Clean up C build dependencies (and bump Qiskit) #32
Conversation
* Remove muparserx which is not needed as of version 0.13 * Ignore run exports from spdlog. Only its headers are used at build time. * Make numpy a run dependency only. Qiskit Aer interfaces with numpy through the pybind11 headers and does not link directly to numpy. * Make the nlohmann_json bounds consistent across environemnts instead of special for osx arm64. * Increase qiskit minimum version
…nda-forge-pinning 2024.02.01.14.05.18
|
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 ( |
|
@leofang I am curious what you think about these changes. The warnings in the build log about DSO's and exports made me look loser at the linking of the C extension. I noticed that muparserx was dropped in the last release, so that was an easy change (as was realizing that I missed bumping the minimum qiskit version). I also noticed that the extension does not link against the spdlog library. I think it only uses the headers at build time and doesn't need spdlog at run time. I also noticed that the extension does not link against numpy and looking at the pybind11 documentation I read that pybind has its own numpy headers and does not need numpy at build time. I am a little concerned about numpy 2 based on this issue, but I think the pybind11 feedstock should set some kind of export on numpy if that is a problem rather than bounding numpy here. However, maybe just depending on numpy is more standard for getting the right run constraints any way for conda-forge? I also bumped up the nlohmann_json version -- I just thought if we are using something different from upstream for one variant and it is working okay we might as well use it for all the variants. |
| @@ -23,6 +23,8 @@ source: | |||
| # This patch was accepted in https://github.com/Qiskit/qiskit-aer/pull/1940 | |||
| # and can be removed from the receipe with the 0.14 release. | |||
| - 0001-windows-find-blas.patch | |||
| # muparserx usage was removed in https://github.com/Qiskit/qiskit-aer/pull/1884 | |||
| - 0001-Remove-muparserx-from-cmake-configuration.patch | |||
There was a problem hiding this comment.
The patch is in Qiskit/qiskit-aer#2045. I hadn't opened that when I pushed here or I would have put the link a comment. I don't want to churn through the whole CI for that, but I will put it in if I do another push.
|
Thanks, Will, a bit overwhelmed this week, will try to take a close look asap. |
|
Thanks! No rush. The changes seem to pass the tests at least. |
|
@wshanks feel free to resolve the conflicts and merge |
|
@conda-forge-admin, please rerender |
|
When I rerendered locally, all the CUDA 12 builds got removed, so I am trying rerendering with the bot here. |
…nda-forge-pinning 2024.03.29.12.36.21
This was missed before and being pulled in by Qiskit (which recently dropped it as a dependency).
The pybind11 issue (pybind/pybind11#4606) requiring a deviation from the upstream migration in conda-forge-pinning-feedstock should be fixed in the latest release (2.12).
…nda-forge-pinning 2024.04.05.13.18.18
Checklist
0(if the version changed)conda-smithy(Use the phrase@conda-forge-admin, please rerenderin a comment in this PR for automated rerendering)