Files
pkh/README.md
T
vhaudiquet 93943dcce0 deb: do not retain staged trees over the retention size cap
The per-target retention kept the newest five staged trees however
large they are: five carried kernel trees are three-digit GiB, most of
it compiled objects whose resume value does not scale with their size.

A staged tree larger than the retention cap (a user setting, see
below) is now dropped at the end of its own build — the record
degrades to metadata-only and the next build re-stages from scratch —
and the retention sweep also degrades older oversized trees, covering
records kept by older pkh versions and interrupted builds. The cap of
0 disables the size rule; the count cap is unchanged.

The cap is the first setting of a user configuration file,
~/.config/pkh/config.json (the directory already holds contexts.json):
missing file and missing keys mean defaults, a corrupt file is backed
up and the defaults apply, and PKH_STAGED_MAXSIZE overrides the file.
The default is 20G, the scale where a carried kernel tree stops
earning its disk.

Every retained tree's size is measured once at retention (the same
concurrent, mount-crossing-safe walker the prune report uses) and
stored in the record manifest, so sweeps reuse it instead of re-walking
the trees on every build.
2026-10-03 18:28:12 +02:00

278 lines
11 KiB
Markdown

# pkh
`pkh` is a packaging helper for Debian/Ubuntu packages.
![](.github/pkh.gif)
## Installation
From crates.io:
```
cargo install pkh
```
Or build from source (the same system packages are needed either way):
```
sudo apt install pkg-config libssl-dev libgpg-error-dev libgpgme-dev
git clone https://git.vhaudiquet.fr/vhaudiquet/pkh.git
cd pkh
cargo install --path .
```
At runtime pkh shells out to the Debian packaging toolchain (git,
dpkg-dev, quilt, mmdebstrap, lintian, pristine-tar, autopkgtest, ...).
Install the ones your workflows use, or build the classic snap from
`snap/snapcraft.yaml` (`snapcraft pack`), which carries them.
## Usage and features
### Basic concepts
`pkh` wraps the different Debian tools and workflows
(`git`, `git-ubuntu`, `pull-debian-sources`, `pull-lp-sources`, `pull-ppa-sources`,
`dch`, `dpkg-buildpackage`, `sbuild`, `dpkg-source`, `quilt`, ...) into
one tool with the same interface for everything, which also handles each
workflow's specifics.
The same options work across subcommands (with very few command-specific options):
```
Options:
-s, --series <series> Target package distribution series
-d, --dist <dist> Target package distribution (debian, ubuntu)
-v, --version <version> Target package version
-a, --arch <arch> Target architecture (amd64, arm64, riscv64, ...)
-p, --pocket <pocket> Target distribution pocket (updates, security, proposed, ...)
--ppa <ppa> Do the action in/for a specific PPA
```
Commands and workflows include:
```
Commands:
new Scaffold a new Debian source package (buildable right away)
pull Pull a source package from the archive or git
chlog Auto-generate changelog entry, editing it, committing it afterwards
build Build the source package (into a .dsc)
put Upload the built source package to a PPA
deb Build the source package into binary package (.deb)
test Run the package's DEP-8 as-installed tests (autopkgtests)
lint Lint the package (lintian wrapper + pkh-native checks)
prune Prune residual pkh build artifacts and caches
help Print this message or the help of the given subcommand(s)
Options:
-h, --help Print help
-V, --version Print version
```
### Examples
A typical workflow to patch an Ubuntu package `hello` on the development release could be:
```
# Obtain the package source
git ubuntu clone hello
cd hello
# Apply the patch to the package
...
dpkg-source --commit
# Increment version number
dch
# Test that the patch builds
git ubuntu export-orig
dpkg-buildpackage -S -I -i -nc -d
sbuild --dist resolute --arch amd64 ../hello_xxx.dsc
# Upload the package to a ppa
dput ppa:user/hello_xxx ../hello_xxx_source.changes
# Commit the changes to git
git add debian/patches/xxx.patch
git commit -m "Applied patch xxx"
git add debian/changelog
git commit -m "changelog"
git checkout -b xxx
git push xxx user-fork
```
That is a lot of different tools and operations. With pkh, the same workflow:
```
# Obtain the package source (and orig tarball)
pkh pull hello # needs -d ubuntu if you are not running Ubuntu
# Apply the patch to the package
...
git add debian/patches/xxx.patch
git commit -m "Applied patch xxx"
pkh chlog
git add debian/changelog
git commit -m "d/changelog"
# Test that the package builds
pkh build
pkh deb
# Upload the package to a ppa
pkh put --ppa user/hello_xxx
# Push the commits to your fork
git push xxx user-fork
```
`pkh chlog` edits `debian/changelog` by default; point it at another file
with `--file <path>` (`-f`).
`pkh chlog` walks the commit history back to the previous changelog entry.
Without a version tag, that search is capped at the last 500 commits
(`--limit <n>`, `0` = unlimited): beyond the cap, the entry simply starts
at the oldest commit searched, which keeps the run fast on large histories.
### Incremental builds (build records)
Every `pkh deb` build records its metadata under
`/var/tmp/pkh/sessions/<target>/<id>/`: the manifest (identity, fork
lineage, outcome, the sha of the chroot tarball it used) and a full
`log/build.log` tee of the build output. The chroot itself is per-run
scratch, extracted fresh from the sha-verified mmdebstrap tarball cache
on every run and removed at the end, plain and `--incremental` alike,
so records stay small. `--incremental` keeps the record's staged tree
(`tree-upper/`), the overlay upperdir where the build's writes land,
holding the compiled objects.
Fresh chroots share their downloads. Every chroot's
`/var/cache/apt/archives` is a bind mount over a shared apt download
cache at `/var/tmp/pkh/apt-cache/<series>-<arch>/`, which the installs
of one run fill and the next runs reuse (apt is configured to keep the
downloaded packages, and an `apt-get autoclean` at the end of each run
drops the debs the archive no longer serves). `pkh prune` removes the
caches untouched for a week.
Fresh chroots share their compiles too. Every run's chroot
gets a shared ccache (at `/var/tmp/pkh/ccache/<target>/`, mounted at
`/var/cache/pkh-ccache`) wired in through a compiler PATH shim that
resolves `gcc`, `g++`, `cc` and `c++` to ccache, with content-keyed
cache checks (`CCACHE_COMPILERCHECK=content`), so cache hits survive
the fresh toolchain of a new chroot. The cache is capped at 10G per
target (override with `PKH_CCACHE_MAXSIZE`); `pkh deb --no-ccache`
bypasses the shim and cache entirely, an escape hatch for rules that
mishandle compiler wrappers (kbuild-style flavour configs).
`--incremental` forks a copy of a previous record and never modifies
the base, so a failed run cannot destroy the last-good state it was
built from. The staged tree is carried only when the base's recorded
tarball sha still matches the cache (the toolchain did not move) and the
changelog version is unchanged; otherwise the fork builds from a fresh
tree:
```
pkh deb # plain run: fresh chroot, fresh tree; the
# record keeps only metadata and the log
# (nothing is kept even on failure or
# interrupt)
pkh deb --incremental # fork the newest successful record of the
# target (among the last five carrying a staged
# tree) and keep this run's staged tree, for
# the next fork to carry
pkh deb --incremental <id>
# fork that exact record (see `pkh deb list`);
# a failed record is fine if its staged tree is
pkh deb list # the records of this tree: id, outcome,
# version, fork lineage, staged-tree state,
# dep-install and build times, age
pkh deb log [-f] [<id>] # print a record's build log; -f follows a
# live build
pkh deb profile [<id>] # a record's per-phase timing profile: how
# long every step took (chroot bootstrap,
# apt phases, patch application, build-user
# setup, build and binary steps, artifact
# retrieval), with their shares of the total
pkh deb --no-ccache # build without the shared compile cache
```
Each target keeps at most five staged trees: when a sixth is retained,
the oldest degrades to a metadata-only record (manifest and log stay).
A staged tree larger than 20 GiB is never kept at all — a tree that
size costs more disk than the resume it buys, so its record degrades
too and the next build re-stages from scratch. The cap is a setting:
```json
// ~/.config/pkh/config.json (missing = defaults; keys may be omitted)
{ "staged_maxsize": "20G" }
```
`"0"` disables the size rule entirely, and `PKH_STAGED_MAXSIZE`
overrides the file when set. `pkh prune` removes records older than a
week. `pkh prune --dry-run` previews everything it would remove,
grouped by kind with sizes; `--all` ignores the age and removes
everything a live build does not hold, recent records and staged trees
included; `-f/--force` skips the size pass and the confirmation
prompt, removing immediately. `pkh deb --incremental` never forks a
record built for a different series/architecture.
Packages whose `debian/rules` keeps an input-untracked stamp cache (the
kernels' `debian/stamps`) declare it in the quirks data
(`resume_clear`). Forked builds drop the cache, so the wrapper steps,
including the kernel's flavour config export, re-run over the
incremental inner builds and a config edit is never silently ignored.
A build killed hard (OOM) can leave its chroot behind; the next run of
that target sweeps it (unmounting and escalating through sudo when
needed), and `pkh prune` collects the rest.
On a pinned runner, a machine that must resolve every package from one
mirror (typically a caching proxy), `pkh deb --mirror <url>` builds the
chroot and resolves every package from that base URL. The mirror is
passed to mmdebstrap at bootstrap, so the chroot tarball's apt sources
carry it and every apt inside the build follows. The mirror must serve
every suite and architecture the build resolves (the security pocket
and, for `--cross` builds, the target architecture's archive included);
PPAs and other third-party sources are left alone. The tarball cache is
keyed per mirror, so pinned and unpinned builds keep separate tarballs
(and `--incremental` records never fork across a mirror change).
### Running the DEP-8 tests
`pkh test` runs the package's as-installed tests (`debian/tests/control`,
declared by `Testsuite: autopkgtest` in `debian/control`) through a
native runner on a pkh testbed, and reports one verdict per test:
```
pkh test # test this tree: builds it first if needed
pkh test --mode local # on the current context, no isolation
pkh test --mode vm --image img.qcow2 # in a VM, via autopkgtest
pkh test --json --fail-on skip # CI-shaped
pkh test --list-tests # what does this package test?
```
By default the tests run in an ephemeral unshare chroot bootstrapped
from the same cached tarballs as `pkh deb`, against the binaries pkh deb
just built next to the tree (stale or missing output is rebuilt first;
`--no-build` refuses, `--debs` overrides). Pockets and PPAs resolve test
dependencies like in `pkh deb` (`-p`, `--ppa`), and `--test-name`,
`--skip-test`, `--shell-fail` and `--setup-commands` cover the day-to-day
debugging loop. Tests needing a real machine (`isolation-machine`,
`needs-reboot`) run only in `--mode vm`, which execs the installed
`autopkgtest` with its qemu runner.
Exit codes: 0 all tests passed (skips and flaky allowed), 1 at least one
test failed (or a `--fail-on` trigger), 2 runtime errors (no declared
tests, no usable testbed, ...).
## End-to-end package harness
`tests/e2e/` exercises pkh the way it is used in anger: a catalog of ~60
real source packages (from `hello` and `micro` to `dracut`, `grub2` and
the kernel) run through the full `pull` → `chlog --rebuild` → `build` →
`deb` workflow, each stage timed, logged and checked for artifacts:
```
tests/e2e/run.py --list # what would run
tests/e2e/run.py # smoke + standard tiers
tests/e2e/run.py --tier long # the multi-hour builds
```
See [tests/e2e/README.md](tests/e2e/README.md) for prerequisites,
filters (`--only`, `--category`, `--tag`), resume support and the run
output layout.
## Future improvement ideas
- pull: try to fetch the correct git branch for series on Debian
- deb: asynchronous build, detachable and monitorable
- put: allow uploads to Debian or Ubuntu archives
- pull: cache Sources.gz files to improve speed
- pull: 'pkh pull' in a package tree should git pull and re-fetch orig tgz