 2026

👷 (feat): Add ARM64 Builds

b69b486

KrisJohnstone force-pushed the arm64-builds branch from ed8627e to b69b486 Compare March 6, 2026 13:22

cstamas added 2 commits March 6, 2026 15:54

Split checks too

77f9549

Align release.yaml too

6daba52

KrisJohnstone force-pushed the arm64-builds branch from c3890f1 to 6b562a0 Compare March 7, 2026 04:38

cstamas mentioned this pull request Mar 12, 2026

Release 1.0.4 #1500

Closed

cstamas removed this from the 1.0.4 milestone Mar 13, 2026

This comment was marked as off-topic.

Sign in to view

feat: migrate to 24.04 to match x86. add missing profile to client.po…

339d7a7

…m for arm64

KrisJohnstone force-pushed the arm64-builds branch from c57e562 to 339d7a7 Compare March 16, 2026 12:12

👷 (chore): Add arm64 artifact upload to release pipeline

e4a25c4

cstamas requested a review from gnodet March 16, 2026 13:25

fix: typo in x86 client pom profile

31ed551

gnodet requested changes Mar 17, 2026

View reviewed changes

fix: missed changing regex in release.yaml

d112333

gnodet approved these changes Mar 17, 2026

View reviewed changes

cstamas added this to the 1.0.5 milestone Mar 17, 2026

cstamas merged commit bfa45f1 into apache:mvnd-1.x Mar 17, 2026
6 checks passed

github-actions bot assigned cstamas Mar 17, 2026

gnodet mentioned this pull request Mar 17, 2026

Add Linux ARM64 (aarch64) native build support #1561

Open

3 tasks

Conversation

KrisJohnstone commented Feb 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

KrisJohnstone commented Feb 5, 2026

Uh oh!

cstamas commented Mar 4, 2026

Uh oh!

gzm55 commented Mar 4, 2026

Uh oh!

cstamas commented Mar 6, 2026

Uh oh!

KrisJohnstone commented Mar 6, 2026

Uh oh!

cstamas commented Mar 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

KrisJohnstone commented Mar 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cstamas commented Mar 6, 2026

Uh oh!

cstamas commented Mar 6, 2026

Uh oh!

KrisJohnstone commented Mar 7, 2026

Uh oh!

gzm55 commented Mar 7, 2026

Uh oh!

KrisJohnstone commented Mar 7, 2026

Uh oh!

cstamas commented Mar 7, 2026

Uh oh!

cstamas commented Mar 11, 2026

Uh oh!

cstamas commented Mar 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment was marked as off-topic.

cstamas commented Mar 13, 2026

Uh oh!

KrisJohnstone commented Mar 16, 2026

Uh oh!

KrisJohnstone commented Mar 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cstamas commented Mar 16, 2026

Uh oh!

cstamas commented Mar 16, 2026

Uh oh!

cstamas commented Mar 16, 2026

Uh oh!

KrisJohnstone commented Mar 16, 2026

Uh oh!

gzm55 commented Mar 17, 2026

Uh oh!

KrisJohnstone commented Mar 17, 2026

Uh oh!

gnodet left a comment

Choose a reason for hiding this comment

Review

Must Fix

BUG: Verification regex in release.yaml allows glibc up to 2.28, not 2.17

Suggestions (non-blocking)

Uh oh!

KrisJohnstone commented Mar 17, 2026

Uh oh!

gnodet left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

KrisJohnstone commented Feb 5, 2026 •

edited

Loading

cstamas commented Mar 6, 2026 •

edited

Loading

KrisJohnstone commented Mar 6, 2026 •

edited

Loading

cstamas commented Mar 13, 2026 •

edited

Loading

KrisJohnstone commented Mar 16, 2026 •

edited

Loading

BUG: Verification regex in `release.yaml` allows glibc up to 2.28, not 2.17