boost-cpp 1.70 is making conda solve 70x slower about boost-cpp-feedstock HOT 10 CLOSED

I think we should be able to sort something out here. I'll see what we can do. This seems like a solver bug, or at least a non-optimal handling of the situation.

from boost-cpp-feedstock.

In this case, do you have libboost installed?

from boost-cpp-feedstock.

Sorry, I've updated the issue description and added a comment in the PR, I've just realized that there is this libboost package in Anaconda main, so it seems unrelated.

from boost-cpp-feedstock.

We're also in the middle of a migration of the boost-cpp pinning from 1.68 to 1.70. Maybe not all your dependencies were already updated: https://conda-forge.org/status/

from boost-cpp-feedstock.

I'm using an environment.yml with everything pinned down (with ==), so it would fail if some dependency was not updated already. We use some mirrors and some custom channels here, but I'll try to craft an environment.yml that reproduces the problem with only Anaconda and conda-forge, and post more debug info.

from boost-cpp-feedstock.

Here is an environment.yml: https://gist.github.com/tadeu/49c6de435dae6764f6616eb10efde2a4/d5d9063f87dd71e81ec9cc199fb288382967ce1b#file-environment-boost-1-70-yml

The command used is conda env update --file environment-boost-1.68.yml --prune (or conda env update --file environment-boost-1.70.yml --prune). With boost-cpp ==1.68.0 it's taking 10s to solve, with boost-cpp ==1.70.0 it's taking a lot of time (11 minutes).

from boost-cpp-feedstock.

Just FYI, I've solved the issue locally by adding boost-cpp packages to a higher priority internal channel, with higher build numbers and removing this from index.json:

  "constrains": [
    "libboost <0"
  ],

but I understand that this is not a solution for everyone, in our case we don't have libboost in our Anaconda main mirror because we filter some packages out.

from boost-cpp-feedstock.

Note: I've tested the same conda env update with conda 4.7.3 from conda-canary, and it's down to 5.5 minutes (from the old 11 minutes), but it's still much slower than for Boost 1.68 packages.

from boost-cpp-feedstock.

can you retry with latest packages and check that is stll a valid issue ?

from boost-cpp-feedstock.

Given this issue is a few years old, several versions of Boost have come out, the original use cases involves mixing conda-forge & defaults (which is no longer supported), and possibly other changes I've missed, going to close this out. If this is still coming up, would recommend starting a new issue with a reproducer. Thanks!

from boost-cpp-feedstock.