Repository navigation
Release Notes
Added
clonetakes a Rust crate'starget. pando proposes it beside a
Cargo.toml(a Maven or sbttargetis left alone), and a new
worktree starts with the main checkout's build, copy-on-write: its
dependencies' fingerprints name no path of the checkout, so cargo
finds every one fresh and builds only the workspace's own crates,
where before each worktree built every dependency again on disk of
its own.pando checksays what a worktree costs on disk, for any project:
the gitignored trees its install and processes built from nothing,
each marked when the main checkout has oneclonecould share, and
the space the tools took outside it, such as a build cache keyed by
the checkout's path that fills again for every worktree. It prints a
line, andcheck --jsonhas it asdisk. pando judges none of it:
the agent brief says how to read it — which trees are safe to clone,
how to find a tool's setting that makes its cache independent of the
path and give it to the process through itsenv— and to check
again to see the difference.- The TUI takes the mouse. A click selects a worktree, a dialog's
option or a line of the log viewer, and a second click is⏎on it;
a click on a key hint presses its key,esc cancelcancelling; a
click on a tab of the log viewer, or on a process the log preview
names, shows that log; the wheel scrolls what is under the pointer.
Help lists all of it. Hold shift to select text while pando has the
mouse (option in iTerm2);[ui] mouse = falsein
~/.pando/config.tomlleaves it to the terminal, and the setup
screen always does, so its prompt can be selected and pasted. scripts/dev new, for contributors: a branch's worktree starts with
the main checkout'starget, cloned copy-on-write, so it builds only
pando and none of its dependencies again. Only ever a clone: on macOS
only where both sit on one APFS volume (cp -cwould copy
elsewhere), on Linux only as a reflink; otherwise cargo builds from
nothing, as before.scripts/dev use release, for contributors:pandois the release
you installed again, Homebrew's or the install script's, after
use <branch>made it a branch's build.pando updateon a branch's
build names it, andscripts/dev rmof the branchpandoruns goes
back to the release rather than to the main checkout's build.pando rmtakes several worktrees:pando rm feat/a feat/b fix/c
removes them side by side, says each as it finishes, and exits 1
naming any it could not remove;--yesand--forceapply to each.
One name reads exactly as before.pando start,stopandrestarttake several worktrees too:
pando start feat/a feat/bsettles every question first, one name
after another, then starts them side by side, each said as it
finishes; one that fails leaves the others, and the exit code is then- A question still open is the whole command's exit 3, and nothing is
started. Every flag applies to each; one name, or none, reads exactly
as before.
- A question still open is the whole command's exit 3, and nothing is
- Marks in the TUI:
+marks the row under the cursor, or unmarks it,
and moves down. A marked row has a+before the cursor's column,
and the list's title and the footer say how many are marked. A mark
stays on its worktree through a refresh, a re-sort or a filter, and
goes with it; esc unmarks every row, after clearing a filter. dwith rows marked removes every marked worktree from one dialog.
It lists each with what its removal stops or takes with it, says
which onlyFremoves (uncommitted changes) and which stay (locked,
the main checkout), and one confirm starts every removal at once.
ykeeps the ones that needFmarked, sodthenFfinishes
them.s,xandrwith rows marked start, stop or restart every marked
worktree at once, each on its own worker.xandrask for their
second press once for the whole set, while any of it runs; a marked
worktree that is up already, or busy with something else, is left
alone and said once. The other keys still act on the row under the
cursor.xon a worktree being started or restarted in the TUI cancels it,
where it saidalready busy starting …and a start could wait a
minute for a service. The start stops at its next safe point — a
readiness wait looks every quarter second, a hook already running
finishes first — through the undo a failure there runs, and the
worktree ends stopped.qtwice with a start in flight cancels it
and quits once it is undone, rather than leaving it half up;q
again quits at once.- The TUI shows a worktree another pando is removing —
pando rmin
another pane — as◌ removing, as it shows its own, ands,x,
r,dand sharing leave it alone, saying it is being removed. One
whose removal was stopped half way readsrm interrupted: nothing
starts it, anddfinishes the removal.
Fixed
pando updatesaid the update failed when Homebrew upgraded pando and
then exited 1 over something else: its periodicbrew cleanup, which
runs after the upgrade, refuses a cask from an untrusted tap. Whether
the update happened is now read from the installed pando's version, so
it reports the new version and says that what failed came after.- A panic in the TUI no longer leaves the terminal reporting focus
changes, which the shell then printed as^[[Iand^[[O. ls,status,open,start --waitand the TUI no longer wait for
another pando's lock: while armor astopheld it — adocker compose downcan take minutes — they printed nothing, andpando
opened onto nothing, until it let go. They now answer from the state
as it stands. A command that does have to wait sayswaiting for another pando to finishafter a second, rather than looking hung,
and the socket scan behind every read no longer holds the lock.pando stop --alland the TUI'sXstop every worktree at once:
one after another, each dev server slow to stop took its own five
seconds' grace, under the lock every other command waits for — four
of them, twenty seconds; now five. Their compose projects are asked
to stop together too.- A
pando stopno longer holds the lock while its processes take
their grace period to end, so two stops of two worktrees — the TUI's,
or two panes' — end together, and a refresh or a start of another
worktree never waits behind one: two worktrees whose servers ignore
SIGTERM stop in 5.3s at once, from 10.3s. A process a stop is taking
down is never shown as failed while it waits, nor a native database
it is stopping said to have exited, and a start of the same worktree
waits for the stop, says so, and then starts it fresh.restart,
stop --alland the TUI'sXstop the same way. - A folder of repositories is read one repository at a time no more:
its listing on every refresh and the TUI's first frame, its branches,
its pull requests, each set's git status, and each repository's base
fornew(a fetch, sometimes) are asked of every repository at once.
On six repositorieslstakes 0.15s, from 0.57s, and the TUI shows a
set's status as soon as its own repositories have answered. pando checkandpando newstop their install when they are told
to stop. A Ctrl-C that landed just before the install started, or a
killof pando alone — an agent's, CI's — reached pando and not the
install, and the check sat until the install ended by itself. Now no
hook starts once pando has been told to stop, and outside a terminal
a running one leads a process group of its own, which pando stops.- Removing a worktree no longer holds up every other command.
rm
stops what runs and marks the worktree as being removed under the
state lock, then takes its containers, its tree and its namespaces
down without it, so several removals — from the TUI, or from several
terminals — run side by side, and nothing else waits for one. A
worktree being removed refuses a start, a share, a secondrmand a
newof its name; a removal stopped half way, by Ctrl-C or a crash,
is finished by the nextpando rmof it, containers and namespaces
included. rmof a running worktree stops it asstopnow does, waiting for
its processes without the lock: two removals of worktrees whose
servers ignore SIGTERM end in 5.3s together, from 10.3s.unshareno longer holds every other command up while the tunnel and
its proxy get their five seconds to exit; the share stays recorded
until both are down.- The TUI no longer removes one worktree at a time:
don another
while one was being removed saidalready busy removing …. A removal
keeps the branch and writes no ref, so it runs beside another
removal, a new worktree being made, or a git move elsewhere. - A worktree being made in the TUI holds up git moves only while git
checks it out, not through the minutes its install takes, andX
goes ahead while one is being made: a create starts nothing a stop
of everything would stop. A create beside another's install no
longer fails when that install holds the repository's config a
moment (husky installsettingcore.hooksPath):newtries git's
lock errors again for up to three seconds.
Install pando-cli 0.11.2
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/mertkaradayi/pando/releases/download/v0.11.2/pando-cli-installer.sh | shInstall prebuilt binaries via Homebrew
brew install mertkaradayi/tap/pandoDownload pando-cli 0.11.2
| File | Platform | Checksum |
|---|---|---|
| pando-cli-aarch64-apple-darwin.tar.xz | Apple Silicon macOS | checksum |
| pando-cli-x86_64-apple-darwin.tar.xz | Intel macOS | checksum |
| pando-cli-aarch64-unknown-linux-musl.tar.xz | ARM64 MUSL Linux | checksum |
| pando-cli-x86_64-unknown-linux-musl.tar.xz | x64 MUSL Linux | checksum |
Verifying GitHub Artifact Attestations
The artifacts in this release have attestations generated with GitHub Artifact Attestations. These can be verified by using the GitHub CLI:
gh attestation verify <file-path of downloaded artifact> --repo mertkaradayi/pandoYou can also download the attestation from GitHub and verify against that directly:
gh attestation verify <file-path of downloaded artifact> --bundle <file-path of downloaded attestation> --repo mertkaradayi/pando