enable threadsafe and production

[marqh]

we use hdf5 in threaded processing, which can lead to segfaults if hdf is built without the threadsafe flag and the production flag

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[jakirkham]

Does this result in an SONAME change?

[marqh]

Does this result in an SONAME change?

I do not know, I am afraid, will the CI testing give us a clue on that?

[marqh]

checking for thread safe support... configure: error: --enable-cxx and --enable-threadsafe flags are incompatible. Use --enable-unsupported to override this error.
Command failed: /bin/bash -x -e /feedstock_root/build_artefacts/work/hdf5-1.8.17/conda_build.sh

I'm not sure how to proceed here.
what does enable-cxx do for us?
are we content with enable-unsupported?

[jakirkham]

We probably can figure it out from the logs. Just noting that if it does, we need to get everything organized for rebuilding with a new HDF5 (assuming this change is agreeable). That will be a fair bit of working considering how important HDF5 is to conda-forge.

If it does, I don't think this should be a priority as Travis CI is changing images on us in less than a month. ( conda-forge/conda-forge.github.io#249 ) Also we are still on conda-build 1 and it is causing us all sorts of grief. So we need to switch to conda-build 2 soon. These two IMHO are higher priority.

This isn't to say that we won't do this at some point. Just that this month already has too many big ticket items for us to provide any promise of getting this done too.

cc @pelson

[jakirkham]

That all being said, this warning is troubling to me. I rely on C++ support from HDF5.

checking for thread safe support... configure: error: --enable-cxx and --enable-threadsafe flags are incompatible. Use --enable-unsupported to override this error.

xref: https://circleci.com/gh/conda-forge/hdf5-feedstock/130

[marqh]

cc @bjlittle

[marqh]

@jakirkham

thank you for the feedback. This is a blocking issue for our use of conda-forge, so we'll need to make other arrangements whilst this gets discussed; all updates are appreciated

[jakirkham]

Sorry to hear that. Unfortunately there is already a full docket of issues that we are trying to address and we need to prioritize them. I hope you understand.

[marqh]

i have checked a local build process for hdf5 and it explicitly sets

--disable-static \
    --enable-linux-lfs --with-zlib --with-ssl \
    --enable-threadsafe --with-pthread=yes \
    --enable-production --enable-cxx \
    --enable-unsupported

so we've got prior with enable-unsupported

[jakirkham]

If you want to help with those other issues that are slowing this down, I think that will make it much easier to get back to this sooner. The Travis CI change is being discussed in issue ( conda-forge/conda-forge.github.io#249 ). Also much of the conda-build 2 switch is being discussed in issue ( conda-forge/conda-forge.github.io#171 ). In particular, we have encountered issues with cmake and conda-build 2. Much more detail on this latter point is in issue ( conda-forge/cmake-feedstock#21 ). I understand if you don't have time to sink into these issues. Just providing you info about them in case you are willing to lend a hand on these org wide problems.

[gillins]

I think the C++ part has no thread safe support hence the need to specify --enable-unsupported to ensure you know that only the C part of the library is thread safe.

[marqh]

i've pushed a build with --enable-unsupported to see if this enables all the builds to succeed

for discussion...

[marqh]

i've pushed a build with --enable-unsupported to see if this enables all the builds to succeed

hmm, so progress; i get a build, but the hdf tests fail, on CircleCI

[marqh]

updated to disable-static and with-ssl to see if this enables the testing to pass for Linux

[ryanvolz]

I was all set to make this very pull request when I just saw that one already existed. I've been tracking down a segfault running some of our HDF5-using code in a conda environment, and I finally figured out it was because we needed a threadsafe HDF5 build.

I made the change and built it locally, and it seems to be working fine. The resulting SONAMEs are the same. I didn't rebuild anything I'm using that uses HDF5 and it still works.

I will note that Debian/Ubuntu build HDF5 with --enable-threadsafe and --enable-unsupported, the latter presumably because it is necessary to be able to keep --enable-cxx and the high-level library.