I think this demonstrates one drawback of conda-forge right now: it's best to go 'all in' on conda-forge for all packages or you run the risk of a broken environment.
In this Mac example, I think the issue may just be due to minor discrepancies between the recipes on the conda-forge and rdkit channels, so it would hopefully just be a short-term problem while people transition and the rdkit channel is phased out. However, on linux, conda-forge currently uses older compilers than the defaults channel, so there may be ongoing issues with mixing conda-forge and defaults packages. They are currently migrating the entire conda-forge ecosystem to the new compilers (RDKit packages are actually already done, hidden under the "gcc7" label, waiting to be activated when all other packages are ready) so this will hopefully be resolved soon too. To use conda-forge for the time being, I'd actually recommend adding conda-forge as the highest priority channel in your config (conda config --add channels conda-forge) and then install into a clean environment (e.g. conda create -n rdenv rdkit). But once the compiler migration is complete, I'd hope that it would be relatively safe to install with "conda install -c conda-forge rdkit" into an existing environment with other packages from a mix of different channels. Matt On Oct 18, 2018, at 06:41 PM, "Peter St. John" <peterc.stj...@gmail.com> wrote: So I'm certainly fine with a conda-forge only option, I don't happen to use the rdkit channel version. But I'll point out that I'm in the minority: comparing downloads of recent versions on https://anaconda.org/rdkit/rdkit/files vs https://anaconda.org/conda-forge/rdkit/files shows that the rdkit channel is definitely used more frequently. One concern: on a mac when I uninstall rdkit from the rdkit channel and subsequently do a `conda install -c conda-forge rdkit`, I'll get a non-working rdkit installation. I.e., >>> import rdkit.Chem throws a SegFault. That also appears to happen if you run conda create -n rdkit_test conda-forge::rdkit. (see attachment) So it appears that for the time being, mac conda-forge builds are going to be best done as fresh installs with a bunch of the added conda-forge dependencies: $ conda create -n rdkit_test -c conda-forge rdkit The following NEW packages will be INSTALLED: blas: 1.0-mkl boost: 1.68.0-py36h3e44d54_1 conda-forge boost-cpp: 1.68.0-h3a22d5f_0 conda-forge bzip2: 1.0.6-1 conda-forge ca-certificates: 2018.10.15-ha4d7672_0 conda-forge cairo: 1.14.12-h276e583_5 conda-forge certifi: 2018.10.15-py36_1000 conda-forge clangdev: 6.0.1-default_1 conda-forge fontconfig: 2.13.1-hce039c3_0 conda-forge freetype: 2.9.1-h6debe1e_4 conda-forge gettext: 0.19.8.1-h1f1d5ed_1 conda-forge glib: 2.56.2-h464dc38_0 conda-forge icu: 58.2-hfc679d8_0 conda-forge intel-openmp: 2019.0-118 jpeg: 9c-h470a237_1 conda-forge libcxx: 6.0.1-0 conda-forge libffi: 3.2.1-hfc679d8_5 conda-forge libgfortran: 3.0.1-h93005f0_2 libiconv: 1.15-h470a237_3 conda-forge libpng: 1.6.35-ha92aebf_2 conda-forge libtiff: 4.0.9-he6b73bb_2 conda-forge libxml2: 2.9.8-h422b904_5 conda-forge llvm-meta: 6.0.1-0 conda-forge llvmdev: 6.0.1-h2d50403_2 conda-forge mkl: 2019.0-118 mkl_fft: 1.0.6-py36_0 conda-forge mkl_random: 1.0.1-py36_0 conda-forge ncurses: 6.1-hfc679d8_1 conda-forge numpy: 1.15.2-py36h6a91979_1 numpy-base: 1.15.2-py36h8a80b8c_1 olefile: 0.46-py_0 conda-forge openssl: 1.0.2p-h470a237_1 conda-forge pandas: 0.23.4-py36hf8a1672_0 conda-forge pcre: 8.41-hfc679d8_3 conda-forge pillow: 5.3.0-py36hc736899_0 conda-forge pip: 18.1-py36_1000 conda-forge pixman: 0.34.0-h470a237_3 conda-forge pycairo: 1.17.1-py36h4d1f301_0 conda-forge python: 3.6.6-h5001a0f_0 conda-forge python-dateutil: 2.7.3-py_0 conda-forge pytz: 2018.5-py_0 conda-forge rdkit: 2018.03.4-py36h557c172_1000 conda-forge readline: 7.0-haf1bffa_1 conda-forge setuptools: 40.4.3-py36_0 conda-forge six: 1.11.0-py36_1001 conda-forge sqlite: 3.25.2-hb1c47c0_0 conda-forge tk: 8.6.8-ha92aebf_0 conda-forge wheel: 0.32.1-py36_0 conda-forge xz: 5.2.4-h470a237_1 conda-forge zlib: 1.2.11-h470a237_3 conda-forge -- Peter On Thu, Oct 18, 2018 at 11:02 AM Thomas Holder <thomas.hol...@schrodinger.com> wrote: Note that it makes a difference whether you do "conda install -c conda-forge rdkit" or "conda install conda-forge::rdkit". The latter will pull in less packages from conda-forge. I'm on Linux, using a somewhat recent environment which has mostly default channel packages + rdkit::rdkit + openbabel::openbabel + schrodinger::pymol. This is what I get: ------------------------------------------------------------------------------ $ conda install conda-forge::rdkit The following packages will be downloaded: package | build ---------------------------|----------------- boost-cpp-1.66.0 | 1 18.8 MB conda-forge boost-1.66.0 | py36_1 317 KB conda-forge pycairo-1.15.4 | py36h1b9232e_1 67 KB rdkit-2018.03.4 | py36ha4bbe77_0 19.4 MB conda-forge ------------------------------------------------------------ Total: 38.5 MB The following NEW packages will be INSTALLED: boost: 1.66.0-py36_1 conda-forge boost-cpp: 1.66.0-1 conda-forge pycairo: 1.15.4-py36h1b9232e_1 The following packages will be DOWNGRADED: rdkit: 2018.03.4.0-py36h71b666b_1 rdkit --> 2018.03.4-py36ha4bbe77_0 conda-forge ------------------------------------------------------------------------------ $ conda install -c conda-forge rdkit The following packages will be downloaded: package | build ---------------------------|----------------- ca-certificates-2018.10.15 | ha4d7672_0 135 KB conda-forge certifi-2018.10.15 | py36_1000 138 KB conda-forge pycairo-1.17.1 | py36h4d1f301_0 202 KB conda-forge boost-cpp-1.66.0 | 1 18.8 MB conda-forge openssl-1.0.2p | h470a237_1 3.1 MB conda-forge boost-1.66.0 | py36_1 317 KB conda-forge rdkit-2018.03.4 | py36ha4bbe77_0 19.4 MB conda-forge ------------------------------------------------------------ Total: 42.0 MB The following NEW packages will be INSTALLED: boost: 1.66.0-py36_1 conda-forge boost-cpp: 1.66.0-1 conda-forge pycairo: 1.17.1-py36h4d1f301_0 conda-forge The following packages will be UPDATED: ca-certificates: 2018.03.07-0 --> 2018.10.15-ha4d7672_0 conda-forge certifi: 2018.10.15-py36_0 --> 2018.10.15-py36_1000 conda-forge openssl: 1.0.2p-h14c3975_0 --> 1.0.2p-h470a237_1 conda-forge The following packages will be DOWNGRADED: rdkit: 2018.03.4.0-py36h71b666b_1 rdkit --> 2018.03.4-py36ha4bbe77_0 conda-forge ------------------------------------------------------------------------------ Cheers, Thomas On Thu, Oct 18, 2018 at 4:43 PM Greg Landrum <greg.land...@gmail.com> wrote: > > > > On Thu, Oct 18, 2018 at 2:21 PM Eric Jonas <jo...@ericjonas.com> wrote: >> >> Greg, I'm all for anything that makes the release process on developers easier; my main question is : With conda-forge, how hard is it to install just _one_ package without having everything else (say numpy, pandas, etc) upgraded to the latest conda-forge version? I've had situations in the past where i'm like "oh I'd just like the latest ___" and suddenly everything in my conda env has been upgraded to the bleeding edge. > > > That's a great question, and it's one I don't really know the answer to. > > On my PC (I'm on the train, and this is what I have with me), here's what I did: > - create a new conda environment that includes an rdkit-channel RDKit install > - uninstall the RDKit from that > - install the RDKit from the conda-forge channel > > Here's what ends up getting changed: > > ## Package Plan ## > > environment location: C:\Users\glandrum\Anaconda3\envs\py36_tmp > > added / updated specs: > - rdkit > > > The following NEW packages will be INSTALLED: > > boost: 1.67.0-py36_vc14_0 conda-forge [vc14] > boost-cpp: 1.67.0-vc14_0 conda-forge [vc14] > pycairo: 1.16.3-py36_vc14_0 conda-forge [vc14] > rdkit: 2018.03.4-py36h857267b_1000 conda-forge > > The following packages will be UPDATED: > > certifi: 2018.10.15-py36_0 --> 2018.10.15-py36_1000 conda-forge > jpeg: 9b-hb83a4c4_2 --> 9b-vc14_2 conda-forge [vc14] > tk: 8.6.8-hfa6e2cd_0 --> 8.6.8-vc14_0 conda-forge [vc14] > > The following packages will be DOWNGRADED: > > icu: 58.2-ha66f8fd_1 --> 58.2-vc14_0 conda-forge [vc14] > libpng: 1.6.35-h2a8f88b_0 --> 1.6.34-vc14_0 conda-forge [vc14] > libtiff: 4.0.9-h36446d0_2 --> 4.0.9-vc14_0 conda-forge [vc14] > pillow: 5.3.0-py36hdc69c19_0 --> 5.2.0-py36h08bbbbd_0 > pixman: 0.34.0-hcef7cb0_3 --> 0.34.0-vc14_2 conda-forge [vc14] > vc: 14.1-h0510ff6_4 --> 14-0 conda-forge > zlib: 1.2.11-h8395fce_2 --> 1.2.11-vc14_0 conda-forge [vc14] > > > That's a fair amount of change, but is less than what I thought might happen (I was worried about numpy+pandas+... being updated). > So that's one data point. What's your take? > > > I will try the same thing on my Mac and Linux boxes tomorrow if no one else has done it by then. > > -greg > > > _______________________________________________ > Rdkit-discuss mailing list > Rdkit-discuss@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/rdkit-discuss _______________________________________________ Rdkit-discuss mailing list Rdkit-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/rdkit-discuss Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ Rdkit-discuss mailing list Rdkit-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
_______________________________________________ Rdkit-discuss mailing list Rdkit-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
