devlaunch 0.0.25__tar.gz → 0.0.27__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (37) hide show
  1. devlaunch-0.0.27/PKG-INFO +1587 -0
  2. devlaunch-0.0.27/README.md +1565 -0
  3. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/completions/dl.bash +1 -1
  4. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/devpod_provider.py +22 -5
  5. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/disk_usage.py +32 -1
  6. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/dl.py +2273 -152
  7. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/gh_auth.py +16 -11
  8. devlaunch-0.0.27/devlaunch/timing.py +410 -0
  9. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/tools.py +477 -58
  10. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/workspace_id.py +3 -2
  11. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/workspace_state.py +5 -0
  12. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/worktree/branch_manager.py +32 -9
  13. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/worktree/config.py +1 -4
  14. devlaunch-0.0.27/devlaunch/worktree/git_errors.py +35 -0
  15. devlaunch-0.0.27/devlaunch/worktree/locks.py +142 -0
  16. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/worktree/migration.py +60 -19
  17. devlaunch-0.0.27/devlaunch/worktree/repo_manager.py +640 -0
  18. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/worktree/storage.py +10 -3
  19. devlaunch-0.0.27/devlaunch/worktree/workspace_clone.py +808 -0
  20. {devlaunch-0.0.25 → devlaunch-0.0.27}/pyproject.toml +17 -1
  21. devlaunch-0.0.25/PKG-INFO +0 -786
  22. devlaunch-0.0.25/README.md +0 -764
  23. devlaunch-0.0.25/devlaunch/worktree/locks.py +0 -65
  24. devlaunch-0.0.25/devlaunch/worktree/repo_manager.py +0 -351
  25. devlaunch-0.0.25/devlaunch/worktree/workspace_clone.py +0 -484
  26. {devlaunch-0.0.25 → devlaunch-0.0.27}/.gitignore +0 -0
  27. {devlaunch-0.0.25 → devlaunch-0.0.27}/LICENSE +0 -0
  28. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/__init__.py +0 -0
  29. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/aid.py +0 -0
  30. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/completion.py +0 -0
  31. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/completion_loader.py +0 -0
  32. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/completions/__init__.py +0 -0
  33. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/devpod_ssh.py +0 -0
  34. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/tty_session.py +0 -0
  35. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/worktree/__init__.py +0 -0
  36. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/worktree/models.py +0 -0
  37. {devlaunch-0.0.25 → devlaunch-0.0.27}/devlaunch/xdg.py +0 -0
@@ -0,0 +1,1587 @@
1
+ Metadata-Version: 2.5
2
+ Name: devlaunch
3
+ Version: 0.0.27
4
+ Summary: DevLaunch - A streamlined CLI for devpod workspaces
5
+ Project-URL: Source, https://github.com/blooop/devlaunch
6
+ Project-URL: Home, https://github.com/blooop/devlaunch
7
+ Author-email: Austin Gregg-Smith <blooop@gmail.com>
8
+ License-Expression: MIT
9
+ License-File: LICENSE
10
+ Requires-Dist: iterfzf>=1.0.0
11
+ Requires-Dist: tomli>=2.0.0
12
+ Provides-Extra: test
13
+ Requires-Dist: coverage<=7.14.1,>=7.5.4; extra == 'test'
14
+ Requires-Dist: hypothesis<=6.155.1,>=6.104.2; extra == 'test'
15
+ Requires-Dist: prek<0.5.0,>=0.2.28; extra == 'test'
16
+ Requires-Dist: pylint<=4.0.5,>=3.2.5; extra == 'test'
17
+ Requires-Dist: pytest-cov<=7.1.0,>=4.1; extra == 'test'
18
+ Requires-Dist: pytest<=9.0.3,>=7.4; extra == 'test'
19
+ Requires-Dist: ruff<=0.15.15,>=0.5.0; extra == 'test'
20
+ Requires-Dist: ty<=0.0.19,>=0.0.12; extra == 'test'
21
+ Description-Content-Type: text/markdown
22
+
23
+ # devlaunch
24
+
25
+ A streamlined CLI for [devpod](https://devpod.sh) with intuitive autocomplete and fzf fuzzy selection.
26
+
27
+ ## Continuous Integration Status
28
+
29
+ [![Ci](https://github.com/blooop/devlaunch/actions/workflows/ci.yml/badge.svg?branch=main)](https://github.com/blooop/devlaunch/actions/workflows/ci.yml?query=branch%3Amain)
30
+ [![Codecov](https://codecov.io/gh/blooop/devlaunch/branch/main/graph/badge.svg?token=Y212GW1PG6)](https://codecov.io/gh/blooop/devlaunch)
31
+ [![GitHub issues](https://img.shields.io/github/issues/blooop/devlaunch.svg)](https://GitHub.com/blooop/devlaunch/issues/)
32
+ [![GitHub pull-requests merged](https://badgen.net/github/merged-prs/blooop/devlaunch)](https://github.com/blooop/devlaunch/pulls?q=is%3Amerged)
33
+ [![GitHub release](https://img.shields.io/github/release/blooop/devlaunch.svg)](https://GitHub.com/blooop/devlaunch/releases/)
34
+ [![PyPI](https://img.shields.io/pypi/v/devlaunch)](https://pypi.org/project/devlaunch/)
35
+ [![Conda](https://img.shields.io/badge/conda-v0.0.9-brightgreen?logo=anaconda)](https://prefix.dev/channels/blooop/packages/devlaunch)
36
+ [![License](https://img.shields.io/github/license/blooop/devlaunch)](https://opensource.org/license/mit/)
37
+ [![Python](https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12%20%7C%203.13%20%7C%203.14-blue)](https://www.python.org/downloads/)
38
+ [![Pixi Badge](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/prefix-dev/pixi/main/assets/badge/v0.json)](https://pixi.sh)
39
+
40
+ ## Installation
41
+
42
+ ### Pixi (Recommended)
43
+
44
+ ```bash
45
+ pixi global install --channel conda-forge --channel https://prefix.dev/blooop devlaunch
46
+ ```
47
+
48
+ This installs `devlaunch` along with `devpod` and all dependencies automatically.
49
+
50
+ ### Pip
51
+
52
+ ```bash
53
+ pip install devlaunch
54
+ ```
55
+
56
+ Note: When using pip, you must install [devpod](https://devpod.sh/docs/getting-started/install) separately.
57
+ If `devpod` is not on `PATH`, every command that needs it prints a single install hint on stderr and exits `127`
58
+ (the shell's "command not found" code). `dl --help` and `dl --version` keep working without it.
59
+
60
+ A `devpod` that is installed but cannot answer is a different failure and gets a different exit code. If
61
+ `devpod list` exits non-zero, or prints something that is not a `--output json` workspace listing, `dl` quotes
62
+ what devpod said on stderr and exits `1` rather than reporting that you have no workspaces — so `dl --purge`
63
+ stops instead of deleting caches it never checked. Shell completion is the deliberate exception: `dl --install`,
64
+ `dl --refresh` and `dl --completion-data` log the failure and carry on with the repos and branches they can
65
+ still discover on local disk, so an unreachable devpod costs you workspace-name completion and nothing more.
66
+
67
+ ### Shell Completions
68
+
69
+ After installation, set up shell completions for `dl` and `aid`:
70
+
71
+ ```bash
72
+ dl --install
73
+ source ~/.bashrc # or restart your terminal
74
+ ```
75
+
76
+ ## Usage
77
+
78
+ ```bash
79
+ dl # Interactive workspace selector (fzf)
80
+ dl <user/repo> # Start workspace and attach shell
81
+ dl <user/repo> <cmd> # Run workspace command (stop, code, etc.)
82
+ dl <user/repo> -- <command> # Run shell command in workspace
83
+ ```
84
+
85
+ ### Commands that need a terminal
86
+
87
+ `dl <ws> -- <command>` gives the command a terminal whenever `dl` itself has one,
88
+ so interactive programs — a coding agent, `htop`, `git rebase -i`, a REPL — start
89
+ and stay up instead of exiting immediately. Redirect the output and the terminal
90
+ goes away again, so `dl <ws> -- ls > files.txt` stays free of escape sequences.
91
+
92
+ This needs the ssh host alias `devpod up` writes to `~/.ssh/config`. If a
93
+ workspace has none, `dl` says so and falls back to the plain `devpod ssh`
94
+ transport, which has no terminal; `dl <ws> restart` republishes the alias. Set
95
+ `DEVLAUNCH_NO_TTY=1` to force the fallback everywhere.
96
+
97
+ ## aid: start a coding agent in a workspace
98
+
99
+ `aid` is `dl` with a coding agent started for you:
100
+
101
+ ```bash
102
+ aid <user/repo>[@branch] [prompt...] # Open the workspace, start the agent
103
+ ```
104
+
105
+ It is a shortcut, not a second launcher. `aid` rewrites its command line into a
106
+ `dl` one and hands it to `dl` itself, so
107
+
108
+ ```bash
109
+ aid blooop/devlaunch@fix/42 fix the flaky test
110
+ ```
111
+
112
+ is exactly
113
+
114
+ ```bash
115
+ dl blooop/devlaunch@fix/42 -- IS_SANDBOX=1 claude --dangerously-skip-permissions 'fix the flaky test'
116
+ ```
117
+
118
+ That means an `aid` workspace *is* the `dl` workspace: same clone, same workspace
119
+ id, same container — started if stopped, attached to if already running, and never
120
+ rebuilt just because `aid` asked for it. Anything `dl` learns, `aid` gets.
121
+
122
+ `claude` is started with `--dangerously-skip-permissions`. The agent is already
123
+ inside a disposable container holding only this repo, so the per-tool prompts it
124
+ would ask on the host protect nothing here and would stall an unattended run.
125
+ `IS_SANDBOX=1` rides along because `claude` otherwise refuses that flag outright
126
+ under `uid 0`, and devcontainers that run as root are ordinary. The variable is
127
+ scoped to the agent process, not exported into your shell.
128
+
129
+ The trade is worth stating plainly: an agent started this way edits, runs and
130
+ deletes inside the container without asking. It cannot reach your host, but it can
131
+ rewrite the checkout it is in, so review an `aid` workspace before pushing rather
132
+ than treating it as a sandbox that will stop it for you. `--codex` and `--gemini`
133
+ are unaffected, and `dl <ws> -- claude` still runs exactly what you typed.
134
+
135
+ | Option | Description |
136
+ |--------|-------------|
137
+ | `--claude`, `--codex`, `--gemini` | Pick the agent (default: `claude`) |
138
+ | `--devcontainer <variant\|path>` | Passed through to `dl` |
139
+ | `DEVLAUNCH_AID_AGENT=<agent>` | Change the default agent |
140
+
141
+ Everything after the workspace is the prompt, flags and all, so it never needs
142
+ quoting to survive `aid`'s own parsing. Managing workspaces — listing, stopping,
143
+ deleting, VS Code — stays with `dl`.
144
+
145
+ The agent's CLI has to be installed in the container; `aid` runs it there, it does
146
+ not install it.
147
+
148
+ ## Workspace Sources
149
+
150
+ ```bash
151
+ dl myproject # Existing workspace by name
152
+ dl user/repo # Create from GitHub repo
153
+ dl user/repo@branch # Create from specific branch
154
+ dl ./path # Create from local path
155
+ ```
156
+
157
+ ## Workspace IDs
158
+
159
+ `dl user/repo@branch` derives one id that names both the devpod workspace (what you
160
+ see in `dl --ls`) and the clone directory under `~/.cache/devlaunch/repos/`:
161
+
162
+ ```
163
+ <repo-slug>-<branch-slug>-<syllables> at most 38 characters
164
+
165
+ blooop/devlaunch@main -> devlaunch-main-zovomobo
166
+ blooop/devlaunch@feature/auth -> devlaunch-feature-auth-poliseno
167
+ blooop/devlaunch@feature-auth -> devlaunch-feature-auth-nesatabe
168
+ blooop/test_renv@nb4 -> test-renv-nb4-polenita
169
+ kinisi-robotics/kinisi_ros@ags-devcontainer-tooling-support
170
+ -> kinisi-ros-ags-devcontainer-t-lenevere
171
+ blooop/devlaunch@dependabot/github_actions/codecov/codecov-action-6
172
+ -> devlaunch-dependabot-codecov-sifivasa
173
+ ```
174
+
175
+ The eight-character syllable suffix is a hash of the full `(owner, repo, branch)` triple.
176
+ It is what makes the id unique: the readable part is shortened to fit the length limit,
177
+ and shortening it does not affect whether two branches share an id. Long branch names
178
+ drop whole `/`-separated middle segments before losing characters, so the part that
179
+ identifies the branch survives. Note the third and fourth lines above: `feature/auth` and
180
+ `feature-auth` read the same once slugged but are different branches, and they get
181
+ different ids.
182
+
183
+ Owner and repo are matched case-insensitively, the way GitHub treats them, so
184
+ `dl NVIDIA/cuda-samples@main` and `dl nvidia/cuda-samples@main` are the same workspace.
185
+ Branch names are case-sensitive, because git refs are.
186
+
187
+ URL specs (`dl github.com/owner/repo`) get an id in the same shape, with the suffix
188
+ hashed over the URL.
189
+
190
+ The id is also the container hostname, so it stays well inside the 38-character budget
191
+ to leave room for tools that add their own prefixes.
192
+
193
+ Branch names must be safe as both git refs and directory names — a name with a space or
194
+ a leading dash is rejected rather than quietly rewritten.
195
+
196
+ ### Upgrading from an older devlaunch
197
+
198
+ This id format is new, and the directories and containers on your machine were named by
199
+ the previous scheme. The first `dl user/repo…` command after upgrading migrates the cache
200
+ once and prints what it did. `dl --help`, `dl --version`, `dl --ls` and opening an existing
201
+ workspace by name do not trigger it.
202
+
203
+ **Your clone directories are renamed.** What was
204
+ `~/.cache/devlaunch/repos/blooop/devlaunch/main` becomes
205
+ `~/.cache/devlaunch/repos/blooop/devlaunch/devlaunch-main-zovomobo`. A workspace is a git
206
+ clone whose `origin` points at the `.bare` cache next to it, and `.bare` does not move, so
207
+ this is a plain rename: branches, history and **uncommitted changes all survive** — only
208
+ the folder name changes. `metadata.json` is updated in the same pass, so nothing is left
209
+ pointing at the old name.
210
+
211
+ **Your existing devpod containers keep their old ids and are orphaned — and they can
212
+ often be repaired rather than replaced.** An orphaned container is sourced at the path this
213
+ migration just renamed, with the real clone sitting next to it under the new name, which is
214
+ precisely what [`dl --reconcile`](#reconciling-records-that-disagree) is for: it re-points
215
+ devpod's record at the renamed clone, and `dl <workspace> recreate` finishes the repair.
216
+ That gives you back the clone association and the workspace's identity — not state that
217
+ lived only inside the old container, which nothing can bring back. The repair is
218
+ order-dependent: relaunching the branch claims the renamed clone for a fresh container,
219
+ and reconcile never re-points a clone a live container holds — so reconcile first, then
220
+ relaunch. Left alone, the next `dl user/repo@branch` simply builds a fresh container
221
+ under the new id, and deleting the old one is all that remains for it.
222
+
223
+ dl does not delete containers for you — deleting by id is how a running sidecar got
224
+ destroyed the last time something tried ([kinisi_ros#9766](https://github.com/kinisi-robotics/kinisi_ros/pull/9766)) —
225
+ so it prints a one-line notice with the count and writes the old ids to
226
+ `~/.cache/devlaunch/orphaned-workspaces.txt`. For the workspaces you are finished with,
227
+ the disposal command reads from that listing:
228
+
229
+ ```bash
230
+ xargs -r -n1 devpod delete < ~/.cache/devlaunch/orphaned-workspaces.txt
231
+ ```
232
+
233
+ **A clone directory with no metadata record is left alone.** Nothing records which branch
234
+ it was cloned for, and the old directory name cannot be turned back into one — `feature/auth`
235
+ and `feature-auth` both became `feature-auth` — so a guessed name would be worse than no
236
+ rename. Those directories stay exactly where they are and are listed in
237
+ `~/.cache/devlaunch/unmigrated-clones.txt`.
238
+
239
+ Running dl again changes nothing: the migration is keyed on the `version` field in
240
+ `metadata.json`, not on directory names, so a branch that happens to look like a new-scheme
241
+ id is never mistaken for one. If a migration is interrupted, the next run finishes it — the
242
+ version is written last, in the same atomic save as the new paths, so it never claims more
243
+ than the filesystem has actually done. A rename the filesystem refuses (a read-only mount,
244
+ tightened permissions) is treated the same way: the version stays put and every later run
245
+ retries the refused directories and repeats the notice until the underlying refusal is
246
+ fixed by hand.
247
+
248
+ ## Workspace Commands
249
+
250
+ | Command | Description |
251
+ |---------|-------------|
252
+ | `dl <user/repo> up` | Start (or create) the workspace without attaching — for prewarming a container before a session wants it |
253
+ | `dl <user/repo> stop` | Stop the workspace |
254
+ | `dl <user/repo> rm, prune` | Delete the workspace |
255
+ | `dl <user/repo> code` | Open in VS Code |
256
+ | `dl <user/repo> restart` | Stop and start (no rebuild) |
257
+ | `dl <user/repo> recreate` | Recreate container |
258
+ | `dl <user/repo> reset` | Clean slate (remove all, recreate) |
259
+ | `dl <user/repo> dotfiles` | Refresh dotfiles in the running workspace (`chezmoi update`) |
260
+ | `dl <user/repo> -- <command>` | Run shell command in workspace (with a terminal, when `dl` has one) |
261
+
262
+ ## Options
263
+
264
+ | Option | Description |
265
+ |--------|-------------|
266
+ | `--devcontainer <variant\|path>` | Use a non-default `devcontainer.json`. A bare name means `.devcontainer/<name>/devcontainer.json`. Stored with the workspace, so pass it once. |
267
+ | `DEVLAUNCH_NO_TTY=1` | Never give a workspace command a terminal; always use the plain `devpod ssh` transport. |
268
+ | `DEVLAUNCH_DOTFILES_ON_ATTACH=1` | Refresh dotfiles before handing over an interactive shell. Off by default; see below. |
269
+
270
+ Projects with demanding devcontainers — several variants, compose sidecars, or a
271
+ host-side `initializeCommand` that has to tell branch workspaces apart — are
272
+ covered in [docs/devcontainer-projects.md](docs/devcontainer-projects.md).
273
+
274
+ ### Refreshing dotfiles on attach
275
+
276
+ devpod applies dotfiles when it *provisions* a workspace, so a workspace that has
277
+ been up for a fortnight still has the dotfiles it was born with. `dl <ws>
278
+ dotfiles` fixes that when you think of it; `DEVLAUNCH_DOTFILES_ON_ATTACH=1` makes
279
+ `dl` think of it for you, running the same refresh just before it hands you the
280
+ shell.
281
+
282
+ ```bash
283
+ DEVLAUNCH_DOTFILES_ON_ATTACH=1 dl someone/repo
284
+ ```
285
+
286
+ It is off unless you set it, and that is the point rather than caution. The
287
+ refresh is a `devpod ssh` round-trip — measured at ~1.7s, almost all of it
288
+ connection setup — with a `git pull` behind it, and it would otherwise be charged
289
+ to every attach on every machine to close a gap most people do not have.
290
+
291
+ Two things it deliberately does not do:
292
+
293
+ - **It never runs for `dl <ws> -- <command>`.** A one-shot command renders no
294
+ prompt and sources no interactive shell, so refreshing in front of it would
295
+ buy that command nothing and cost it the round-trip. That path is the one
296
+ agent launchers use, and it stays exactly as fast as it was.
297
+ - **It never holds the shell hostage.** The refresh gets 60 seconds; an
298
+ unreachable dotfiles remote, or one that wants a password nobody is there to
299
+ type, means a pause and then your shell, not a hang. Failure is a warning —
300
+ you get the workspace either way.
301
+
302
+ Refreshes run every time you attach, with no cooldown, because you asked for
303
+ them. If that is too often, unset the variable and use `dl <ws> dotfiles`.
304
+
305
+ ## GitHub Authentication
306
+
307
+ Every workspace `dl` opens inherits the host's GitHub login, so `gh` is already
308
+ authenticated inside the container and the devcontainer.json does not have to
309
+ arrange anything for it. devpod forwards the ssh agent and git credentials on its
310
+ own, but nothing else carries `gh`.
311
+
312
+ devlaunch takes the token from `GH_TOKEN`, `GITHUB_TOKEN`, or `gh auth token`,
313
+ whichever answers first, and hands it to the container as `GH_TOKEN`. That reaches
314
+ any image and any container user, unlike a bind-mount of `~/.config/gh`, and it
315
+ works whether the host keeps its token in `hosts.yml` or in a keyring. The token
316
+ is passed to devpod through a private file and through devpod's own environment,
317
+ never on a command line, so it does not appear in `ps`. `dl` installs `gh` itself
318
+ (see [Tools in every workspace](#tools-in-every-workspace)), so the login has
319
+ something to be spent on whatever the image ships. Check a workspace with:
320
+
321
+ ```bash
322
+ dl <workspace> -- gh auth status
323
+ ```
324
+
325
+ If no token can be found, `dl` warns on stderr and opens the workspace anyway rather
326
+ than failing — and the warning names the config directory `gh` consulted, because the
327
+ usual cause is a shell that scoped `XDG_CONFIG_HOME` somewhere `gh` has no login,
328
+ not a host that is actually logged out.
329
+
330
+ ### Who gets the token
331
+
332
+ Everything running in the container does — including a `postCreateCommand` from a
333
+ repo you did not write. `dl someone/repo` builds and runs that project's
334
+ devcontainer with your GitHub token in its environment, and a `gh auth login` token
335
+ usually carries `repo`, `workflow`, `gist` and `read:org` scopes. devpod already
336
+ forwards the ssh agent to every workspace, so this is not a new trust boundary, but
337
+ it is a wider one. Skip it for a repo you have not read:
338
+
339
+ ```bash
340
+ DEVLAUNCH_NO_GH_TOKEN=1 dl someone/repo
341
+ ```
342
+
343
+ | Variable | Description |
344
+ |----------|-------------|
345
+ | `DEVLAUNCH_NO_GH_TOKEN=1` | Do not forward the host's GitHub login into workspaces |
346
+
347
+ ### When the token changes
348
+
349
+ `dl` refreshes the token on every start, so rotating it on the host is enough for
350
+ any workspace that gets started or restarted afterwards. Attaching to a workspace
351
+ that is *already running* skips that step, and the token it was given at startup
352
+ stays in place — including one it was given before you set
353
+ `DEVLAUNCH_NO_GH_TOKEN`. Run `dl <workspace> restart` to replace it.
354
+
355
+ ## Tools in every workspace
356
+
357
+ `gh` and `claude` are available in every workspace `dl` opens, in every kind of
358
+ session — an interactive `dl <workspace>`, a one-shot `dl <workspace> -- <command>`,
359
+ and `aid`. The repo's `devcontainer.json` does not have to provide them, and most
360
+ do not: `dl` launches arbitrary repos, so a guarantee that depended on the image
361
+ would not be a guarantee.
362
+
363
+ ### How they get there
364
+
365
+ On `devpod up`, at most three round trips, each one earning the next.
366
+
367
+ **1. The setup pass — the only trip a ready workspace ever pays.** One trip
368
+ carries everything the host wants done on the way into a running container: the
369
+ stages first, then the probe. Naming the container — the hostname your shell
370
+ prompt shows — is the one stage today, and it costs nothing extra because the
371
+ probe was paying for the trip anyway. Each stage reports `ok`, `failed` with its
372
+ exit status, or *not reached*; one that fails stops neither the stages behind it
373
+ nor the probe, and `dl` says which one it was.
374
+
375
+ The probe is the tail of that trip. The container reports
376
+ what only it can know: whether both tools answer at all, where its `claude`
377
+ resolves to, and where `~/.local/share/claude/versions` in its own home resolves
378
+ to. It reports those and names no verdict; the host reads them, so "a real
379
+ `claude`" is defined in exactly one place. The reading is one of three:
380
+
381
+ - **provisioned** — `gh` answers on the login PATH and `claude` resolves to a
382
+ binary the official installer put in the versions directory. Nothing else
383
+ happens.
384
+ - **lendable** — both names answer, but that `claude` is a shim or a wrapper.
385
+ - **absent** — a tool is genuinely missing.
386
+
387
+ **2. A lend, for *lendable* and *absent*.** `dl` streams its own `gh` and `claude`
388
+ into the container as a tar over the `devpod ssh` channel it already holds — a
389
+ local pipe, no network and no download. Nothing lands outside a staging directory
390
+ until both binaries have been run there once, so a container that cannot execute
391
+ them (a different libc, a different architecture) is left exactly as it was.
392
+
393
+ **3. The network install, for *absent* only.** When the host had nothing to lend,
394
+ or the lend was refused, `pixi global` installs both tools — and `pixi` itself
395
+ first if the image has none. A *lendable* container never reaches this trip: it
396
+ stops after the lend, or — when the host had nothing to lend — after the probe
397
+ itself. A `claude` already answers there, and this install decides what to do
398
+ with the same `command -v` that a shim satisfies, so the trip would install
399
+ nothing.
400
+
401
+ Tools reach the PATH of a login shell through whichever of `~/.bash_profile`,
402
+ `~/.bash_login` or `~/.profile` bash actually reads — it sources only the first of
403
+ those that exists, so an image shipping a `~/.bash_profile` never reads
404
+ `~/.profile`.
405
+
406
+ An install that fails costs the workspace its tools, not its launch: `dl` logs a
407
+ warning and hands you the session anyway.
408
+
409
+ ### What to bake so a launch does no work at all
410
+
411
+ To make every `dl` launch of an image stop at trip 1. The probe asks a **login**
412
+ shell to resolve each name, so every bullet here is about what a login shell can
413
+ find:
414
+
415
+ - **`gh`** anywhere on the login PATH.
416
+ - **`claude`** in the layout its official installer creates — the binary at
417
+ `~/.local/share/claude/versions/<version>`, a **direct child** of that
418
+ directory named for the version, with `~/.local/bin/claude` symlinked to it.
419
+ Nested any deeper — `versions/<version>/bin/claude`, the shape a downloader
420
+ parked there would take — is read as somebody else's tree that merely starts
421
+ with the official path, and does not count.
422
+ - **`~/.local/bin` on the login PATH**. The symlink above is how `claude`
423
+ answers at all; a login shell that cannot find that directory reads the image
424
+ as *absent* however carefully the rest was baked, and it pays the full lend.
425
+ Ubuntu's stock `~/.profile` prepends `~/.local/bin` itself — but an image
426
+ shipping a `~/.bash_profile` never reads `~/.profile` (above), and then
427
+ nothing does.
428
+
429
+ Nothing else counts as a `claude`, and that is the point. A *shim* — a small
430
+ launcher that downloads the real binary the first time it is called — answers
431
+ `command -v claude` exactly as the real thing does, while the workspace still
432
+ owes a multi-hundred-megabyte download at the least convenient moment. So `dl`
433
+ resolves the name rather than running it (running a shim *is* the download), reads
434
+ a shim as *lendable*, and sends the host's real binary. The lend prepends
435
+ `~/.local/bin` to the login PATH, which is what puts the lent binary in front of
436
+ the shim from then on — intended, and the reason the next launch probes
437
+ *provisioned* and the transfer is paid once rather than forever.
438
+
439
+ **This repo's own devcontainer feature bakes a shim.**
440
+ `.devcontainer/claude-code/install.sh` installs `claude-shim`, so an image built
441
+ from it does *not* meet the contract by itself: its first `dl` launch is lent a
442
+ real `claude`, and only launches after that do nothing. Build the official layout
443
+ into the image if you want the first launch free too.
444
+
445
+ ### What this deliberately does not do
446
+
447
+ - **No per-tool transfer.** The lend is all-or-nothing — an image with a real `gh`
448
+ but a shimmed `claude` is sent both. Splitting the payload would save part of
449
+ one transfer, paid once per workspace, in exchange for a matrix of half-lent
450
+ states every later step would have to reason about. (The *network* install is
451
+ already per tool: each install guards itself with its own `command -v`.)
452
+ - **No version sync.** A real `claude` already in the container is left alone
453
+ whatever its version. `dl` lends what is missing; it is not a package manager,
454
+ and keeping versions in step would mean deciding what to do when the container
455
+ is the newer one. The official binary self-updates in a long-lived workspace,
456
+ and rebuilding one re-provisions it from scratch. The single upgrade `dl` does
457
+ perform is replacing a shim with a real binary.
458
+
459
+ ### Turning it off
460
+
461
+ ```bash
462
+ DEVLAUNCH_NO_TOOLS=1 dl someone/repo
463
+ ```
464
+
465
+ | Variable | Description |
466
+ |----------|-------------|
467
+ | `DEVLAUNCH_NO_TOOLS=1` | Do not install `gh` or `claude` into workspaces. The setup pass still runs — one trip per `up`, which still names the container; only the installing is skipped |
468
+
469
+ Attaching to a workspace that is *already running* skips `devpod up`, and so skips
470
+ this too. A workspace started by something other than `dl` — or created before this
471
+ existed — picks the tools up on its next `dl <workspace> restart`.
472
+
473
+ ## A terminal beside the agent
474
+
475
+ Every workspace `dl` opens also has [zellij](https://zellij.dev) on `PATH`, which
476
+ buys one thing the other tools do not: an agent running in a container can open a
477
+ **second terminal next to itself**, in the same container, and you can attach to it
478
+ from anywhere to watch or to type.
479
+
480
+ Nothing has to cooperate for this. It does not come from your dotfiles, it does not
481
+ need an edit to any repo's `devcontainer.json`, and it works in images `dl` has never
482
+ seen — the same argument the rest of "Tools in every workspace" makes, for the same
483
+ reason: `dl` launches arbitrary repos.
484
+
485
+ ### Opening a pane from inside a session
486
+
487
+ From anywhere inside the container — including from a completely non-interactive
488
+ command, with no terminal attached to anything:
489
+
490
+ ```bash
491
+ zellij -s devlaunch action new-pane -- htop
492
+ ```
493
+
494
+ `-s <name>` is the form to use and the only one worth depending on. Bare
495
+ `zellij action new-pane` happens to work by falling back to the single running
496
+ session, which stops being a single session the moment there are two of them.
497
+
498
+ `devlaunch` is the session name `dl` creates and the one to name here.
499
+
500
+ ### Switching the wrap on
501
+
502
+ The session an agent opens panes into has to exist first, and creating it is
503
+ **off by default**:
504
+
505
+ ```bash
506
+ DEVLAUNCH_ZELLIJ=1 dl someone/repo -- claude -p "do the thing"
507
+ ```
508
+
509
+ | Variable | Description |
510
+ |----------|-------------|
511
+ | `DEVLAUNCH_ZELLIJ=1` | Before running `dl <spec> -- <command>`, make sure a zellij session named `devlaunch` exists in the container, so the command can open panes into it |
512
+
513
+ With it off, no invocation changes meaning at all — that is what off means here, and
514
+ it is why the switch exists rather than the behaviour simply being on.
515
+
516
+ **The command runs beside the session, not inside a pane of it.** That is deliberate.
517
+ Putting the command in a pane would hand its stdin, stdout and exit status to zellij,
518
+ and all three are things `dl` promises to leave alone: `dl <ws> -- cmd > file` has to
519
+ put the command's own output in the file, and a failing command has to come back with
520
+ its own status. Since `zellij -s <name> action new-pane` works perfectly well from a
521
+ command that is in no session at all, running beside the session costs nothing and
522
+ delivers the same pane.
523
+
524
+ **The interactive session of a bare `dl <workspace>` is untouched, switched on or
525
+ off.** An interactive attach sends no command for the wrap to attach to — that is
526
+ exactly what gets it a terminal from devpod — and giving it one would cost either the
527
+ terminal or a round trip in front of every shell. You land in an ordinary login shell
528
+ with `zellij` on `PATH`, so `zellij attach -c devlaunch` gets you the session, and any
529
+ panes an agent has opened in it, whenever you want them.
530
+
531
+ There is one exception, and it is a pleasant one: if you also run with
532
+ `DEVLAUNCH_DOTFILES_ON_ATTACH=1`, that refresh is a command, so it gets wrapped like
533
+ any other and the session is already there when the shell arrives.
534
+
535
+ ### Existing workspaces
536
+
537
+ zellij arrives on the setup pass, which runs on every `devpod up`. So a workspace
538
+ that predates this picks it up on its next **`dl <workspace> restart`** — a full
539
+ `dl <workspace> recreate` also works but is not needed, because nothing here is a
540
+ bind mount and mounts are the thing that only lands at container creation.
541
+
542
+ Attaching to a workspace that is *already running* skips `devpod up` and so skips
543
+ this too, which is what makes the restart necessary rather than automatic.
544
+
545
+ ### What it costs
546
+
547
+ Almost nothing, and that was measured rather than assumed. zellij is a conda-forge
548
+ package installed by pixi into the container, so it lands in the shared package cache
549
+ above and every container after the first extracts rather than downloads:
550
+
551
+ | | |
552
+ |---|---|
553
+ | **Warm install** (shared cache populated) | 0.56s / 0.23s / 0.23s over three fresh containers |
554
+ | **Cold install** (empty cache) | 3.0s, filling 167MB of shared cache |
555
+ | **Every launch after the first** | one `command -v`; the whole setup pass measured at 50ms |
556
+
557
+ **It can never fail a launch.** Provisioning zellij is a stage of the setup pass, so a
558
+ container with no network, no pixi and no way to get either reports the stage as
559
+ failed, by name, and then opens exactly as it would have. A container that ends up
560
+ without zellij still works; with the wrap on, the command still runs, because the
561
+ session setup is allowed to fail and the command runs regardless.
562
+
563
+ `DEVLAUNCH_NO_TOOLS=1` turns this off along with the rest of tool provisioning —
564
+ installing zellij is tool provisioning, where naming a container is not.
565
+
566
+ ## The shared pixi package cache
567
+
568
+ Every container `dl` creates gets one host directory bound into it, and
569
+ `PIXI_CACHE_DIR` pointed at it, so that dotfiles which provision their tools with
570
+ `pixi global sync` download each package once per machine instead of once per
571
+ container:
572
+
573
+ | | |
574
+ |---|---|
575
+ | **On the host** | `$XDG_CACHE_HOME`, or `~/.cache`, then `devlaunch/pixi` |
576
+ | **In the container** | `/var/tmp/devlaunch-pixi` |
577
+
578
+ Measured on the profile this was built for — 23 pixi-global environments — a
579
+ container with a cold cache spends 62–113 s and downloads 1.2 GB; one that finds
580
+ the packages already there finishes in 18–28 s and fetches nothing. Two containers
581
+ syncing against it at the same time is fine: the downloads are content-addressed
582
+ and rattler takes a lock per package.
583
+
584
+ **Deleting it is always safe, at any moment, including while containers are
585
+ running.** It holds nothing but downloaded package archives — every one of them
586
+ re-fetchable from the network, and none of them referenced by a path anything
587
+ inside a container has stored. The worst a deletion costs is the next container's
588
+ download.
589
+
590
+ ```bash
591
+ rm -rf ~/.cache/devlaunch/pixi
592
+ ```
593
+
594
+ `dl --purge` takes it away with the rest of `~/.cache/devlaunch/`, for the same
595
+ reason.
596
+
597
+ Two things it deliberately is not. It is **not the host's own**
598
+ `~/.cache/rattler/cache`: containers write into it as their own remote user,
599
+ whose uid only happens to match yours, and a `pixi clean cache` you run for your
600
+ own reasons must not be able to pull packages out from under a running container.
601
+ And it is **not a shared `PIXI_HOME`** — the installed environments and their
602
+ trampolines are baked with absolute paths, and two containers sharing one
603
+ environment tree is [pixi#5476](https://github.com/prefix-dev/pixi/issues/5476).
604
+ Only the download cache is shared, which is the part that is safe to share.
605
+
606
+ If the directory cannot be created, or is not there when the launch reaches it —
607
+ a full disk, a read-only cache home, a cache swept between the two — the launch
608
+ goes ahead without the mount and the container downloads its own packages,
609
+ exactly as it did before this existed.
610
+
611
+ **Sharing requires the container's user to be able to write the directory**,
612
+ which in practice means its uid matches yours or it is root. The mount carries
613
+ host ownership through unchanged, and pixi does not degrade to reading a cache
614
+ it cannot write: pointing `PIXI_CACHE_DIR` at a directory owned by another uid
615
+ fails the install outright (`Permission denied` on the repodata, exit 1) even
616
+ when every package it wants is already in there. So an image whose remote user
617
+ is neither root nor your uid does not merely lose the sharing — its `pixi global
618
+ sync` fails, and its tools do not get provisioned.
619
+
620
+ `dl` cannot see the container's uid before it launches, so it cannot decide this
621
+ for you. In practice the common case is safe: every mainstream base declares a
622
+ remote user at uid 1000, which is the first human user on a Linux host. If you
623
+ hit the failure, the fixes available to you are to run that image as your own
624
+ uid, or to take the cache out of play for it (`rm -rf ~/.cache/devlaunch/pixi`
625
+ recovers a directory an earlier container left owned by someone else).
626
+
627
+ ### Existing containers, and what a recreate is for
628
+
629
+ **A mount lands only when a container is created.** devpod re-applies
630
+ `--workspace-env` on every `up`, but it will not add a bind mount to a container
631
+ that already exists — passing `--mount` there is a silent no-op. So a container
632
+ built before this feature, or before a change to where the mount lands, keeps
633
+ whatever it was created with until `dl <workspace> recreate`, and only then
634
+ picks the current arrangement up.
635
+
636
+ In between, `PIXI_CACHE_DIR` points at `/var/tmp/devlaunch-pixi` with nothing
637
+ mounted on it. That is a working private cache, not a failure — `/var/tmp` is
638
+ world-writable in every image, so pixi creates the directory and fills it. The
639
+ container re-warms itself and simply never shares, abandoning whatever pixi had
640
+ already warmed in its default location. **This is the reason the container-side
641
+ path is under `/var/tmp` rather than somewhere tidier like `/var/cache`:** a
642
+ target whose parent is root-owned is a hard `pixi global sync` failure on every
643
+ container that predates it, not a lost optimisation.
644
+
645
+ One older breakage needs the recreate rather than a restart. Devlaunch briefly
646
+ mounted this cache inside `~/.cache`, which left that directory root-owned in
647
+ any image that ships no `~/.cache` of its own. `$HOME` lives on the container's
648
+ own layer, so `dl <workspace> stop` and a fresh `up` keep the root-owned
649
+ directory; `dl <workspace> recreate` gets a new layer where `~/.cache` is the
650
+ user's own again.
651
+
652
+ ## Global Commands
653
+
654
+ | Command | Description |
655
+ |---------|-------------|
656
+ | `dl --ls` | List all workspaces |
657
+ | `dl --ls --json` | The same list as JSON, with each workspace's repo, branch, state and [unsaved work](#cleaning-up-workspaces) — for tools that decide what to clean up |
658
+ | `dl --ls --size` | Add [what deleting each workspace would free](#how-much-disk-a-workspace-costs). Opt-in: it walks every file in the clone |
659
+ | `dl --install` | Install shell completions |
660
+ | `dl --prune [-y] [--force]` | Remove [the clone directories no workspace opens any more](#pruning-the-clones-nothing-opens) — and nothing else |
661
+ | `dl --reconcile [-y]` | Re-point [devpod workspaces whose recorded source folder no longer holds a checkout](#reconciling-records-that-disagree) at the clone that does |
662
+ | `dl --purge [-y]` | Remove all devlaunch data — [the workspaces devlaunch created](#what-purge-deletes), and its caches |
663
+ | `dl --refresh` | Refresh completion cache |
664
+ | `dl --help, -h` | Show this help |
665
+ | `dl --version` | Show version (an editable install also names the tree it runs from) |
666
+
667
+ A released install prints the version and nothing else. An install made in
668
+ editable mode says so and names the checkout it resolves to, so two builds of
669
+ the same version are told apart at a glance:
670
+
671
+ ```bash
672
+ $ dl --version
673
+ dl 0.0.9
674
+
675
+ $ dl-next --version # editable install of a working tree
676
+ dl 0.0.9 (dev, editable from /path/to/your/devlaunch)
677
+ ```
678
+
679
+ `aid --version` reports the same thing under its own name. The provenance comes
680
+ from the installed package's own PEP 610 metadata; an install that records none
681
+ just prints the bare version.
682
+
683
+ ### What purge deletes
684
+
685
+ devpod's workspace list is shared. A workspace you made with `devpod up`, or that
686
+ another tool made, sits in the same list as the ones `dl` made, and `dl --purge`
687
+ has no business destroying it. So it deletes only the workspaces devlaunch
688
+ created — the clones it made under its own cache directory (`$XDG_CACHE_HOME` or
689
+ `~/.cache`, then `devlaunch/repos/<owner>/<repo>/<id>`), which is exactly the
690
+ directory the purge is about to remove anyway. Everything else keeps working
691
+ afterwards, because nothing a purge touches backs it.
692
+
693
+ Anything it is leaving is named before it asks:
694
+
695
+ ```
696
+ $ dl --purge
697
+ This will remove all devlaunch data:
698
+ - 4 DevPod workspace(s)
699
+ - /home/you/.cache/devlaunch/ (workspace clones, repo caches, the shared pixi cache, completions)
700
+
701
+ Leaving 2 workspace(s) devlaunch did not create:
702
+ - pythontemplate
703
+ - my-hand-made-workspace
704
+
705
+ Are you sure? [y/N]
706
+ ```
707
+
708
+ Three things `dl` does create are in that second list rather than the first.
709
+ `dl ./some/path` and `dl <git-url>` open a source `dl` did not clone, so it
710
+ cannot tell them from a workspace you made by hand — and a `config.toml` that
711
+ points `repos_dir` outside the cache puts the clones somewhere `--purge` does not
712
+ remove either, so those are left too. Delete any of them with `dl <workspace> rm`.
713
+ Erring this way is deliberate — a purge that skips one of your own workspaces
714
+ costs you a command, and the other kind of mistake costs you work you cannot get
715
+ back.
716
+
717
+ #### When part of the cache will not go
718
+
719
+ A container writes into its clone as its own user — `vscode`, uid 1000, in the
720
+ standard devcontainer base image. Where your host user is uid 1000 too, nothing
721
+ here comes up. Where it is not — CI, a shared machine, a container running as
722
+ root, or devlaunch developed inside its own devcontainer — the directories the
723
+ container made cannot be emptied by you, and the purge cannot remove them.
724
+
725
+ It removes everything else anyway, and names what is left:
726
+
727
+ ```
728
+ $ dl --purge -y
729
+ Removed what was permitted under /home/you/.cache/devlaunch. These refused:
730
+ - /home/you/.cache/devlaunch/repos/blooop/bencher/bencher-main-kivagede: Permission denied
731
+
732
+ Usually this means a container wrote them as a different user, and:
733
+ sudo rm -rf '/home/you/.cache/devlaunch'
734
+ clears them. Check the reasons above first -- it does not fix all of them.
735
+ devlaunch does not manage Docker images or volumes: the containers these workspaces used may still hold disk, and `docker system df` shows what Docker is holding.
736
+ ```
737
+
738
+ That last line ends every purge, including one that found nothing to purge and
739
+ one you answered `n` to — `dl --prune` ends on the same one, in the same words.
740
+ See [the disk neither command frees](#the-disk-neither-command-frees).
741
+
742
+ Exit status is `1`, because a clone you were told would go is still on disk. It
743
+ used to be `1` with the *whole* cache still standing: the first refusal stopped
744
+ the purge, so the completion caches, `metadata.json` and every other clone
745
+ survived on account of one directory.
746
+
747
+ When **none** of it goes — nothing under the cache came away at all, which is
748
+ what a symlinked cache root gives you, or one that cannot even be looked at, or
749
+ one whose every entry refused — the headline says that instead of claiming a
750
+ partial success:
751
+
752
+ ```
753
+ $ dl --purge -y
754
+ Removed nothing under /home/you/.cache/devlaunch. These refused:
755
+ - /home/you/.cache/devlaunch: Permission denied
756
+ ```
757
+
758
+ The report underneath is the same one, and so is the exit status: `0` means the
759
+ cache is gone and nothing else does, which is the only distinction a script can
760
+ act on. Removed everything, removed what it was permitted to and removed nothing
761
+ are three outcomes rather than two, and the sentence is where the third one
762
+ lives — because it is the one that decides whether you still have clones to go
763
+ and look for.
764
+
765
+ What is listed is the directory, once — not the hundreds of files inside it.
766
+ Unlinking needs write permission on the directory rather than on the file, so
767
+ every entry in that clone refuses separately and they are all the same fact.
768
+ Two *separately* unwritable directories on one path are two lines, though,
769
+ because clearing the inner one would leave the outer one just as stuck.
770
+
771
+ Each line carries what the system actually said. A container running as another
772
+ user is the common cause, but a read-only mount, `chattr +i` and a busy
773
+ mountpoint all land here too — and `sudo rm -rf` does not fix those, which is
774
+ why the report offers the cause rather than asserting it.
775
+
776
+ If you have **moved your cache** by making `~/.cache/devlaunch` a symlink, a
777
+ purge refuses it and names the target rather than following it. Remove the real
778
+ directory yourself if you meant to: following the link would empty a directory
779
+ you never named, and removing just the link would report a clean sweep while
780
+ your clones sat on the other volume.
781
+
782
+ ### Pruning the clones nothing opens
783
+
784
+ A workspace per branch means clone directories accumulate under the cache, and
785
+ until now nothing removed them: measured on one host, **52 clone directories for
786
+ 17 live devpod workspaces — 37 of them attached to nothing, 4.00 GB, against
787
+ 7.86 GB still in use.** `--purge` is the wrong tool for that, being
788
+ all-or-nothing: the only way to get the 4 GB back was to destroy the 7.86 GB
789
+ too, and every bare cache with it.
790
+
791
+ `dl --prune` removes exactly the clone directories no live workspace opens. It
792
+ never deletes a devpod workspace, a container, an image or a volume, never
793
+ touches a repo's `.bare` cache (0.08 GB for seven repos, and it is what makes
794
+ the next clone of a repo fast), and never looks outside
795
+ `<cache>/devlaunch/repos`. Every directory it finds is one of three things:
796
+
797
+ - **a live workspace opens it** — kept, and named with the workspace that has
798
+ it. "Opens" means at *or under*: a workspace opened on a subdirectory of a
799
+ clone still needs the clone;
800
+ - **nothing opens it** — removed, unless it holds work that exists nowhere else,
801
+ or `git` would not say what it holds. A clone a container wrote as another
802
+ user is unreadable rather than empty, and "cannot tell" is kept, not removed;
803
+ - **`dl`'s records and devpod's disagree about it** — kept, always. This is
804
+ [#88](https://github.com/blooop/devlaunch/issues/88)'s shape. On that ticket's
805
+ host, 36 devpod workspaces out of 39 recorded a source folder that was gone or
806
+ was a config-only stub, while the real checkout sat beside it under a newer
807
+ naming scheme — so a perfectly healthy clone was opened by nobody, and the
808
+ stub was the only thing anything pointed at. `--prune` will not guess which
809
+ clone such a workspace needs: it keeps every clone of that repository and
810
+ names the record to go and fix. `--force` does not move any of them.
811
+ [`dl --reconcile`](#reconciling-records-that-disagree) is what fixes them.
812
+
813
+ Note that *every* directory two levels under `<cache>/devlaunch/repos` is a
814
+ candidate — a stray directory somebody left there is looked at like any other.
815
+ The cache is `dl`'s to manage; things that are not clones do not belong in it.
816
+ But `git` cannot say what a directory that is not a repository holds, and
817
+ "cannot say" is kept rather than removed, so clearing junk out of the cache
818
+ takes `--force`. That is the same refusal a clone with a half-written `.git`
819
+ gets, and deliberately so: telling the two apart would mean `--prune` forming
820
+ its own opinion about a directory `dl <workspace> rm` already refuses on.
821
+
822
+ ```
823
+ $ dl --prune
824
+ Clone directories under /home/you/.cache/devlaunch/repos:
825
+
826
+ Removing 2 that nothing references -- 1.4 GiB:
827
+ - /home/you/.cache/devlaunch/repos/blooop/bencher/bencher-test1-pipagito (1.1 GiB)
828
+ - /home/you/.cache/devlaunch/repos/blooop/devlaunch/devlaunch-t1-vebilote (317.0 MiB)
829
+
830
+ Leaving 3:
831
+ - /home/you/.cache/devlaunch/repos/blooop/devlaunch/devlaunch-main-zovomobo: workspace devlaunch-main-zovomobo still opens it
832
+ - /home/you/.cache/devlaunch/repos/blooop/wayfinder/wayfinder-devlaunch-kilarabo: holds 2 unpushed commit(s) -- add --force to remove it anyway
833
+ - /home/you/.cache/devlaunch/repos/blooop/rockerc/rockerc-main-ludomane: devpod lists workspace rockerc-main-ludomane and sources it at /home/you/.cache/devlaunch/repos/blooop/rockerc/main; see devlaunch#88
834
+
835
+ Dropping 12 record(s) of directories already gone.
836
+
837
+ Are you sure? [y/N]
838
+ ```
839
+
840
+ `-y` skips the question. **A clone holding uncommitted or unpushed work is kept
841
+ and named**, in the same words [`dl <workspace> rm`](#cleaning-up-workspaces)
842
+ refuses in — 13 of those 37 stale clones did, two of them with real unpushed
843
+ commits, so this is load-bearing rather than a formality. `--force` promotes
844
+ that one case and nothing else. Erring this way costs you a flag; erring the
845
+ other way costs work that cannot be recovered.
846
+
847
+ The sizes are the same *exclusive* bytes `dl --ls --size` reports, and they mean
848
+ [the same thing](#how-much-disk-a-workspace-costs): what removing that directory
849
+ would actually free, not what `du` would print. Where a walk could not read
850
+ something the figure reads `≥` and so does the total, because a floor printed as
851
+ a total is a cleanup tool telling you a directory is small when it is not.
852
+
853
+ Directories that will not come away are named the same way [a purge names
854
+ them](#when-part-of-the-cache-will-not-go), the rest still go, and the exit
855
+ status is `1`.
856
+
857
+ **Nothing here runs on its own.** A full scan measured 1017 ms on that host —
858
+ about two warm launches — and it gets slower exactly as the cache gets fuller,
859
+ so it is never on a launch path and never folded into `dl --ls`. Answering `n`
860
+ *is* the read-only view; there is no separate flag for it. It costs one
861
+ `devpod list` to build the plan and no `devpod status` at all, because whether a
862
+ workspace is running has no bearing on whether a directory is opened by one. A
863
+ run you say yes to pays a second `devpod list` before it removes anything, and
864
+ classifies every directory again: a launch that finishes while the report is on
865
+ screen registers a workspace for one of the directories in the plan, and that is
866
+ the one thing the plan cannot be re-checked against from disk. The set you
867
+ approved can shrink between the report and the act. It can never grow.
868
+
869
+ It also drops the `metadata.json` records of directories that are already gone.
870
+ That file was append-only in practice — 49 records for 17 live workspaces on the
871
+ same host — and this is the first thing that prunes it.
872
+
873
+ #### The disk neither command frees
874
+
875
+ Both commands end on the same line, in the same words:
876
+
877
+ ```
878
+ $ dl --prune -y
879
+ ...
880
+ Removed 2 clone director(ies) -- 1.4 GiB.
881
+ devlaunch does not manage Docker images or volumes: the containers these workspaces used may still hold disk, and `docker system df` shows what Docker is holding.
882
+ ```
883
+
884
+ The gigabytes a cleanup reports are usually not the ones you are looking for. On
885
+ the host this was measured, `--prune` had 4.00 GB of stale clones to give back
886
+ while `docker system df` read **86.5 GB of reclaimable images, 43.18 GB of
887
+ volumes and 13.88 GB of build cache** — an order of magnitude more, sitting
888
+ behind a command that had just said "Removed". Saying nothing is what makes a
889
+ freed figure read as *all* of it, so both commands say this instead, whether they
890
+ removed 40 clones, found nothing to remove, or were answered `n` at the
891
+ confirmation. The report you get for saying `n` is a reason to print it, not an
892
+ exception: that is where somebody is deciding what is worth deleting.
893
+
894
+ **It is a sentence, not a measurement.** `dl` runs no `docker` command to print
895
+ it, so there is nothing to be slow and nothing to fail where Docker is absent,
896
+ stopped, or reachable only as another user. The figures above are this README's,
897
+ from the host it was measured on, not from your machine — `docker system df` is
898
+ where yours are.
899
+
900
+ **And it points rather than offers.** There is deliberately no `dl` flag that
901
+ removes an image, and no list of image ids here to paste into `docker image rm`.
902
+ Images devpod builds carry no devlaunch or devpod label, so any list `dl` printed
903
+ would be a guess at which of them belong to these workspaces, and `docker image
904
+ prune -a` is not scoped to devlaunch at all — it would take images built by
905
+ everything else on the machine. Deleting them is a decision with your own
906
+ containers on the other side of it, and `docker system df` is the tool that shows
907
+ you what it costs.
908
+
909
+ ### Reconciling records that disagree
910
+
911
+ `dl` keeps its own record of every workspace, and devpod keeps one too. They
912
+ agree until the naming that connects them moves — and it did move once, when
913
+ workspace ids and clone-directory names gained a hashed suffix. `dl`'s records
914
+ were migrated to the new naming; devpod's were not, because nothing knew to
915
+ touch them. On the host that reported it, **36 of 39 devpod workspaces recorded
916
+ a source folder that was missing, or was a stub with no `.git` in it**, while the
917
+ real checkout sat next to it under the new name. Nothing was deleted and nothing
918
+ was corrupted: `dl` was simply asking devpod about workspaces devpod had never
919
+ been given, and devpod was answering correctly that there were none.
920
+
921
+ Two things fix that, and they are different jobs. `dl` now **writes the devpod
922
+ workspace id down** when it creates a workspace, so the naming can move again
923
+ without taking anything with it — that is automatic and needs no command. It
924
+ does nothing for the records that already disagree, because they were written
925
+ before there was a field to write it in. `dl --reconcile` is for those:
926
+
927
+ ```
928
+ $ dl --reconcile
929
+ devpod workspaces sourced under /home/you/.cache/devlaunch/repos at something that is not a clone:
930
+
931
+ Re-pointing 2:
932
+ - devlaunch-main: .../blooop/devlaunch/main -> .../blooop/devlaunch/devlaunch-main-zovomobo
933
+ - bencher-test1: .../blooop/bencher/test1 -> .../blooop/bencher/bencher-test1-pipagito
934
+
935
+ Each of these needs `dl <workspace> recreate` afterwards: the container
936
+ still has the old source bind-mounted, and no record change moves a mount.
937
+
938
+ Leaving 1, which dl will not guess at:
939
+ - rockerc-main (.../blooop/rockerc/main): no clone of that repository answers to this name
940
+
941
+ Nothing here is deleted. `dl <workspace> rm` is how one goes, if it should.
942
+ ```
943
+
944
+ It matches the two sides **by path, never by id** — the id is the thing that
945
+ changed, so it connects nothing, while the source folder devpod kept still names
946
+ the owner and the repository exactly, and its last component still names the
947
+ branch in one of the three ways `dl` has named a clone directory. Where that
948
+ match is not unique it is refused rather than guessed: a clone a live workspace
949
+ already opens — at it, or anywhere under it — is never taken from it, a clone two
950
+ dead records both match is claimed by neither, and a name that two clones answer
951
+ to (the old flattened spelling turned `feature/auth`, `feature auth` and
952
+ `feature:auth` all into `feature-auth`) adopts neither of them. If a live
953
+ workspace's source cannot be followed at all, the whole command stops the way
954
+ `dl --prune` does, because such a workspace could be holding any of the clones on
955
+ offer. **Nothing is ever deleted.** A workspace `dl` cannot match
956
+ is named and left exactly where it is, because whether a workspace is finished
957
+ with is not something `dl` can know, and the two mistakes are not the same size.
958
+
959
+ Run it as often as you like — a repaired workspace is no longer sourced at a
960
+ non-checkout, so a second run finds nothing to do.
961
+
962
+ **A re-pointed workspace still needs rebuilding.** Its container was built with
963
+ the dead path bind-mounted into it, and changing a record does not move a mount.
964
+ `dl <workspace> recreate` is what finishes the repair, and it is the step that
965
+ needs Docker.
966
+
967
+ **Do not point an old `dl` at a reconciled cache.** A `dl` from before the naming
968
+ changed derives the old directory name, does not find it, and treats the launch
969
+ as a cold one: it clones a second directory under the old name, registers a
970
+ second devpod workspace, and rewrites that branch's record with the old naming
971
+ and an empty workspace id — undoing the repair for that one workspace, and
972
+ leaving you two clones of the branch. It is not destructive and the next
973
+ `dl --reconcile` sorts it out, but a machine that runs both builds against one
974
+ cache will keep re-breaking. Upgrade the old one, or give it its own
975
+ `XDG_CACHE_HOME`.
976
+
977
+ ### Cleaning up workspaces
978
+
979
+ One workspace per branch means workspaces accumulate, and `--purge` is the wrong
980
+ tool for tidying: it is all-or-nothing and takes the caches with it.
981
+
982
+ **devlaunch does not decide which workspaces are finished.** Whether a piece of
983
+ work is over is a fact about a ticket, a review, or somebody's intent, and `dl`
984
+ knows about clones and containers. Inferring it from the branch — merged into
985
+ the default, or deleted from the remote — was tried and dropped: it reads like a
986
+ git fact but is a guess at intent, and it cannot tell a squash-merged branch
987
+ from an abandoned one. So `dl` supplies the two halves a tool that *does* know
988
+ needs, and that tool drives the cleanup:
989
+
990
+ ```bash
991
+ dl --ls --json # what exists, and what each workspace holds
992
+ dl --ls --json --size # ...and what removing each one would free
993
+ dl <workspace> rm # remove one
994
+ ```
995
+
996
+ The JSON reports, per workspace: `id`, `devlaunch` (did `dl` create it),
997
+ `repo`, `branch` (what the workspace was made for), `checkedOut` (what its clone
998
+ is on now, which can differ), `path`, `state`, `lastUsed`, and — the field a
999
+ cleanup tool must not ignore — `unsaved`:
1000
+
1001
+ ```json
1002
+ {
1003
+ "id": "devlaunch-wayfinder-devlaunch-80-ladepomi",
1004
+ "devlaunch": true,
1005
+ "repo": "blooop/devlaunch",
1006
+ "branch": "wayfinder/devlaunch-80",
1007
+ "state": "Stopped",
1008
+ "unsaved": {
1009
+ "wouldLose": "2 uncommitted change(s) (pixi.lock, notes.md) and 1 unpushed commit(s)"
1010
+ }
1011
+ }
1012
+ ```
1013
+
1014
+ `unsaved` is an object with exactly one key, and the key says which of three
1015
+ answers it is:
1016
+
1017
+ | `unsaved` | Meaning |
1018
+ | --- | --- |
1019
+ | `{"nothingToLose": true}` | Everything in the clone exists on a remote too. Deleting it costs nothing. |
1020
+ | `{"wouldLose": "<what>"}` | Uncommitted changes (untracked files included), commits no remote has, or both. |
1021
+ | `{"couldNotTell": "<why>"}` | `git` could not read the clone as a repository — a half-removed `.git`, an interrupted delete. The files are still there and nothing has established that they exist anywhere else. |
1022
+
1023
+ The changed paths are named, not just counted, and that matters more than it
1024
+ looks: a devcontainer that runs a package install in its `postCreateCommand` can
1025
+ leave a tracked lockfile modified in *every* workspace it builds — this repo's
1026
+ own does — and as a bare count that is indistinguishable from an hour of unsaved
1027
+ work. A cleanup tool believing the count would then never clean anything. Named,
1028
+ it is judgeable. A workspace `dl` did not create reports `devlaunch: false` and
1029
+ no `unsaved` — `unsaved` is `null` exactly where `devlaunch` is `false`, and
1030
+ nowhere else: there is no clone of `dl`'s to protect, and it has no business
1031
+ inspecting your checkout. (`repo` and `branch` are a weaker test and not the
1032
+ same set: they come from `dl`'s metadata record, and a clone `dl` owns can have
1033
+ lost its record while the clone and the work in it are still on disk. That clone
1034
+ is inspected and reported like any other.)
1035
+
1036
+ **`dl <workspace> rm` refuses to delete a clone it would lose work from — when
1037
+ the recorded clone holds unsaved work, and when it cannot tell what that clone
1038
+ holds**, so a caller that forgets to read the field is still caught. (Recorded,
1039
+ because that is the directory the guard reads; the case with no record is
1040
+ neither, and is described below.)
1041
+
1042
+ ```
1043
+ $ dl blooop/repo@feature rm
1044
+ error: devlaunch-repo-feature-xyz holds 1 unpushed commit(s).
1045
+ Push or commit it, or run: dl blooop/repo@feature rm --force
1046
+ ```
1047
+
1048
+ ```
1049
+ $ dl blooop/repo@feature rm
1050
+ error: devlaunch-repo-feature-xyz: git could not read /home/…/repo/feature:
1051
+ fatal: not a git repository. devlaunch will not delete a clone it cannot
1052
+ check. Look at it, or run: dl blooop/repo@feature rm --force
1053
+ ```
1054
+
1055
+ That refusal is the only judgement `dl` makes here, and it is not about finished
1056
+ work — it is `dl` declining to destroy the only copy of something, including
1057
+ when it cannot prove there is another copy. Say `--force` if you mean it.
1058
+
1059
+ `--force` changes one more answer: an already-absent workspace counts as
1060
+ deleted, like `rm -f`. Unforced, `rm` reports devpod's refusal to delete a
1061
+ workspace it does not have; forced, the contract is the state afterwards, not
1062
+ that a delete happened — which is what lets the [cold benchmark's per-run
1063
+ reset](#measuring-launch-time) run before the first launch, when there is
1064
+ nothing to remove yet.
1065
+
1066
+ The guard reads `dl`'s metadata record, so the recorded directory is the one it
1067
+ asks about. (The delete does not always remove that same directory: when the
1068
+ recorded path is not on disk it falls back to a derived one. That divergence is
1069
+ older than this guard and is tracked as devlaunch#174.) One case is therefore
1070
+ neither a refusal nor a delete: a clone under `dl`'s cache that has **no** record — a metadata write
1071
+ that failed, a record pruned, a cache restored without one. The listing still
1072
+ reports what that clone holds, so `unsaved` is the field to read; but `rm`
1073
+ removes the devpod workspace, exits `0` without asking for `--force`, and leaves
1074
+ the clone on disk, because there is no recorded directory for it to remove
1075
+ either. Nothing is destroyed, and nothing then points at the clone: it is yours
1076
+ to keep or to `rm -rf` by hand.
1077
+
1078
+ [`wf`](https://github.com/blooop/wayfinder) is the caller this was built for: it
1079
+ names its branches after its tickets, so it knows which workspaces belong to
1080
+ finished work and removes those.
1081
+
1082
+ ### How much disk a workspace costs
1083
+
1084
+ `dl --ls --size` adds a `SIZE` column, and `dl --ls --json --size` adds a `disk`
1085
+ object beside the other per-workspace facts:
1086
+
1087
+ ```
1088
+ $ dl --ls --size
1089
+ WORKSPACE TYPE SOURCE SIZE LAST USED
1090
+ kinisi-ros-main-lubadaha local /home/…/repos/kinisi-robotics/kinisi_ros/main 64.9 MiB 2026-08-08 11:43:27
1091
+ my-own-checkout local /home/…/projects/scratch - 2026-08-01 09:12:04
1092
+ ```
1093
+
1094
+ **The number is what deleting that workspace would give back, not what `du`
1095
+ prints.** Those differ, and the gap is the point of the design. A repo is cloned
1096
+ once into a bare cache and every workspace clone hardlinks its git objects out
1097
+ of that one copy, so the objects exist once on disk however many workspaces
1098
+ share them. A size that walked each workspace on its own — which is what `du`
1099
+ does when you point it at one directory, counting the blocks every file in it
1100
+ occupies — bills each workspace for the whole shared pool.
1101
+
1102
+ The measurement the row above comes from, taken with the shipped code on one
1103
+ machine (Ubuntu 24.04, ext4, warm page cache) on a real clone of that repo made
1104
+ by `git clone` from the bare in `dl`'s own cache:
1105
+
1106
+ | | bytes |
1107
+ | --- | --- |
1108
+ | `du -s --block-size=1` on the clone alone | 353,230,848 |
1109
+ | what `dl --ls --size` reports for it | 68,050,944 |
1110
+ | what `dl --ls --size` reports for the bare it clones from | 651,264 |
1111
+ | `du -sc --block-size=1` over both together | 353,882,112 |
1112
+
1113
+ `du` bills that workspace **5.2x** what deleting it would actually free. The
1114
+ difference is a single 270,823,424-byte pack file with one link in the clone and
1115
+ one in the bare, so removing either end frees none of it.
1116
+
1117
+ **That sharing is a promise, not a coincidence, and a test holds it to that.**
1118
+ `git clone <path> <path>` hardlinks pack files by default, and the default is
1119
+ all that was ever keeping it true — a `file://` URL, an intermediate copy, or an
1120
+ explicit `--no-hardlinks` would each forfeit it with nothing failing and no
1121
+ warning printed. Measured on this repo — `du -sc` over the cache and each
1122
+ clone's `.git`, ext4, git 2.55.0 — that is 2400 KB for the cache plus one
1123
+ workspace against 4472 KB unshared, and 196 KB rather than 2268 KB of `.git` for
1124
+ every workspace after the first. So an integration test asserts the pack files are
1125
+ the cache's — same inode, more than one link — and that assertion goes red on
1126
+ all three. No clone flag is used to guard it: `--local` is already the default
1127
+ and does not even reject a `file://` source, and `--shared`/`--reference` were
1128
+ measured to leave a workspace that fails `git fsck` once the cache has fetched
1129
+ and gc'd, for a 2 KB saving.
1130
+
1131
+ Sharing does erode, in one measured way that is a safety property rather than a
1132
+ fault: when the cache repacks, an existing workspace's pack loses its second
1133
+ link and becomes that workspace's own complete copy, still passing `git fsck`.
1134
+ The workspace stops being cheap and never stops being valid — which is the trade
1135
+ `--shared` and `--reference` get wrong, and the reason they are not used.
1136
+
1137
+ **Large files are shared the same way, but nothing about `git clone` does it for
1138
+ you.** git-lfs objects are not git objects: the clone does not carry them at
1139
+ all, so a workspace of an LFS repo used to download the entire payload from the
1140
+ forge and keep a private copy of it in `.git/lfs/objects` — every workspace,
1141
+ every time, on top of the worktree copy. `dl` now makes the bare cache the
1142
+ repo's LFS store as well: the payload is fetched once into `<repo>/.bare/lfs`
1143
+ for the branch being launched, and each workspace materializes out of *that*,
1144
+ which git-lfs does by hardlinking. Measured with git-lfs 3.7.1 on ext4: the
1145
+ workspace's object file is the same `(st_dev, st_ino)` as the cache's, so its
1146
+ store costs nothing, and the materialization succeeds with the remote deleted
1147
+ from disk — the second workspace of an LFS repo touches the network for its
1148
+ large files not at all. What remains per workspace is the worktree copy, which
1149
+ is real bytes and cannot be shared: a container build has to be able to read
1150
+ them. If the cache cannot supply an object — a first launch offline, a payload
1151
+ the branch alone introduces — the old download from `origin` still runs, and a
1152
+ workspace left holding pointer files is retried on the next launch rather than
1153
+ written off.
1154
+
1155
+ Nothing about that is written into the workspace's `.git/config`, and that
1156
+ restraint is load-bearing rather than tidy: `dl` bind-mounts the *clone*
1157
+ directory into the devcontainer and `.bare` is a sibling that is not mounted, so
1158
+ an `lfs.storage` entry or an added remote naming a host path would break every
1159
+ `git checkout` of an LFS repo inside the container while working perfectly on
1160
+ the host. A test asserts the clone keeps exactly one remote, still pointing at
1161
+ the forge, and no `lfs.storage` at all.
1162
+
1163
+ So `dl` counts a file only when every one of its hardlinks lies inside the
1164
+ workspace being measured. Two consequences, both deliberate:
1165
+
1166
+ - **The sizes do not add up to the size of the cache.** Bytes shared between
1167
+ workspaces belong to none of them, because deleting any one frees none of
1168
+ them. They become the last workspace's the moment it is the last one — which
1169
+ is exactly when deleting it *would* free them. In the table above that is the
1170
+ last two rows read against each other: 68,702,208 reported bytes against
1171
+ 353,882,112 held.
1172
+ - **A workspace's size can change without the workspace changing**, when a
1173
+ sibling that was sharing with it goes away. That is the truth about shared
1174
+ storage.
1175
+
1176
+ A workspace `dl` did not create reads `-` (`null` in JSON): there is no clone of
1177
+ `dl`'s there to measure, and walking your own project directory is not `dl`'s to
1178
+ do. The table and the JSON decide that from the same rule — is the clone one
1179
+ `dl` put in its own cache, the same question `--purge` deletes by — so the two
1180
+ always name the same set of workspaces as measurable. Where a walk hits a
1181
+ directory it cannot read — a container writes into its
1182
+ clone as its own user, so this happens — the answer is a floor rather than a
1183
+ total: `≥2.0 MiB` in the table, and `{"atLeastBytes": …, "unreadable": 1}` in
1184
+ JSON instead of `{"exclusiveBytes": …}`. A partial measurement never comes back
1185
+ looking like a complete one.
1186
+
1187
+ **It is opt-in because it walks the whole clone.** Plain `dl --ls` is one devpod
1188
+ round-trip and no filesystem work at all, and the walk is O(files) with no
1189
+ ceiling. Measured with the shipped code on one machine — Ubuntu 24.04, ext4,
1190
+ warm page cache, five runs after a warm-up, the machine otherwise busy — a real
1191
+ 8,309-entry clone walked in 24–28 ms, this repo's own tree with its built
1192
+ environment inside it (9,124 entries) in 17–21 ms, and a 114,817-entry tree in
1193
+ 232–239 ms. No cold-cache figure is quoted because none was taken: dropping the
1194
+ page cache needs root on that machine. Those are one machine's numbers on warm
1195
+ cache and yours will differ, but the shape is the point — it grows with the file
1196
+ count, and a devcontainer that builds its environment *inside* the clone (this
1197
+ repo's own does) is most of that count. That is not a bill a listing should
1198
+ present unasked.
1199
+
1200
+ Docker images and named volumes are not counted: `dl` did not create the layer
1201
+ store and does not manage volumes. `docker system df` is the tool that knows —
1202
+ the same boundary [`--prune` and `--purge` name](#the-disk-neither-command-frees)
1203
+ when they finish.
1204
+
1205
+ ## Examples
1206
+
1207
+ ```bash
1208
+ dl # Select workspace with fzf
1209
+ dl devpod # Open existing workspace
1210
+ dl loft-sh/devpod # Create from GitHub
1211
+ dl blooop/devlaunch@main # Create from specific branch
1212
+ dl ./my-project # Create from local folder
1213
+ dl blooop/devlaunch code # Open in VS Code
1214
+ dl blooop/devlaunch -- make test # Run command in workspace
1215
+ dl blooop/devlaunch stop # Stop workspace
1216
+ ```
1217
+
1218
+ ## Features
1219
+
1220
+ - **Fuzzy Selection**: When called without arguments, uses fzf for interactive workspace selection
1221
+ - **Smart Completion**: Tab completion for workspaces, GitHub repos (owner/repo format), and paths
1222
+ - **GitHub Shorthand**: Use `owner/repo` instead of full URLs - automatically expands to `github.com/owner/repo`
1223
+ - **Branch Support**: Specify branches with `owner/repo@branch` syntax
1224
+ - **Fast Autocomplete**: Completion cache for ~3ms response time (vs ~700ms without cache)
1225
+ - **One Round-Trip Per Question**: every `devpod` call costs ~0.45s, far more than `dl` itself, so a command reads the workspace list at most once — and everything a container needs on the way in (naming it, then the tools probe) rides one setup pass, so an interactive `dl <ws>` and a one-shot `dl <ws> -- <cmd>` cost the same trips
1226
+
1227
+ ## Measuring launch time
1228
+
1229
+ Set `DEVLAUNCH_TIMING=1` and a `dl` command ends with one summary on stderr,
1230
+ naming each subprocess round trip and the total. Unset (or `0`) records nothing
1231
+ and prints nothing.
1232
+
1233
+ Captured from a real warm launch (the launch's own output elided):
1234
+
1235
+ ```bash
1236
+ $ DEVLAUNCH_TIMING=1 dl-next blooop/mcp-devtasks -- true
1237
+ ...
1238
+ dl-timing: devpod status 0.454s
1239
+ dl-timing: gh auth token 0.036s
1240
+ dl-timing: devpod ssh 1.952s
1241
+ dl-timing: total 2.444s (in-process, excluding interpreter startup)
1242
+ ```
1243
+
1244
+ ### The same launch, machine-readable
1245
+
1246
+ `DEVLAUNCH_TIMING=json` swaps that prose for one document on a single
1247
+ `dl-timing-json:` line, so a trend job can read a launch without scraping
1248
+ prose. It decomposes the launch into five **ownership-boundary stages** — one
1249
+ per party that could actually make it faster — with the round trips nested
1250
+ inside the stage that paid for them:
1251
+
1252
+ | stage | what it owns |
1253
+ |---|---|
1254
+ | `handoff` | the gap between the keystroke that resolved to this exec and dl starting (see the stamps below) |
1255
+ | `host-prep` | the host's own git work — the bare clone and its fetches, the lock waits, the LFS probe and, for an LFS repo, the cache's LFS fetch and the workspace's materialization out of it — and the `gh auth token` trip, wherever on the launch it falls |
1256
+ | `devpod-up` | the arm that gets a container running: the existence probe and, when it is not running, the `up` itself. On a warm launch that arm is the probe alone |
1257
+ | `tools` | the probe trip and the conditional lend, including staging the payload tar |
1258
+ | `attach` | the last trip, into the running command |
1259
+
1260
+ Two rules are worth knowing before reading one:
1261
+
1262
+ - **A stage that never ran is absent, not zero.** A warm launch reports no
1263
+ `host-prep` at all, because it did none. A stage that failed is present,
1264
+ timed up to the failure, and marked `failed`.
1265
+ - **A stage totals over its whole arm**, not just over its round trips, so the
1266
+ host-side work between two spawns is attributed rather than lost. Stages
1267
+ never double-count each other: `tools` runs inside the launch `devpod-up`
1268
+ brackets, and those seconds are charged to `tools` alone. The in-process
1269
+ stages therefore add up to the total — measured on a real cold launch below,
1270
+ they came to 20.834s against a 20.834s total.
1271
+
1272
+ `handoff` is the exception to that sum, and the only one: it ends where
1273
+ `total` begins, so it is time the process could not have measured from inside
1274
+ itself. A consumer adding stages up against the total leaves it out.
1275
+
1276
+ Two optional environment variables let whatever launches `dl` — a shell
1277
+ function, an agent front-end — close the loop on the time before dl existed.
1278
+ Both are Unix epoch seconds, which is what `date +%s.%N` prints:
1279
+
1280
+ | variable | meaning |
1281
+ |---|---|
1282
+ | `DEVLAUNCH_HANDOFF_T0` | the keystroke that resolved to this exec. Becomes the `handoff` stage — the only measurement of exec plus interpreter startup there is, since `total` begins after both |
1283
+ | `DEVLAUNCH_PREWARM_FIRED_AT` | when a prewarm (`dl <ws> up`) was fired for this workspace, if one was |
1284
+
1285
+ With the prewarm stamp set, the document also reports what that prewarm was
1286
+ worth: the head start it bought, and which shape the launch then took — `hit`
1287
+ (the workspace was already up), `partial` (this launch queued behind a prewarm
1288
+ still running) or `miss` (this launch ran the `up` itself). dl decides that,
1289
+ not the firer: a prewarm is fired and forgotten, so only the launch that
1290
+ followed can see whether it helped. **A stamp that is missing, unreadable, or
1291
+ ahead of this clock reports nothing** rather than a zero — an absent handoff
1292
+ and an instantaneous one are different facts, and a trend cannot tell them
1293
+ apart once one is written as the other.
1294
+
1295
+ Captured from a real warm launch with both stamps set (one line, wrapped and
1296
+ elided here for reading):
1297
+
1298
+ ```bash
1299
+ $ DEVLAUNCH_TIMING=json DEVLAUNCH_HANDOFF_T0=$(date +%s.%N) \
1300
+ DEVLAUNCH_PREWARM_FIRED_AT=... dl-next blooop/mcp-devtasks -- true
1301
+ ...
1302
+ dl-timing-json: {"total": 2.210768, "total_epoch": "in-process, excluding interpreter startup",
1303
+ "stages": [{"stage": "handoff", "seconds": 0.130542, "outcome": "ok", "spans": []},
1304
+ {"stage": "host-prep", "seconds": 0.027799, "outcome": "ok",
1305
+ "spans": [{"label": "gh auth token", "seconds": 0.02747}]},
1306
+ {"stage": "devpod-up", "seconds": 0.455188, "outcome": "ok",
1307
+ "spans": [{"label": "devpod status", "seconds": 0.455158}]},
1308
+ {"stage": "attach", "seconds": 1.726961, "outcome": "ok",
1309
+ "spans": [{"label": "devpod ssh", "seconds": 1.72661}]}],
1310
+ "prewarm": {"head_start_seconds": 42.489243, "shape": "hit"}}
1311
+ ```
1312
+
1313
+ Stages appear in the order the launch first entered them, and only the ones it
1314
+ reached appear at all — this warm launch built nothing and lent nothing, so
1315
+ there is no `tools` stage and no `devpod up` inside `devpod-up`. That
1316
+ `handoff: 0.131s` is the exec and the interpreter start, which nothing else
1317
+ measures.
1318
+
1319
+ The cold launch of the same repo, same host and session, decomposed as:
1320
+ `host-prep` 2.257s (`git clone --bare` 1.602 + `git fetch` 0.427 + workspace
1321
+ `git clone` 0.065 + LFS probe 0.002 + token 0.034), `devpod-up` 5.848s (`devpod
1322
+ up` 5.113 of it), `tools` 10.224s (probe trip 1.584 + `tools tar` 0.111 +
1323
+ transfer 8.445), `attach` 2.505s — 20.834s of stages against a 20.834s total.
1324
+
1325
+ For before/after numbers, `scripts/bench_launch.py` runs a command N times and
1326
+ reports the median — one command per side of a change:
1327
+
1328
+ ```bash
1329
+ python3 scripts/bench_launch.py -n 5 -- dl-next owner/repo -- true # warm launch
1330
+ ```
1331
+
1332
+ (`pixi run bench -n 5 -- ...` in the devcontainer.) It reports no median if any
1333
+ run fails, so a broken launch cannot pass as a fast one. See `bench_launch.py
1334
+ --help` for `--before` — the per-run reset that makes a *cold* median cold and
1335
+ whose `rm --force` also succeeds on the first run, when there is nothing to
1336
+ remove yet — and for why its wall clock and `dl-timing: total` are not the
1337
+ same quantity. For scale, on the host the session above was captured on, the
1338
+ warm median over 5 runs was 2.176s. Running the cold recipe exactly as the
1339
+ epilog writes it — `-n 5`, container recreated per run — gave a median of
1340
+ 15.899s (runs: 15.9, 20.0, 15.2, 15.7, 17.8). Read that as the cost of
1341
+ recreating a container, not of a first-ever launch: the reset removes the
1342
+ workspace but leaves the docker image layers and the bare clone cache, so
1343
+ every run after the first starts from both. A machine that must also pull or
1344
+ build the image pays more, by an amount this recipe does not measure — but the
1345
+ gap is large: an earlier 3-run median on this same host, reported as its first
1346
+ real launch, was 33.204s.
1347
+
1348
+ Every number in the two paragraphs above was copied into this prose by hand.
1349
+ `--record` is how that stops: it writes the same invocation as one JSON object
1350
+ a trend job can upload without anyone reading it.
1351
+
1352
+ ```bash
1353
+ python3 scripts/bench_launch.py -n 5 --record warm.json --shape warm \
1354
+ -- dl-next owner/repo -- true
1355
+ ```
1356
+
1357
+ The record holds the command, the run count, each run's wall time and
1358
+ per-stage seconds, and the medians of those — and nothing else, because a CI
1359
+ job stamps its own commit, clock and host better than this script can. Four
1360
+ things about it are load-bearing:
1361
+
1362
+ - **The median is the point, the runs are its evidence.** A trend compares a
1363
+ point against the immediately previous one, so N runs published as N points
1364
+ would read ordinary spread as a regression.
1365
+ - **A stage no run reported is absent, not zero.** A warm launch legitimately
1366
+ has no cold-path stages; a zero would claim the work happened instantly
1367
+ rather than not at all. A stage only *some* runs reported is a median over
1368
+ those runs, carrying a count of how many — a median of two and a median of
1369
+ five are not the same claim.
1370
+ - **Recording asks the launch for its stages** (`DEVLAUNCH_TIMING=json`), so a
1371
+ run that reports no timing document is an error and no record — same
1372
+ discipline as no median over a failed run.
1373
+ - **`--shape` labels the trend line** (`warm`, `cold-recreate`). It is the
1374
+ caller's to say: the same command benches either shape depending on
1375
+ `--before`, and a wrong label is worse in a trend than a missing one.
1376
+
1377
+ ### The trend on main
1378
+
1379
+ Every push to `main` runs `.github/workflows/bench.yml`, which benches both
1380
+ shapes on the runner and publishes one point per stage to
1381
+ <https://blooop.github.io/devlaunch/dev/bench/>. It can also be dispatched by
1382
+ hand. Reading it needs nothing but the chart; what follows is for changing it.
1383
+
1384
+ `scripts/bench_points.py` is the step between the two formats — bench records
1385
+ in, one flat array of trend cases out:
1386
+
1387
+ ```bash
1388
+ python3 scripts/bench_points.py warm.json cold-recreate.json --out bench.json \
1389
+ --require-stages-on cold-recreate
1390
+ ```
1391
+
1392
+ (`pixi run bench-points ...` in the devcontainer.) One case per stage the shape
1393
+ reported, plus that shape's own total: `warm / host-prep`,
1394
+ `cold-recreate / devpod-up`, `warm / total`. Six properties of it are
1395
+ load-bearing:
1396
+
1397
+ - **The published value is the median, and the case name is a key.** The trend
1398
+ compares a point against the immediately previous point and nothing older, so
1399
+ the de-noising has to have happened before publishing. Renaming a stage
1400
+ starts a new, empty series beside a frozen old one.
1401
+ - **The spread rides along as the point's error bar**, and the outside
1402
+ stopwatch (`wall=`) as part of its `extra`. Evidence beside the number,
1403
+ rather than a second trend line for the same launch that disagrees with the
1404
+ first by a constant.
1405
+ - **An absent stage is absent.** A warm launch lends nothing, so there is no
1406
+ `warm / tools` case at all — never a zero, which would claim an instantaneous
1407
+ lend and would drag the line down exactly where a regression should show.
1408
+ - **`--require-stages-on` fails the job rather than the trend.** The way this
1409
+ decomposition is expected to break is an absence, not a wrong number: a stage
1410
+ stops being emitted and the total keeps working. So the cold-recreate shape,
1411
+ where every stage is known to be present, asserts them all, and a run that
1412
+ lost one publishes nothing and goes red. Naming a shape that was not benched
1413
+ fails too — an assertion that covers nothing reads exactly like one that
1414
+ passed.
1415
+ - **A record that is missing or unreadable refuses the same way.** The step
1416
+ that writes the records can exit 0 without having written one, so the absence
1417
+ arrives here — and it prints `bench_points: <reason>` naming the file and
1418
+ writes nothing, like every other refusal, rather than a traceback.
1419
+ - **A regression alerts; it never gates.** The workflow is deliberately not a
1420
+ job in `ci.yml`: a job there would join the CI gate's `needs` by house
1421
+ convention and turn a noisy wall-clock measurement into a merge gate. A point
1422
+ above the threshold leaves a commit comment and a red mark on the chart, and
1423
+ the build stays green.
1424
+
1425
+ ## Worktree Backend
1426
+
1427
+ For git repositories, devlaunch uses an efficient worktree backend by default:
1428
+
1429
+ - **Efficient Storage**: Repos are cloned once to `~/.cache/devlaunch/repos/owner/repo/`, then git worktrees are created for each branch
1430
+ - **Shared Git Objects**: All branches share git objects, saving disk space
1431
+ - **Targeted Fetch**: A launch fetches only the one branch it is launching, so no launch waits on a repo-wide refresh
1432
+
1433
+ ### What you get when you push and immediately launch
1434
+
1435
+ - **Attaching to a workspace devpod already knows**: no git at all. The workspace
1436
+ is exactly as you left it; freshness inside it is your own `git pull`.
1437
+ - **A cold launch** (first time this branch is launched on this machine, or a
1438
+ clone devpod has forgotten): one targeted fetch of that branch, every time.
1439
+ Push upstream and immediately `dl` the branch and you get the pushed tip.
1440
+ - **A branch that does not exist yet**: created from the default branch's freshly
1441
+ fetched tip.
1442
+ - **Offline**: a warning, and the launch proceeds from whatever the cache holds.
1443
+ It only fails when there is nothing cached to launch from.
1444
+ - **Everything else** (other branches, tags, prunes) is refreshed by the
1445
+ background updater within the configured interval (default: 1 hour), which
1446
+ never blocks a launch.
1447
+
1448
+ ### Container Sharing Mode
1449
+
1450
+ Use `--shared` to share a single container across multiple branches of the same repo:
1451
+
1452
+ ```bash
1453
+ dl --shared owner/repo@branch1 # Creates container "owner-repo"
1454
+ dl --shared owner/repo@branch2 # Reuses "owner-repo" container
1455
+ ```
1456
+
1457
+ ### Pre-warming
1458
+
1459
+ Use `--warm` to prepare a workspace without attaching a shell:
1460
+
1461
+ ```bash
1462
+ dl --warm owner/repo@branch # Creates container in background
1463
+ ```
1464
+
1465
+ ## Shell Completion
1466
+
1467
+ After running `dl --install`, you get intelligent tab completion:
1468
+
1469
+ - Workspace names from your devpod list
1470
+ - Known GitHub owners and repositories from your workspaces
1471
+ - File/directory paths when starting with `./`, `/`, or `~`
1472
+ - All global flags (`--ls`, `--install`, etc.) and workspace commands
1473
+
1474
+ ### How the completion cache stays current
1475
+
1476
+ The data behind completions lives in `~/.cache/devlaunch/completions.json`, and
1477
+ building it means a `git ls-remote` per known repo — seconds of work. So it is
1478
+ rebuilt in the background at most once an hour (the same interval the worktree
1479
+ backend's background fetch sweep uses), and at most once per `dl` invocation. Commands
1480
+ that change your workspaces (starting, stopping or deleting one) rebuild it as
1481
+ soon as they finish, regardless of when it was last built. Commands with no use
1482
+ for it — `dl --help`, `dl --version` — do not touch it at all.
1483
+
1484
+ A branch created on a remote in the last hour may therefore not be offered yet.
1485
+ `dl --refresh` rebuilds the cache immediately and ignores the interval.
1486
+
1487
+ ## Development
1488
+
1489
+ This project uses [pixi](https://pixi.sh) for environment management.
1490
+
1491
+ ```bash
1492
+ # Run tests
1493
+ pixi run test
1494
+
1495
+ # Run the e2e suite: real devpod, real containers
1496
+ pixi run test-e2e
1497
+
1498
+ # Run full CI suite
1499
+ pixi run ci
1500
+
1501
+ # Format and lint
1502
+ pixi run style
1503
+ ```
1504
+
1505
+ `pixi run test` skips the e2e tests, which need devpod and a Docker daemon and
1506
+ build real containers. CI runs `pixi run test-e2e` in a job of its own, outside
1507
+ the Python matrix, on a throwaway runner — on every push to `main` and on every
1508
+ pull request, whatever branch that pull request targets. Stacked chains, where
1509
+ each link targets its predecessor rather than `main`, get the same CI as anything
1510
+ else.
1511
+
1512
+ Alongside the matrix and e2e there is a `gate` job that does nothing but fail
1513
+ unless every other job in that workflow succeeded. It exists so that a branch
1514
+ ruleset has one stable name to require rather than a list: requiring the jobs one
1515
+ by one means literal strings in a repository setting, which nobody reviews and
1516
+ which goes stale the moment a job is added or renamed — and a required check that
1517
+ no longer exists does not turn a merge red, it stops gating it. Adding a job
1518
+ means adding it to `gate`'s `needs`, in the same pull request, where it can be
1519
+ seen. It reaches only as far as its own workflow file, so the `prek` lint job is
1520
+ not behind it and has to be required alongside it.
1521
+
1522
+ Running it yourself is a different proposition. This repo's devcontainer carries
1523
+ a Docker daemon of its own, through the `docker-in-docker` feature, and pins the
1524
+ same devpod a host installs, so `pixi run test-e2e` from inside it builds its
1525
+ containers in there rather than on your Docker. You can also run it on a machine
1526
+ you do not mind it writing to — an ephemeral CI runner, say. It is skipped by
1527
+ default rather than gated on a container, because what it needs is a daemon, not
1528
+ nesting. Either way the suite exercises `dl --purge`, so it gives itself a
1529
+ private devpod namespace before collection begins — but the containers it builds
1530
+ are real ones, and it wants several minutes and a 1.25 GB image pull the first
1531
+ time.
1532
+
1533
+ Its skips mean one thing only. A test that opts out does so through
1534
+ `fixtures.e2e_guard.opt_out`, and any other skip is reported as a failure,
1535
+ because a run that could not reach a registry used to be indistinguishable from
1536
+ a healthy one. Every run also prints what it actually built:
1537
+
1538
+ ```
1539
+ --------------------------------- e2e session ---------------------------------
1540
+ 22 e2e tests attempted, 5 workspaces created: e2e-test-create, e2e-test-lifecycle, e2e-test-git, e2e-purge-devlaunchs, e2e-purge-hand-made
1541
+ ```
1542
+
1543
+ A run whose workspace-building tests built nothing does not pass: the shortfall
1544
+ is counted into the last line of the run, so `4 passed, 18 skipped` becomes
1545
+ `1 failed, 4 passed, 18 skipped`. A run with no workspace-building tests in it —
1546
+ `pytest -m e2e test/e2e/test_interactive_session.py`, say — has nothing to
1547
+ answer for and says so instead.
1548
+
1549
+ `DEVLAUNCH_E2E_WORKSPACE=<id>` opts in to the interactive-session tests, which
1550
+ attach to a workspace you already have running rather than building one.
1551
+
1552
+ The nested daemon is also why the devcontainer does not join the host's network
1553
+ namespace: a nested daemon needs a namespace of its own, or it co-manages the
1554
+ host's `docker0` bridge and writes its NAT rules into the host's netfilter
1555
+ tables.
1556
+
1557
+ ### Disk cost of the dev container
1558
+
1559
+ Opening a devcontainer for a branch costs about **2 GB on the host before you do
1560
+ anything in it**: ~600 MB of image layers unique to this image, a ~680 MB container
1561
+ writable layer, and a ~520 MB `<workspace>-pixi` volume.
1562
+
1563
+ The container carries its own Docker daemon, and that daemon's `/var/lib/docker`
1564
+ lives on a second named volume. One `pixi run test-e2e` plus a couple of nested
1565
+ workspaces puts **~2.3 GB** in there, and nothing garbage-collects it — the inner
1566
+ daemon reports ~45% of its images reclaimable with no reclaimer. Nested daemons
1567
+ share no layers with the host or with each other, so this is paid once per branch.
1568
+
1569
+ **Budget ~4 GB per branch you are actively developing and e2e-testing — about 12 GB
1570
+ for three concurrent branches.**
1571
+
1572
+ The time cost is cold pulls in a fresh nested daemon: the first `devpod up` inside a
1573
+ new container takes ~25s, ~16s of which is pulling a base image the host already has.
1574
+ Workspaces after that reuse it and take ~8s.
1575
+
1576
+ **These volumes are not reclaimed automatically.** `devpod delete` removes the
1577
+ container with `docker rm` and never touches volumes, and Docker never
1578
+ garbage-collects a *named* volume — so `<workspace>-pixi` and
1579
+ `dind-var-lib-docker-*` outlive the workspace that created them. To see what has
1580
+ piled up:
1581
+
1582
+ ```bash
1583
+ docker system df -v # under Local Volumes, LINKS 0 means no container uses it
1584
+ ```
1585
+
1586
+ Cross-check a name against `devpod list` before removing it with `docker volume rm`:
1587
+ a volume belonging to a live workspace shows `LINKS 1`.