Uh oh!
There was an error while loading. Please reload this page.
Fix conda include reporting [do not merge] - #2072
Conversation
* Add AutoWIG to list of binding generators
henryiii
commented
Jan 15, 2020
This fixed the problem with Conda here: conda-forge/pybind11-feedstock#43 |
henryiii
commented
Jan 15, 2020
The failure with CI is due to a ruby error in homebrew, and is unrelated. |
The thing I have not tested is that if a pip install in a conda environment works properly with this change. Edit: Yes, it seems to report correctly. |
Thank you for the report. As added here, I cannot reproduce the problem with conda (on conda-forge) that you report and this diff likely breaks the includes. Also note that upcoming releases got again a logic change in #1995 - maybe you are mismatching versions between development and latest stable with corresponding install logic locally? Maybe you are mixing a homebrew install with a local conda install to build upon? |
Yes, this is targeted at stable (which I didn't notice), it may be fixed in master. |
Got it, it's a conda-forge, Unix-only install problem since we passed a contradicting flag to CMake. Thank you for spotting this! ✨ |
wjakob
commented
Jun 30, 2020
The logic in |
henryiii
commented
Jul 1, 2020
I think so. I'll open an issue if anything is still broken. |
The Conda-forge package is reporting the wrong directory. From local testing, the standard way of accessing the directory seems to be working correctly; it's just the injected special logic that is causing issues.
Is the virtualenv logic needed, or could it be removed as well?
Testing this in Conda-forge before making the PR non-draft.