condarc.example

# ============================================================================
# ~/.condarc — Conda configuration
# ============================================================================
#
# Conda's behavior is shaped almost entirely by ~/.condarc. The defaults
# are conservative (slow solver, narrow channel, no startup discipline),
# which causes the "conda is slow / breaks my env" complaints you'll see
# online. The settings below fix the three biggest pain points:
#
#   1. Slow dependency resolution → libmamba solver
#   2. Narrow package coverage    → conda-forge channel
#   3. Auto-activated base on every shell startup → disabled
#
# Apply by copying this file to ~/.condarc, then running:
#   conda config --show     # verify it was picked up
#
# ============================================================================

# ----------------------------------------------------------------------------
# Channels — where conda looks for packages, in priority order
# ----------------------------------------------------------------------------
# conda-forge is the community-maintained channel with the broadest package
# coverage and the fastest updates. defaults is Anaconda's official channel
# — narrower, slower-moving, but bundled with Miniconda installs.
#
# Listing conda-forge FIRST means it wins on package source when both
# channels have the same package. defaults is kept as a fallback for
# packages that exist only there.
# ----------------------------------------------------------------------------

channels:
  - conda-forge
  - defaults

# ----------------------------------------------------------------------------
# Strict channel priority — prevents "mixed-channel dependency hell"
# ----------------------------------------------------------------------------
# Without strict priority, conda can pull dependencies from any channel
# regardless of order. Result: an env where numpy comes from conda-forge,
# scipy comes from defaults, and the two were built against incompatible
# BLAS libraries. Subtle breakage, hard to debug.
#
# Strict priority forces conda to satisfy ALL deps from the highest-priority
# channel that has them, only falling back to lower-priority channels when
# necessary. Slightly stricter solves, dramatically fewer "why is this
# import segfaulting" mysteries.
# ----------------------------------------------------------------------------

channel_priority: strict

# ----------------------------------------------------------------------------
# Solver — use libmamba (5–10x faster than the classic solver)
# ----------------------------------------------------------------------------
# libmamba is now the default in newer conda versions, but older installs
# still use the classic solver. Setting this explicitly future-proofs the
# config and makes the choice visible.
#
# If conda errors with "no module named libmambapy" when this is set,
# install it once: `conda install -n base conda-libmamba-solver`
# ----------------------------------------------------------------------------

solver: libmamba

# ----------------------------------------------------------------------------
# Don't auto-activate the base env on shell startup
# ----------------------------------------------------------------------------
# Default conda behavior is to activate the `base` env every time you open
# a new shell. This pollutes your PATH, slows shell startup, and means
# every `python` call you make resolves to base's Python instead of the
# system Python. Setting this to false means `conda activate <env>` is an
# explicit, intentional action — not a side effect of opening a terminal.
#
# Tradeoff: you have to `conda activate <env>` every time you want conda's
# tools. The discipline is worth it.
# ----------------------------------------------------------------------------

auto_activate_base: false

# ----------------------------------------------------------------------------
# Don't show channel URLs in the conda list output
# ----------------------------------------------------------------------------
# Cosmetic — keeps `conda list` output clean and readable. Set to true if
# you want to see exactly which channel each package came from (useful
# when debugging cross-channel issues).
# ----------------------------------------------------------------------------

show_channel_urls: false

# ----------------------------------------------------------------------------
# pip + conda interop — keep them separate, on purpose
# ----------------------------------------------------------------------------
# When pip_interop_enabled is true, conda tries to manage pip-installed
# packages as if they were conda packages. This sounds helpful and usually
# causes more problems than it solves — conda's solver can decide a
# pip-installed package is "incompatible" and try to replace it, breaking
# the env.
#
# Better discipline: use conda for the Python version and any C-extension-
# heavy packages (numpy, scipy, pytorch); use pip for everything else
# (most pure-Python libs); accept that conda doesn't track pip's state.
# When in doubt, install via conda first; fall back to pip only if the
# package isn't in conda-forge.
# ----------------------------------------------------------------------------

pip_interop_enabled: false