Speed up shell startup 9x (1.9s → 0.2s)

Lazy-load Intel oneAPI setvars.sh (~1.5s) via wrapper functions that
source the environment on first use of icc/icx/ifort/etc.

Cache all shell completion outputs to ~/.cache/bash.d/ so they are
sourced from disk instead of regenerated via subprocess on every
shell start.  Cache invalidates automatically when the tool binary
is updated.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

00-completion-cache

@@ -0,0 +1,25 @@
+# shellcheck shell=bash
+# Cache completion scripts to avoid subprocess overhead on every shell start.
+#
+# Usage:  cached_completion <tool> <generation-command...>
+# Example: cached_completion gh gh completion -s bash
+#
+# The cache file is regenerated when the tool binary is newer than the cache.
+# Cache location: ${XDG_CACHE_HOME:-$HOME/.cache}/bash.d/
+
+_COMP_CACHE="${XDG_CACHE_HOME:-$HOME/.cache}/bash.d"
+
+cached_completion() {
+    local tool="$1"; shift
+    local cache_file="${_COMP_CACHE}/${tool}.bash"
+    local tool_path
+    tool_path="$(command -v "$tool" 2>/dev/null)" || return
+
+    if [[ ! -f "$cache_file" || "$tool_path" -nt "$cache_file" ]]; then
+        mkdir -p "$_COMP_CACHE"
+        "$@" > "$cache_file" 2>/dev/null || return
+    fi
+
+    # shellcheck disable=SC1090
+    . "$cache_file"
+}

20-oneapi

@@ -1,13 +1,45 @@
 # shellcheck shell=bash
-# Intel oneAPI toolkit — source the environment setup if installed
+# Intel oneAPI toolkit — lazy-loaded to avoid ~1.5s startup penalty.
-# shellcheck disable=SC1090
+# The full environment is sourced on first use of any Intel compiler/tool.
+# To load manually: _load_oneapi
+
 ONEAPI_SETVARS="/opt/intel/oneapi/setvars.sh"
+
 if [[ -f "$ONEAPI_SETVARS" ]]; then
+    _load_oneapi() {
+        # Guard against double-loading
+        if [[ "${_ONEAPI_LOADED:-0}" -eq 1 ]]; then
+            return
+        fi
+        _ONEAPI_LOADED=1
+
+        # Remove the lazy-load wrappers before sourcing
+        local _cmd
+        for _cmd in icc icpc icx icpx ifort ifx dpcpp; do
+            unset -f "$_cmd" 2>/dev/null
+        done
+
+        # shellcheck disable=SC1090
         . "$ONEAPI_SETVARS" --force > /dev/null
+
         if [[ -n "$ONEAPI_ROOT" ]]; then
             case ":${LD_LIBRARY_PATH}:" in
                 *":${ONEAPI_ROOT}/lib:"*) ;;
                 *) export LD_LIBRARY_PATH="${ONEAPI_ROOT}/lib${LD_LIBRARY_PATH:+:$LD_LIBRARY_PATH}" ;;
             esac
         fi
+    }
+
+    # Create wrapper functions for common Intel tools that trigger lazy-load
+    _oneapi_lazy_wrap() {
+        local cmd="$1"
+        # shellcheck disable=SC2139
+        eval "${cmd}() { _load_oneapi; command ${cmd} \"\$@\"; }"
+    }
+
+    for _oneapi_cmd in icc icpc icx icpx ifort ifx dpcpp; do
+        _oneapi_lazy_wrap "$_oneapi_cmd"
+    done
+    unset -f _oneapi_lazy_wrap
+    unset _oneapi_cmd
 fi

50-asdf-completion

@@ -1,4 +1,3 @@
 # shellcheck shell=bash
-# shellcheck disable=SC1090
 # Tab completion for asdf, a tool that manages multiple language runtime versions
-command -v asdf &>/dev/null && . <(asdf completion bash)
+cached_completion asdf asdf completion bash

50-bun-completion

@@ -1,4 +1,3 @@
 # shellcheck shell=bash
-# shellcheck disable=SC1090
 # Enable bash completion for bun (JavaScript runtime & package manager)
-command -v bun &>/dev/null && . <(bun completions)
+cached_completion bun bun completions

50-fj-completion

@@ -1,4 +1,3 @@
 # shellcheck shell=bash
-# shellcheck disable=SC1090
 # Tab completion for fj, the Forgejo CLI (like GitHub CLI but for Forgejo)
-command -v fj &>/dev/null && . <(fj completion bash)
+cached_completion fj fj completion bash

50-gh-completion

@@ -1,4 +1,3 @@
 # shellcheck shell=bash
-# shellcheck disable=SC1090
 # Tab completion for gh, the GitHub CLI
-command -v gh &>/dev/null && . <(gh completion -s bash)
+cached_completion gh gh completion -s bash

50-podman-completion

@@ -1,4 +1,3 @@
 # shellcheck shell=bash
-# shellcheck disable=SC1090
 # Tab completion for Podman, a rootless container engine
-command -v podman &>/dev/null && . <(podman completion bash)
+cached_completion podman podman completion bash

50-tailscale-completion

@@ -1,4 +1,3 @@
 # shellcheck shell=bash
-# shellcheck disable=SC1090
 # Tab completion for Tailscale, a mesh VPN for connecting devices
-command -v tailscale &>/dev/null && . <(tailscale completion bash)
+cached_completion tailscale tailscale completion bash

50-uv-completion

@@ -1,3 +1,3 @@
 # shellcheck shell=bash
 # Enable bash completion for uv (Python package manager)
-command -v uv &>/dev/null && eval "$(uv generate-shell-completion bash)"
+cached_completion uv uv generate-shell-completion bash

CLAUDE.md

@@ -22,6 +22,7 @@ Use these in all scripts — do not manipulate `PATH` or check permissions manua
 - `path_append <dir>` — add directory to end of `$PATH` (checks existence, prevents duplicates)
 - `path_prepend <dir>` — add directory to start of `$PATH`
 - `require_private <file>` — warn if file is group/world-accessible; use in credential files
+- `cached_completion <tool> <gen-cmd...>` — cache completion output to `~/.cache/bash.d/`; regenerates when the tool binary is updated. Use this for all new completion scripts instead of `eval` or `. <()`
 
 ## Writing Scripts
 
@@ -30,6 +31,8 @@ Use these in all scripts — do not manipulate `PATH` or check permissions manua
 - Guard directory-dependent exports with `[[ -d <path> ]]` before exporting
 - Do not suppress stderr from external scripts — only redirect stdout when needed
 - Avoid `eval` when `. <(cmd)` (process substitution) works
+- For shell completions, use `cached_completion` instead of `eval` or `. <()` to avoid subprocess overhead on every shell start
+- For slow environment scripts, use lazy-loading (see `20-oneapi` for the wrapper-function pattern)
 - Prevent `LD_LIBRARY_PATH` duplicates with a `case` guard (see `20-oneapi` for example)
 
 ## Permissions

README.md

@@ -35,6 +35,7 @@ Lower numbers load first, so foundational pieces like helper functions (`00-`) a
 
 - **`00-path-helper`** — Defines `path_append` and `path_prepend` functions that safely add directories to `$PATH` (checking the directory exists and avoiding duplicates).
 - **`00-credential-guard`** — Defines `require_private`, which warns at shell startup if a credential file has overly permissive permissions (anything beyond owner-only access).
+- **`00-completion-cache`** — Defines `cached_completion`, which caches shell completion scripts to `~/.cache/bash.d/` so they are only regenerated when the tool binary is updated. Used by all `50-*-completion` scripts.
 
 ### PATH (`10-`)
 
@@ -46,7 +47,7 @@ Lower numbers load first, so foundational pieces like helper functions (`00-`) a
 
 - **`20-android`** — Sets `$ANDROID_HOME`, `$JAVA_HOME`, and adds Android SDK tools to PATH (only when installed).
 - **`20-ninja`** — Configures Ninja as the default CMake generator and enables ccache for C/C++ builds (only when installed).
-- **`20-oneapi`** — Sources the Intel oneAPI environment and adds its libraries to `LD_LIBRARY_PATH`.
+- **`20-oneapi`** — Lazy-loads the Intel oneAPI environment. Because `setvars.sh` takes ~1.5 seconds, the full environment is deferred until an Intel tool (`icc`, `icx`, `ifort`, etc.) is first invoked. Call `_load_oneapi` to trigger it manually.
 
 ### Prompt (`30-`)
 
@@ -54,9 +55,14 @@ Lower numbers load first, so foundational pieces like helper functions (`00-`) a
 
 ### Completions (`50-`)
 
+All completions (except `50-claude-completion`) use `cached_completion` to avoid subprocess overhead on every shell start. Cache files live in `~/.cache/bash.d/` and are automatically regenerated when the tool binary is updated.
+
 - **`50-asdf-completion`** — Tab completion for the asdf version manager.
+- **`50-bun-completion`** — Tab completion for Bun (JavaScript runtime & package manager).
 - **`50-claude-completion`** — Tab completion for [Claude Code](https://claude.com/claude-code) CLI.
 - **`50-fj-completion`** — Tab completion for the Forgejo CLI (`fj`).
+- **`50-gh-completion`** — Tab completion for the GitHub CLI (`gh`).
+- **`50-podman-completion`** — Tab completion for Podman (rootless container engine).
 - **`50-tailscale-completion`** — Tab completion for Tailscale.
 - **`50-uv-completion`** — Tab completion for [uv](https://github.com/astral-sh/uv) (Python package manager).