FTBFS on hppa - undefined symbols

[malaterre]

Let see if I can get input from you guys on the following issue we have within Debian on hppa arch.

It turns out that the symbol black magic introduced in 1.3.1 makes the compilation of libkcapi fails on hppa arch:

• https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1003088

It looks like the symbols that are messing the builds are the ones duplicated from one version to another, for example: kcapi_cipher_stream_init_enc which is either:

kcapi_cipher_stream_init_enc@LIBKCAPI_0.12.0
kcapi_cipher_stream_init_enc@LIBKCAPI_1.3.1

[malaterre]

In particular compare the hppa output of:

malat@panama ~/libkcapi-1.3.0 % nm -D ./.libs/libkcapi.so | grep kcapi_aead_stream_init_enc
00008688 T kcapi_aead_stream_init_enc@@LIBKCAPI_0.12.0

vs:

malat@panama ~/libkcapi-1.3.1 % nm -D ./.libs/libkcapi.so | grep kcapi_aead_stream_init_enc
         U kcapi_aead_stream_init_enc

---

For comparison on my amd64 system:

libkcapi-1.3.0  % nm -D ./.libs/libkcapi.so | grep kcapi_aead_stream_init_enc
0000000000009700 T kcapi_aead_stream_init_enc@@LIBKCAPI_0.12.0

vs

libkcapi-1.3.1 % nm -D ./.libs/libkcapi.so | grep kcapi_aead_stream_init_enc
000000000000a271 T kcapi_aead_stream_init_enc@LIBKCAPI_0.12.0
000000000000a22c T kcapi_aead_stream_init_enc@@LIBKCAPI_1.3.1

[malaterre]

@simo5 any suggestions ?

[simo5]

sounds like hppa tooling is failing to understand how to deal with symbol vesioning correctly?
what compiler version is this, what linker?

[malaterre]

Staring at the log file:

• https://buildd.debian.org/status/fetch.php?pkg=libkcapi&arch=hppa&ver=1.3.1-1&stamp=1641233787&raw=0

I see:

gcc-11_11.2.0-13

and

binutils-common_2.37-10 binutils-hppa-linux-gnu_2.37-10

[malaterre]

If I cherry-pick commit 71d80bc, the build on hppa succeeds...

[malaterre]

I'll cherry-pick also 3e9a149 just in case and see how it goes on debian buildds machines.

[simo5]

Sounds like you should pull in some gcc expert to see if there are bugs that are specific to hppa. I haven't touched that arch in more than a decade, and I am not aware of differences with it that we need to account for.
Are we sure there are no debian patches that are breaking this?

[smuellerDD]

Am Freitag, 28. Januar 2022, 16:49:40 CET schrieb Mathieu Malaterre: Hi Mathieu,

If I cherry-pick commit 71d80bc, the build on hppa succeeds...

If you say that these patches in the master branch are fixing your issue, then I will see to release a new version so that we have an official version with the fixes. I am a bit behind with the review of a few patches. Once I completed that, I would certainly cut a new release. Ciao Stephan

[simo5]

If I cherry-pick commit 71d80bc, the build on hppa succeeds...

Ah sorry, did not realize these commits hadn't made a release yet, as we applied them downstream in Fedora as well.

[malaterre]

@simo5 the description of 71d80bc refer to the use of LTO which is not my case on hppa (or at least I could not see an explicit reference to flto in the build log)

[malaterre]

FYI, I forwarded the issue upstream:

• https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104363

I believe this is safe now to close this issue. Thanks all.