ci: add big-endian (s390x) build and test workflow via QEMU by brtnfld · Pull Request #6367 · HDFGroup/hdf5

brtnfld · 2026-04-15T16:11:40Z

Summary

Adds .github/workflows/bigendian.yml: a reusable workflow_call job that builds and tests HDF5 on Alpine Linux s390x (IBM Z, big-endian) using jirutka/setup-alpine with QEMU user-space emulation
Wires it into call-workflows.yml as call-release-cmake-bigendian alongside the existing i386 job
Includes a sanity assertion (python3 -c "assert sys.byteorder == 'big'") to fail fast if endianness is wrong

Design notes

Modeled directly after the existing i386.yml pattern — same action, same pinned hash, same CMake flags
Fortran, Java, C++, ZLIB, SZIP, and plugin support are disabled to keep the emulated build lean; the goal is endianness correctness, not full feature coverage
Parallelism capped at --parallel 2 / ctest --parallel 2 since s390x emulation is slower than native x86

Test plan

Verify the bigendian.yml job appears in the Actions tab on a push/PR to develop
Confirm the "Verify big-endian" step prints Confirmed big-endian
Confirm build and ctest pass without failures

Adds bigendian.yml using jirutka/setup-alpine with arch: s390x to run builds and tests under an Alpine Linux s390x environment (big-endian). Wires it into call-workflows.yml alongside the existing i386 job.

brtnfld · 2026-04-15T19:00:58Z

It takes too long to run for a PR check; it should be moved to a daily test.

test/API/H5_api_dataset_test.c: H5Tenum_insert requires a value pointer matching the base type size (int/uint32_t = 4 bytes). Passing &i where i is size_t (8 bytes) causes the function to read the high 4 bytes on big-endian systems, which are always zero for small i, resulting in duplicate enum member errors. Fix by using a properly-typed local variable for both the native (int) and non-native (uint32_t) enum inserts. bigendian.yml: add --timeout 1800 to ctest to accommodate the slower execution speed of QEMU s390x emulation.

This test consistently hits the 1800s per-test ceiling under QEMU s390x emulation (the h5dump command itself returns 0; the image dataset creation is simply too slow). Exclude it rather than inflating the timeout further.

hyoklee · 2026-04-16T02:07:04Z

	2600 - FORTRAN_testhdf5_fortran (Failed)

Crashes before any output due to missing libquadmath.so on Alpine s390x (REAL(16) is compiled in but libquadmath is not available as a package). To be investigated separately.

hyoklee · 2026-04-16T21:24:20Z

@brtnfld , can you patch this and include image test?

--- a/HDF5Examples/C/HL/tfiles/h5ex_image2BE.ddl
+++ b/HDF5Examples/C/HL/tfiles/h5ex_image2BE.ddl
@@ -33878,7 +33878,7 @@ GROUP "/" {
       }
    }
    DATASET "image8bit" {
-      DATATYPE  H5T_STD_U8LE
+      DATATYPE  H5T_STD_U8BE
       DATASPACE  SIMPLE { ( 400, 300 ) / ( 400, 300 ) }
       DATA {
       (0,0): 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
@@ -41397,7 +41397,7 @@ GROUP "/" {
       }
    }
    DATASET "palette" {
-      DATATYPE  H5T_STD_U8LE
+      DATATYPE  H5T_STD_U8BE
       DATASPACE  SIMPLE { ( 256, 3 ) / ( 256, 3 ) }
       DATA {
       (0,0): 0, 0, 255,

hyoklee

2 more changes are desired:

See #6367 (comment) .
Use uraimo/run-on-arch-action instead of jirutka/setup-alpine. Then, fortran test will also pass.

Gemini's explanation:

Recent issues in the jirutka/setup-alpine repository suggest that while s390x is supported, there have been sporadic "segmentation faults" and "Illegal instruction" errors tied to how it installs apk-tools during the initial setup on certain architectures.

Recommendation
If you need stability for s390x: Stick with uraimo/run-on-arch-action. Its Docker-based approach is better at isolating the emulation layer from the host's quirks.

If you need speed and only use x86/ARM: jirutka/setup-alpine is significantly faster because it doesn't have the overhead of Docker image layers, but it is more sensitive to the host environment.

… tests Replace jirutka/setup-alpine with uraimo/run-on-arch-action for more stable s390x QEMU emulation via Docker. Enable FORTRAN_testhdf5_fortran by removing it from the ctest exclusion list.

…22.04

brtnfld · 2026-04-22T21:32:54Z

QEMU instability on GitHub-hosted runners seems to be an issue. Neither Alpine nor Ubuntu made a difference.

At this point, uraimo/run-on-arch-action is not reliable enough for compiling HDF5 on s390x via QEMU on these runners.

hyoklee · 2026-04-23T16:51:12Z

I see a red again if you revert to alpine after 2 greens with uraimo.

The h5ex_image2BE.ddl expected output was updated to H5T_STD_U8BE in eb535be, so there is no longer a need to skip H5_H5DUMP-h5ex_image2.

…an-s390x

Temporary: allows the workflow to run automatically on push to this branch (and provides a Release fallback when build_mode input is absent). Remove the push trigger before merging.

ci: add big-endian (s390x) build and test workflow via QEMU

8cab757

Adds bigendian.yml using jirutka/setup-alpine with arch: s390x to run builds and tests under an Alpine Linux s390x environment (big-endian). Wires it into call-workflows.yml alongside the existing i386 job.

brtnfld requested review from bmribler, derobins, fortnern, glennsong09, jhendersonHDF, lrknox, mattjala, qkoziol and vchoi-hdfgroup as code owners April 15, 2026 16:11

github-project-automation Bot added this to HDF5 - TRIAGE & TRACK Apr 15, 2026

github-project-automation Bot moved this to To be triaged in HDF5 - TRIAGE & TRACK Apr 15, 2026

brtnfld assigned hyoklee and lrknox Apr 15, 2026

brtnfld added 2 commits April 15, 2026 14:09

hyoklee requested changes Apr 15, 2026

View reviewed changes

Comment thread .github/workflows/bigendian.yml Outdated

github-project-automation Bot moved this from To be triaged to In progress in HDF5 - TRIAGE & TRACK Apr 15, 2026

ci(bigendian): enable Fortran build on s390x

6ab2c5e

ci(bigendian): skip FORTRAN_testhdf5_fortran on s390x

d7f6bae

Crashes before any output due to missing libquadmath.so on Alpine s390x (REAL(16) is compiled in but libquadmath is not available as a package). To be investigated separately.

hyoklee previously approved these changes Apr 16, 2026

View reviewed changes

ci(bigendian): skip s390x job on pull requests (too slow)

09ca092

brtnfld dismissed hyoklee’s stale review via 09ca092 April 16, 2026 18:53

hyoklee approved these changes Apr 16, 2026

View reviewed changes

bmribler previously approved these changes Apr 16, 2026

View reviewed changes

hyoklee requested changes Apr 17, 2026

View reviewed changes

Merge branch 'develop' into feature/ci-bigendian-s390x

e66689d

brtnfld added 2 commits April 22, 2026 10:50

updated to LE to BE in test files

eb535be

ci(bigendian): switch to uraimo/run-on-arch-action and enable Fortran…

2c4a00d

… tests Replace jirutka/setup-alpine with uraimo/run-on-arch-action for more stable s390x QEMU emulation via Docker. Enable FORTRAN_testhdf5_fortran by removing it from the ctest exclusion list.

brtnfld dismissed bmribler’s stale review via 2c4a00d April 22, 2026 15:54

brtnfld added 6 commits April 22, 2026 10:55

Merge branch 'develop' into feature/ci-bigendian-s390x

07251dd

ci(bigendian): add workflow_dispatch to allow manual triggering

e02979e

ci: trigger bigendian workflow

ba25c9b

ci(bigendian): reduce build parallelism to 1 to avoid QEMU segfaults

fc06f98

ci(bigendian): switch to ubuntu22.04 distro for more stable QEMU s390x

5865cd7

ci(bigendian): install cmake via pip to get version >= 3.26 on ubuntu…

700cf3d

…22.04

ci(bigendian): revert to jirutka/setup-alpine (uraimo QEMU too unstable)

4c802de

brtnfld added 3 commits May 19, 2026 13:45

ci(bigendian): re-enable image2 test now that DDL uses H5T_STD_U8BE

1011817

The h5ex_image2BE.ddl expected output was updated to H5T_STD_U8BE in eb535be, so there is no longer a need to skip H5_H5DUMP-h5ex_image2.

Merge remote-tracking branch 'origin/develop' into feature/ci-bigendi…

11551f0

…an-s390x

ci(bigendian): add push trigger on feature branch for testing

db22ae0

Temporary: allows the workflow to run automatically on push to this branch (and provides a Release fallback when build_mode input is absent). Remove the push trigger before merging.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

ci: add big-endian (s390x) build and test workflow via QEMU#6367

ci: add big-endian (s390x) build and test workflow via QEMU#6367
brtnfld wants to merge 19 commits into
HDFGroup:developfrom
brtnfld:feature/ci-bigendian-s390x

brtnfld commented Apr 15, 2026 •

edited

Loading

Uh oh!

brtnfld commented Apr 15, 2026

Uh oh!

Uh oh!

hyoklee commented Apr 16, 2026

Uh oh!

hyoklee commented Apr 16, 2026

Uh oh!

hyoklee left a comment •

edited

Loading

Uh oh!

brtnfld commented Apr 22, 2026

Uh oh!

hyoklee commented Apr 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

brtnfld commented Apr 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Design notes

Test plan

Uh oh!

brtnfld commented Apr 15, 2026

Uh oh!

Uh oh!

hyoklee commented Apr 16, 2026

Uh oh!

hyoklee commented Apr 16, 2026

Uh oh!

hyoklee left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

brtnfld commented Apr 22, 2026

Uh oh!

hyoklee commented Apr 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

brtnfld commented Apr 15, 2026 •

edited

Loading

hyoklee left a comment •

edited

Loading