Closed TimoRoth closed 5 years ago
I can confirm this on linux, cc @ocefpaf
Seems there is indeed a problem with rtree. Only trying to install fiona and rtree gives:
joris@joris-XPS-13-9350:~/scipy$ conda create -n test_env -c conda-forge python=3.7 fiona rtree
Collecting package metadata: done
Solving environment: failed
UnsatisfiableError: The following specifications were found to be in conflict:
- rtree
Use "conda search <package> --info" to see the dependencies for each package.
although that if I try to install fiona and libspatialindex (the only dependency of rtree), then it works fine ..
@ocefpaf I see you changed the pinning of libspatialindex recently to 1.9: https://github.com/conda-forge/conda-forge-pinning-feedstock/commit/4b273715fe2c25a60288ceb54318ea61082ba90d#diff-8926f6e7808be490ec92ac38016ab868. But, if you just ask to install it, conda-forge gives you 1.8, and I also don't see any version 1.9 for libspatialindex.
It is probably related to the repo patch I applied earlier this week :unamused:
Looking into it...
@soapy1 do you know what may be going on here? Should we revert https://github.com/conda-forge/conda-forge-repodata-patches-feedstock/pull/14?
I can confirm this on linux, cc @ocefpaf
Can confirm this on Windows too.
so for some reason, I can't reproduce this issue and I get the 1.9 libspatialindex. If someone can test with setting channel_alias: https://conda-web.anaconda.org
in their .condarc. This will make conda get the unpatched repodata. So if that resolves the issue then conda-forge/conda-forge-repodata-patches-feedstock#14 should be reverted. Looking into this a bit more
so for some reason, I can't reproduce this issue and I get the 1.9 libspatialindex.
I just merged libspatialindex 1.9
but the old rtree
should be installable with libspatialindex 1.8
before I merged it :-/
Looking into this a bit more
Thanks!
Sooooo, it looks like the repodata hotfix is the culprit here. I think the pinning can be readjusted @ocefpaf
Sooooo, it looks like the repodata hotfix is the culprit here. I think the pinning can be readjusted @ocefpaf
I'll revert that PR but not sure how to fix the repo hotfix.
@TimoRoth and @gboeing can you please test that again? I did not really "fix it" I only removed the "bad fix" and pushed new packages. Hopefully things should work with libspatialindex 1.9.0
on Linux and macOS and libspatialindex 1.8.5
on Windows.
I'll try to push libspatialindex 1.9
on Windows over the weekend.
so, I think if the pin is changed to <1.9
or 1.8.*
it should be resolved. It seems like conda is trying to match exactly to 1.8 and that doesn't really work.
so, I think if the pin is changed to
<1.9
or1.8.*
it should be resolved. It seems like conda is trying to match exactly to 1.8 and that doesn't really work.
Got it. Thanks! However, libspatialindex
has only 2 children (rtree
and qgis
) so I'm fixing this one "by hand" now and leave repo data patch for the next time this happens.
A basic conda create -n test_env -c conda-forge python=3.7 geopandas
now works again!
A basic
conda create -n test_env -c conda-forge python=3.7 geopandas
now works again!
Thanks for checking (and sorry for yet another broken env). We'll learn how to do this eventually ;-p
Looks like it's working again on Windows. Thanks!
This is directly related to channel_priority=strict being set, without that, it manages to install, but ends up like described in issue #48 where libpoppler.so.76 is missing.
In another, already pre-existing environment, it gives out a bit more descriptive of an error: