Worktrees

Work on several feature branches in parallel without re-cloning, re-configuring, or paying a cold compile per branch.

Why this is cheap here

Two properties of the setup compound nicely:

1. .git/info/exclude and the hooks live in the common git directory — isolation registered once covers every worktree automatically.
2. The ccache is shared across worktrees, while each worktree keeps its own build/ and .dev-env/ state. A new worktree branched from a recently built revision compiles mostly from cache (≥ 80 % hits).

Lifecycle

# create — one command does git worktree add AND bootstrap
qgis-dev worktree add fix-labeling-crash          # sibling: ../QGIS-fix-labeling-crash

# work there
cd ../QGIS-fix-labeling-crash && qgis-dev build   # warm from shared ccache

# overview
qgis-dev worktree status
# WORKTREE                       BRANCH               DIRTY  AHEAD  BUILD
# …/QGIS                         master                   0      0   28G
# …/QGIS-fix-labeling-crash      fix-labeling-crash       2      1   9G

# finish — guarded: refuses if dirty or unpushed (override with --force)
qgis-dev worktree remove ../QGIS-fix-labeling-crash

From Neovim: <leader>pwn create · <leader>pwl switch (:tcd, so tasks, clangd and quickfix all follow) · <leader>pws dashboard · <leader>pwD remove · <leader>pwp prune.

Each worktree is fully independent

Own build tree, own compile database (clangd follows the buffer), own build profile, own throwaway QGIS run-profiles. Only the ccache — and the isolation guarantee — are shared.