mungo 0.1.2

Quick environment solver

mungo

IMPORTANT WARNING mungo is experimental, things are very likely to break.

mungo acts as a drop-in replacement for conda create and conda install - but with way faster environment resolution. The improvements in speed are achieved by employing linear programming instead of sat-solving and caching of intermediate results (see "The ILP" for more information).

Note that we do not want to replace conda - we merely wish to share our solution so that it can be tested and perhaps someday be integrated into conda itself.

Caveats

• mungo only exposes basic functionality:
  • install (package_spec)+ (--name NAME)? (--channel CHANNEL)* (--file FILE)?
  • create  (package_spec)* (--name NAME)? (--channel CHANNEL)* (--file FILE)?
• mungo's solutions will not be the same as conda's.
• mungo cannot handle custom channel urls. Any channel which has repodata.json.bz2 files available from https://conda.anaconda.org/CHANNEL/ARCH is fine, though.
• mungo only reads channels from ~/.condarc (but also uses the --channel arguments, if supplied, of course).
• mungo only supports linux at the moment.

Installation

You need a working conda installation, since actual package installation is delegated to conda.

from PIP

pip install mungo

Examples

# create an environment named 'foo' with the specified packages
mungo create -n foo "python>=3.7" pulp packaging pyyaml

# create an environment from an environment file (such as mungo.yml)
mungo create --file mungo.yml

# install packages into the current environment
mungo install bwa

# install packages into a different environment
mungo install -n foo bwa

The ILP

mungo uses an integer linear program to determine a configuration of compatible packages which maximizes version numbers while also keeping channel order in mind.

After merging, ILP variables ∈ {0, 1} are created for each p-node. These variables relate to (package, version) configurations and tell us whether a configuration is selected for installation (1) or not (0). In a second step, the following ILP constraints are generated from the constrain nodes defined above: For each parent p (variable), the sum of all children (variables) C must be greater or equal to p; in other words: If p is selected, at least one available version of each dependent package must also be selected. If p is not selected for installation (i.e. p = 0), installation status of its dependencies is not relevant (for p).

Changelog

• version 0.1.0: Initial version.