Dear Étienne,

For future reference, I managed to identify the root cause of the
issue by changing the structure of "run-dicom-image-transcode-test" so
as to get the full log of Orthanc [1], which explains the intermediate
release "orthanc-wsi_3.2+dfsg-6" yesterday. Then, inspecting the
detailed logs, I observed the following error line for all the arm64,
riscv64, ppc64el and s390x architectures [2]:

1549s E0427 22:47:24.020832 MAIN wsi:/Plugin.cpp:492] Exception in
startup tests: Internal error

Tracking this line in the upstream source code [3], it appears that
the issue was related to a different behavior in floating-point
arithmetic on these specific platforms. I thus created a patch to
relax the error as a warning [4], which led to
"orthanc-wsi_3.2+dfsg-7".

This patch makes autopkgtest succeed (note however that riscv64 is
still pending at the time of writing). Inspecting the new logs in the
Debian continuous integration [5], I then got:

[arm64] 103s W0428 06:26:25.965016 MAIN wsi:/Plugin.cpp:489] LABColor
test: (0,7.56979e-06,127) vs. (0,0,127)
[ppc64el] 130s W0428 03:57:26.923464 MAIN wsi:/Plugin.cpp:489]
LABColor test: (0,7.56979e-06,127) vs. (0,0,127)
[s390x]  85s W0428 04:07:34.147764 MAIN wsi:/Plugin.cpp:489] LABColor
test: (0,7.56979e-06,127) vs. (0,0,127)

As can be seen, the floating-point error on the "A" component
(7.56979e-06 vs. expected 0) is negligible, so the error was a false
alarm, and I consider that the patch solves this issue.

The upstream code has been relaxed accordingly, so the patch will not
be necessary in future releases [6]. I hope that this investigation at
last unblocks the dcmtk migration to 3.6.9-5 in testing.

Kind Regards,
Sébastien-

[1] 
https://salsa.debian.org/med-team/orthanc-wsi/-/commit/8ef626741c542a463dd15f0e82797a66a3f5e912
[2] https://ci.debian.net/packages/o/orthanc-wsi/testing/riscv64/60287848/
[3] 
https://orthanc.uclouvain.be/hg/orthanc-wsi/file/OrthancWSI-3.2/ViewerPlugin/Plugin.cpp#l473
[4] 
https://salsa.debian.org/med-team/orthanc-wsi/-/commit/56c5003abaa66c26b996f6ac28a682be0a05a0d2
[5] https://ci.debian.net/packages/o/orthanc-wsi/unstable/arm64/60294240/
[6] https://orthanc.uclouvain.be/hg/orthanc-wsi/rev/0045d4f23eba


On Thu, 24 Apr 2025 at 22:51, Étienne Mollier <emoll...@debian.org> wrote:
> We're almost there with getting dcmtk and the Orthanc ecosystem
> to migrate to trixie.  The remaining issues only affect
> orthanc-wsi, and on a particular set of CPU architectures
> (arm64, riscv64, ppc64el and s390x).

Reply via email to