Document security verdict versus Docker

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Charts and reproducible benchmarks, Makefile, expanded .gitignore
2026-10-23 23:26:57 +02:00 · 2026-10-23 22:24:56 +02:00 · 2026-10-23 22:18:42 +02:00
19 changed files with 1087 additions and 16 deletions
@@ -1 +1,33 @@
+# build output
 build/
+cmake-build-*/
+CMakeCache.txt
+CMakeFiles/
+cmake_install.cmake
+compile_commands.json
+*.o
+*.a
+*.so
+*.d
+
+# state and scratch created by running or testing doos
+.doos/
+*.tmp
+*.log
+*.pid
+*.core
+core
+
+# editors and OS
+.vscode/
+.idea/
+*.swp
+*~
+.DS_Store
+
+# local AI tooling state
+.claude/
+
+# python bytecode from docs/charts/make_charts.py
+__pycache__/
+*.pyc
@@ -1,5 +1,7 @@
 # AGENTS.md

+**Docs:** [README](README.md) · [SPEC](SPEC.md) · [ARCHITECTURE](docs/ARCHITECTURE.md) · [RESEARCH](docs/RESEARCH.md) · [COMPARISON](COMPARISON.md) · [AGENTS](AGENTS.md) · [CHANGELOG](CHANGELOG.md)
+
 Rules for any AI agent (or human) changing this repository. The user's global rules
 also apply; this file adds the project-specific ones. Keep this file current: it is
 part of every release.
@@ -11,8 +13,9 @@ part of every release.
 extracted image (image mode) with a per-container upper dir, pivots into it and runs a command.
 Networking is slirp4netns or pasta, ids are mapped through newuidmap when possible, cgroup v2 limits
 and a watchdog enforce `--memory --pids --cpus --size`, capabilities and seccomp confine the command.
-No root at runtime. Authoritative behavior: `SPEC.md`. Design notes: `docs/ARCHITECTURE.md`.
-Measurements and gaps: `docs/RESEARCH.md`.
+No root at runtime. Authoritative behavior: [SPEC](SPEC.md). Design notes: [ARCHITECTURE](docs/ARCHITECTURE.md).
+Measurements and gaps: [RESEARCH](docs/RESEARCH.md). Positioning against other tools: [COMPARISON](COMPARISON.md).
+Usage: [README](README.md). History: [CHANGELOG](CHANGELOG.md).

 ## Mandatory release workflow (every change, same session)

@@ -22,11 +25,12 @@ Measurements and gaps: `docs/RESEARCH.md`.
   - `patch`: bug fix, docs, tests, tooling, no behavior change for users.
   - `minor`: new or changed user-visible behavior (flag, command, mount, limit).
   - `major`: breaking change to CLI, state layout, or container compatibility.
-4. Run `scripts/release.sh <patch|minor|major> "imperative commit subject"`.
+4. Run `make docs` (`charts` plus `check-docs`) (the release script does too: every document must link to every other, every relative link must resolve).
+5. Run `scripts/release.sh <patch|minor|major> "imperative commit subject"`.
   It bumps `VERSION`, builds, runs `tests/run.sh`, commits all changes, and creates
   the annotated tag `vX.Y.Z`. It aborts on any build or test failure.
-5. Verify live: run `build/doos` once against the real thing (the e2e suite does this).
-6. Report honestly what passed and what did not. Do not push unless the user asks.
+6. Verify live: run `build/doos` once against the real thing (the e2e suite does this).
+7. Report honestly what passed and what did not. Do not push unless the user asks.

 Never hand-edit `VERSION` or create tags manually. Never skip the tests to release.

@@ -42,6 +46,9 @@ Never hand-edit `VERSION` or create tags manually. Never skip the tests to relea
 | A new measurement | `docs/RESEARCH.md` with machine, kernel, and what was run |
 | A new known gap or a closed gap | `docs/RESEARCH.md` gaps list and `README.md` |
 | These workflow rules | this file |
+| A measurement, or anything the charts show | `make bench` then `make charts` (rewrites `docs/charts/*.svg` and every `<!--bench:...-->` number and table in `COMPARISON.md` and `docs/RESEARCH.md`); never type those numbers by hand |
+| A feature or limit that appears in the comparison tables | `COMPARISON.md` (tables, the gaps list, the conclusion) |
+| A new or renamed markdown file | `scripts/check-docs.sh` DOCS list, the nav line (`**Docs:**`) of every document, `README.md` Project docs |
 | Image catalog, sources, verification, pinned keys | `SPEC.md` Images, `README.md`, `docs/RESEARCH.md` |
 | Seccomp list, capability set, `/etc` allowlist | `SPEC.md` Hardening and filesystem, `README.md`, `docs/RESEARCH.md` |
 | Limits, cgroup layout | `SPEC.md` Resource limits, `docs/ARCHITECTURE.md` |
@@ -75,6 +82,12 @@ freeze (about 4 minutes in total, needs network, `pasta`, `gpg` and cgroup v2 de
 before release. On Ubuntu 24.04+ the AppArmor profile in `apparmor/doos` must be
 installed once for the binary being tested (`doos apparmor`; needs sudo, ask the user, do not do it yourself).

+## Make targets
+
+`make help` lists them: `build`, `test`, `check-docs`, `bench` (re-measure into `docs/charts/bench.json`),
+`charts` (SVGs and generated numbers from that file), `docs`, `release KIND=patch MSG="..."`, `clean`.
+The Makefile is a thin wrapper; the logic lives in `CMakeLists.txt`, `scripts/` and `docs/charts/`.
+
 ## Environment notes

 - State dir: `$DOOS_HOME`, else `$XDG_DATA_HOME/doos`, else `~/.local/share/doos`.
@@ -1,5 +1,19 @@
 # Changelog

+**Docs:** [README](README.md) · [SPEC](SPEC.md) · [ARCHITECTURE](docs/ARCHITECTURE.md) · [RESEARCH](docs/RESEARCH.md) · [COMPARISON](COMPARISON.md) · [AGENTS](AGENTS.md) · [CHANGELOG](CHANGELOG.md)
+
+## 1.1.4 - 2026-10-23
+- Docs: [COMPARISON](COMPARISON.md) section 6.1 answers whether doos is a security-wise Docker replacement (same class as rootless Podman, not for untrusted code), summarized in the README safety summary.
+
+## 1.1.3 - 2026-10-23
+- Charts: startup time, executable size and CPU limit as SVG (light and dark palettes, doos highlighted), embedded in [COMPARISON](COMPARISON.md) and [RESEARCH](docs/RESEARCH.md).
+- Reproducible measurements: `scripts/bench.sh` writes `docs/charts/bench.json`; `docs/charts/make_charts.py` builds the charts and rewrites every generated number and table in the documents from it, so text, tables and charts cannot drift apart.
+- `Makefile` wrapper (`make help`), expanded `.gitignore`.
+
+## 1.1.2 - 2026-10-23
+- Docs: new [COMPARISON](COMPARISON.md) (doos against Docker, rootless Docker, Podman, bubblewrap, systemd-nspawn, LXC/Incus, Apptainer, Distrobox/Toolbx, gVisor, Kata/Firecracker) with sourced pros and cons, startup and disk measurements against Docker, Podman and bubblewrap on the same machine, a gaps list tied to code, and a conclusion.
+- Docs: every document now has a navigation line and links to every other document. `scripts/check-docs.sh` enforces it (and that relative links resolve) and runs inside `scripts/release.sh`.
+
 ## 1.1.1 - 2026-10-23
 - `doos apparmor` prints a profile for the exact path of the running binary (own profile name per path, so several installs coexist). A start that fails because of AppArmor now prints the install command instead of only `mount bind /: Permission denied`. Verified with a copy in `~/.local/bin`.
 - `apparmor/doos` (packaging) covers more standard install paths.
@@ -0,0 +1,206 @@
+# doos compared with Docker, Podman and others
+
+**Docs:** [README](README.md) · [SPEC](SPEC.md) · [ARCHITECTURE](docs/ARCHITECTURE.md) · [RESEARCH](docs/RESEARCH.md) · [COMPARISON](COMPARISON.md) · [AGENTS](AGENTS.md) · [CHANGELOG](CHANGELOG.md)
+
+This page puts doos next to the tools people usually reach for instead. It separates three kinds of claims, and every row says which one it is:
+
+- **[M] measured** on the development machine, described in "Measurements" below.
+- **[S] sourced** from the linked documentation or articles. Vendor blogs are marked as such and their numbers are not used.
+- **[D] design**: what the doos source and [SPEC](SPEC.md) do today.
+- **[K] general knowledge** that this research did not re-verify.
+
+Nothing here is a roadmap. Where doos lacks something, the gap is listed with the code that would have to change.
+
+Versions compared: doos 1.1.1, Docker 29.1.3 (rootful daemon on this machine), Podman 5.7.0 (netavark, crun, overlay), bubblewrap 0.11.1, on Ubuntu 26.04, kernel 7.0.0-38, 12 cores, one machine.
+
+## 1. At a glance
+
+| | doos | Docker (rootful) | Docker rootless | Podman rootless |
+|---|---|---|---|---|
+| Needs a daemon | no [D] | yes, root daemon [S: [Docker security](https://docs.docker.com/engine/security/)] | yes, per-user daemon [K] | no [S: [rootlesscontaine.rs](https://rootlesscontaine.rs/getting-started/podman/)] |
+| Runs without root | yes [D] | no [S] | yes [S] | yes [S] |
+| Who can use it safely | any user | only trusted users: access to the daemon is root-equivalent [S: [Docker security](https://docs.docker.com/engine/security/), [iximiuz](https://labs.iximiuz.com/tutorials/docker-security-introduction-a859718d)] | any user | any user |
+| Image source | tarball, https URL with sha256, small catalog (Alpine, Ubuntu, Debian) [D] | OCI registries, Dockerfile builds [K] | same as Docker | OCI registries, Containerfile builds [K] |
+| Build images | no [D] | yes (BuildKit) [K] | yes | yes (Buildah) [K] |
+| Host dirs into the container (`-v`) | **no** [D] | yes [K] | yes | yes |
+| Published ports | none for `slirp`; automatic for `pasta` [D] | `-p` [K] | `-p` | `-p` [S: [podman-rootless(7)](https://man.archlinux.org/man/podman-rootless.7.en)] |
+| Rootless network backend | slirp4netns or pasta [D] | n/a | RootlessKit based [S: [Podman blog](https://blog.podman.io/2023/07/podman-performance-root-and-rootless/)] | pasta default since 5.0 [S: [podman-rootless(7)](https://man.archlinux.org/man/podman-rootless.7.en)] |
+| Resource limits | memory, pids, cpus, written data [D][M] | memory, cpu, pids, more [K] | needs cgroup v2 + systemd delegation [S: [Docker rootless tips](https://docs.docker.com/engine/security/rootless/tips/)] | needs cgroup v2 [S: [podman-rootless(7)](https://man.archlinux.org/man/podman-rootless.7.en)] |
+| Default syscall filter | denylist: 48 listed calls plus rules for `clone`, `unshare`, `clone3` [D] | allowlist profile [S: [Docker seccomp](https://docs.docker.com/engine/security/seccomp/)] | same | built-in profile [K] |
+| Default capabilities | Docker's default set [D][M] | that set [S] | that set | a smaller set [S: [dev.to comparison](https://dev.to/mechcloud_academy/docker-vs-podman-an-in-depth-comparison-2025-2eia)] |
+| Orchestration (Compose, Kubernetes) | none [D] | Compose, Swarm, Kubernetes via runtimes [K] | same | Compose, pods, `kube play`, systemd units [K] |
+| Background containers, logs, `stop` | none: a container lives while its first `attach` command lives [D] | yes [K] | yes | yes |
+| Binary size | <!--bench:doos_size-->552 KiB<!--/bench-->, <!--bench:shared_libs-->6<!--/bench--> shared libraries, <!--bench:src_lines-->3,445<!--/bench--> lines of C++ [M] | CLI <!--bench:docker_size-->29.9 MiB<!--/bench-->, plus `dockerd` and `containerd` [M][K] | same | <!--bench:podman_size-->41.2 MiB<!--/bench--> [M] |
+| Platforms | Linux only, tested on x86_64 only [D] | Linux, macOS, Windows (via a VM) [K] | Linux | Linux; macOS and Windows via a VM [K] |
+| Maturity | version 1.1, one test machine | very mature, huge ecosystem | mature | mature |
+
+## 2. Other tools, one line each
+
+| Tool | What it is | Versus doos |
+|---|---|---|
+| bubblewrap (`bwrap`) | Low-level unprivileged sandbox builder, the engine behind Flatpak [S: [bubblewrap README](https://cdn.jsdelivr.net/gh/containers/bubblewrap/README.md), [Julia Evans](https://jvns.ca/blog/2022/06/28/some-notes-on-bubblewrap/)] | Smaller and faster to start [M], but you assemble every mount by hand and it has no image handling, no persistent state and no cgroup limits [K]; it maps a single id unless you add a helper [K]. doos is the "batteries included" layer on the same kernel features. |
+| systemd-nspawn | Boots a whole OS tree as a container [S: [forum overview](https://forums.puri.sm/t/linux-containers-comparison-quest-for-the-best/13344)] | Rootless use is limited to a single UID/GID inside [S]. Not installed here, so not measured. |
+| LXC / Incus (LXD fork) | System containers: a full distro with init and several services, close to a light VM [S: [OneUptime](https://oneuptime.com/blog/post/2026-03-18-choose-between-podman-and-lxc-lxd/view)] | Better for long-lived pet systems and multi-service setups, with API, storage pools and profiles. Heavier to set up. `lxc` binaries exist here, Incus does not, neither was run. |
+| Apptainer (Singularity) | Rootless, daemonless, single-file SIF images, built for HPC (MPI, InfiniBand, GPUs) [S: [Alliance Canada](https://docs.alliancecan.ca/wiki/Apptainer), [NYU HPC](https://services.rt.nyu.edu/docs/hpc/containers/intro/)] | Different audience. It also has `--fakeroot` and signed immutable images [S], which doos lacks. Not installed here. |
+| Distrobox / Toolbx | Podman- or Docker-backed "your distro inside your terminal", home directory shared on purpose, not a sandbox [S: [Arch wiki](https://wiki.archlinux.org/title/Distrobox), [comparison](https://computingforgeeks.com/distrobox-vs-toolbox/)] | Opposite goal: doos hides the host's `/home` by default. |
+| gVisor | A user-space kernel that answers most syscalls itself, cutting host-kernel exposure at an I/O cost [S: [Edera](https://edera.dev/stories/kata-vs-firecracker-vs-gvisor-isolation-compared), [Northflank](https://northflank.com/blog/kata-containers-vs-firecracker-vs-gvisor)] | Stronger isolation than doos, needs its own runtime. Vendor blogs; their numbers are not used here. |
+| Kata Containers / Firecracker | Each workload in its own KVM micro-VM with its own kernel [S: same sources, [MASCOTS 2020 paper](https://cristian.kleinlabs.eu/publications/mascots2020_container_runtime_security.pdf)] | The right tool for hostile or multi-tenant code. doos shares the host kernel, so a kernel bug is a doos escape. |
+
+## 3. Measurements
+
+Method: a trivial command (`true`) started <!--bench:runs-->10<!--/bench--> times in a row after one warm-up run, average wall time per run, the Alpine image already present for every tool, one machine, one run set on <!--bench:bench_date-->2026-10-23<!--/bench-->. Versions: <!--bench:versions-->doos 1.1.2, Docker 29.1.3, Podman 5.7.0, bubblewrap 0.11.1<!--/bench-->. The numbers come from `scripts/bench.sh` (`make bench`); the charts and the tables below are generated from its output (`make charts`), so they cannot drift from each other.
+
+![Horizontal bar chart of the time to start a trivial command for doos, bubblewrap, Podman and Docker, with and without network. doos is highlighted. Exact values are in the table below.](docs/charts/startup.svg)
+
+<!-- bench:startup:begin -->
+| Setup | Network | Time per run |
+|---|---|---|
+| bubblewrap (host /usr) | none | 8.4 ms |
+| doos Alpine image | none | 13.8 ms |
+| doos host mode | none | 18.7 ms |
+| Podman Alpine | none | 208 ms |
+| Docker Alpine (rootful) | none | 344 ms |
+| bubblewrap (host /usr) | default | 7.4 ms |
+| doos Alpine image | default | 52.1 ms |
+| doos host mode | default | 56.9 ms |
+| doos Alpine, pasta | default | 64.6 ms |
+| Podman Alpine | default | 241 ms |
+| Docker Alpine (rootful) | default | 436 ms |
+<!-- bench:startup:end -->
+
+How to read this honestly:
+
+- doos is fast partly because it does less: no image store database, no registry code, no layer unpacking per run, no container metadata service. The extra time over bubblewrap is the id map helpers (`newuidmap`, `newgidmap`), the overlay mounts, a hardening filter and the network helper process.
+- Docker here is a rootful daemon, so its numbers include a client-to-daemon round trip. Rootless Docker was not measured.
+- Run-to-run spread is real. An earlier ad-hoc run of the same loop measured rootful Docker at 590 ms (no network) and 456 ms (default network); the run charted above measured 344 ms and 436 ms. Read the order of magnitude, not a ranking between close values.
+- Startup is one number among many. It matters for scripts that start hundreds of short-lived sandboxes. It does not matter for a service that starts once.
+
+![Horizontal bar chart of executable sizes on a logarithmic axis for bubblewrap, crun, doos, the docker CLI and podman. doos is highlighted. Exact values are in the table below.](docs/charts/size.svg)
+
+<!-- bench:sizes:begin -->
+| Executable | Size |
+|---|---|
+| bubblewrap | 79 KiB (80,424 bytes) |
+| crun | 472 KiB (483,392 bytes) |
+| doos | 552 KiB (565,680 bytes) |
+| docker CLI | 29.9 MiB (31,374,336 bytes) |
+| podman | 41.2 MiB (43,180,896 bytes) |
+<!-- bench:sizes:end -->
+
+Disk, measured with `du` (which cannot read overlayfs' mode-000 work directories, so true usage is slightly higher):
+
+| Item | Size |
+|---|---|
+| doos Alpine image, extracted once and shared | 8.3 MiB |
+| doos Ubuntu base image, extracted once and shared | 98 MiB |
+| one doos container on top of an image | a few KiB of metadata plus its own writes |
+| four containers (one Alpine, three Ubuntu) on shared images | 20 KiB together |
+| doos host-mode container after its first create (skeleton, `/etc` allowlist, dpkg state) | 4 MiB, created in 0.5 s |
+| Podman storage on this machine (many unrelated images) | 26.3 GB images, 1.8 GB containers |
+
+The Podman figure is this machine's accumulated state, not a like-for-like comparison. What the comparison does show is the model: doos extracts an image once and every container adds only its own writes.
+
+## 4. doos versus Docker
+
+**Where doos is better**
+
+- No daemon and no root. With Docker's default install, anyone who can talk to the daemon controls the machine [S]. doos has no such trust boundary: it can only do what the invoking user can do, plus what the `newuidmap` helper allows.
+- Tiny and auditable: <!--bench:doos_size-->552 KiB<!--/bench-->, <!--bench:src_lines-->3,445<!--/bench--> lines of C++ [M], one process tree, no background service, nothing listening.
+- Instant: <!--bench:doos_range-->14 to 65 ms<!--/bench--> against <!--bench:docker_range-->344 to 436 ms<!--/bench--> for a rootful Docker run and <!--bench:podman_range-->208 to 241 ms<!--/bench--> for Podman [M].
+- Safe defaults a user does not have to remember: Docker's default capability set, a seccomp filter, an allowlist of host `/etc`, a memory and pid cap, a written-data cap [D][M].
+- A "host mode" that reuses the host's installed tools (compilers, `apt` state) for throwaway shells with no image download [D].
+- A size cap that freezes instead of filling the disk [D][M]. Docker's writable-layer size limit needs the overlay2 driver on XFS with project quotas [S: [Fabian Lee](https://fabianlee.org/2020/01/15/docker-use-overlay2-with-an-xfs-backing-filesystem-to-limit-rootfs-size/)].
+
+**Where Docker is better**
+
+- Everything around images: registries, `docker build`, layer caching, multi-arch, signing and scanning ecosystems [K]. doos cannot pull an arbitrary `nginx:1.27` from a registry.
+- Host directories and devices into the container (`-v`, `--device`, GPUs). doos has no bind-mount option [D].
+- Long-running services: background containers, restart policies, healthchecks, logs, `exec`, `stop` [K]. In doos a container exists only while its first `attach` command runs [D].
+- Networking: user-defined networks, DNS between containers, published ports on any interface [K].
+- Compose, Swarm and the Kubernetes ecosystem [K].
+- macOS and Windows support through a VM [K].
+- Strongest default syscall posture: an allowlist, which denies unknown new syscalls. The doos filter is a denylist, so a new kernel syscall is allowed until it is added [D][S].
+
+## 5. doos versus Podman
+
+Podman is the closest comparison, since both are daemonless and rootless.
+
+**Where doos is better**
+
+- Smaller, faster to start [M] and one binary with no storage database. Podman's rootless setup brings a store under `~/.local/share/containers` with its own metadata [K].
+- Opinionated limits and hardening out of the box, with a stated, measured effect (OOM kill, pids, cpus, freeze at a size cap) [M].
+- Two network backends with different, documented trade-offs, including a host-like mode that keeps host abstract unix sockets out of the container [M].
+- Image verification is explicit: sha256 plus signatures checked against fingerprints pinned in the source [D][M].
+- Frozen-at-the-cap behaviour with a never-frozen rescue shell [D][M].
+
+**Where Podman is better**
+
+- OCI compatibility end to end: pull, push, build with Buildah, pods, `kube play`, Quadlet and systemd integration [K]. It can also run images made for Docker unchanged.
+- Docker-compatible CLI and socket, so existing tooling mostly works [K].
+- Bind mounts, volumes, secrets, `exec`, `logs`, `stop`, restart policies [K].
+- Years of hardening, a security team, and a large user base finding the sharp edges. doos has one test machine.
+- Podman documents rootless caveats: ports below 1024, `/etc/subuid` setup, cgroup v2 only, and a rootless port forwarder that does not preserve client source IPs [S: [podman-rootless(7)](https://man.archlinux.org/man/podman-rootless.7.en)]. doos shares the subuid and cgroup v2 requirements [D]. Binding ports below 1024 in doos was not tested.
+
+**What both share:** the same kernel features (user, mount, pid, network namespaces, overlayfs, cgroup v2, seccomp), the same pasta/slirp4netns options, and the same weakness: they share the host kernel [S].
+
+## 6. Security, side by side
+
+- **Shared kernel.** doos, Docker, Podman, bubblewrap, nspawn and LXC all run container processes on the host kernel. A kernel vulnerability reachable through allowed syscalls is a host vulnerability. gVisor reduces that exposure at a performance cost; Kata and Firecracker remove it by giving each workload its own kernel [S].
+- **Rootless is the real difference from rootful Docker.** A bug that gives root inside a rootless doos or Podman container yields the invoking user's rights on the host, not root [D][S].
+- **doos filter style.** A denylist built at runtime with an architecture check, x32 guard and `clone3` handling [D]. Docker and Podman ship allowlist profiles [S]. An allowlist is the more conservative design.
+- **Missing layers.** Docker applies an AppArmor profile (`docker-default`) on this machine [M: `docker info` lists apparmor]. doos installs no AppArmor or SELinux profile of its own; its only profile exists to *permit* user namespaces on Ubuntu 24.04+ [D].
+- **Supply chain.** doos pins publisher signing keys for Alpine and Ubuntu and checks digests for Debian [D][M]. Registry-based tools rely on registry trust, content trust or external signing tools [K].
+
+### 6.1 Is doos a security-wise Docker replacement?
+
+Short answer: it is in the same class as rootless Podman and rootless Docker by design, but it is not a replacement for running untrusted code.
+
+| Question | Answer |
+|---|---|
+| Better than rootful Docker on privilege? | Yes. No root daemon, nothing runs as root on the host, so an escape yields the invoking user's rights, not root [D][S]. |
+| Equal to rootless Podman or Docker? | Not proven. The mechanism is the same (user namespaces, seccomp, dropped capabilities, cgroup v2), the maturity is not [D][K]. |
+| What the default confines | 14 capabilities (Docker's default set, `CapEff` `a80425fb`), `no_new_privs`, seccomp denylist of 48 syscalls, cgroup memory/pids/cpu limits, a written-data cap, no host `/home`, host loopback blocked in `slirp` [M]. |
+| Weaker than Docker or Podman | Denylist instead of an allowlist (a new dangerous syscall is allowed by default); no per-container AppArmor or SELinux profile; host mode uses the host's own `/usr` and an `/etc` allowlist; `--net host` shares abstract unix sockets [D]. |
+| Kernel exposure | Shared kernel, and user namespaces enlarge the reachable kernel surface, which is why Ubuntu 24.04 restricts them by default [S]. |
+| Assurance | One developer, one machine (Linux x86_64), no external audit, no CVE history, Debian images verified by digest only [M][D]. |
+
+Use doos for your own tools, builds and experiments, where you accept the risk. For code you do not trust, or for several tenants, use rootless Podman with its default seccomp and SELinux or AppArmor, and gVisor or Kata when the kernel must not be shared (see the conclusion table in section 8). Functionally it is also not a drop-in replacement: see section 7.
+
+Full list of what doos does not cover is at the end of [RESEARCH](docs/RESEARCH.md); the exact hardening is in [SPEC](SPEC.md) under Hardening.
+
+## 7. What doos lacks, and which code would change
+
+These are the gaps that matter when comparing, each next to where it lives. They are descriptions, not promises.
+
+| Gap | Where |
+|---|---|
+| No OCI registry pull, no layers, no build | `src/image.cpp` (`Catalog`, `ImageStore`) |
+| No host directory bind mounts, no devices, no GPU | `setupRoot` in `src/sandbox.cpp`, option parsing in `src/cli.cpp` |
+| No background containers, `stop`, `exec` by name, logs | lifetime model in `Sandbox::start` / `Sandbox::watch` (container lives with its first command) |
+| No user-defined networks, no `-p` | `src/network.cpp` |
+| No Compose or Kubernetes integration | not present |
+| Environment, user, working-directory flags missing | `applyLimitsAndEnv` in `src/sandbox.cpp` |
+| Linux x86_64 tested only | `src/hardening.cpp` guards aarch64, never run there |
+| seccomp is a denylist | `deniedSyscalls` in `src/hardening.cpp` |
+| Test coverage: one machine | `tests/run.sh` |
+
+## 8. Conclusion
+
+doos is not a Docker replacement and does not try to be. It is a small, rootless, daemonless tool for **disposable or per-user Linux shells and sandboxes that start in tens of milliseconds, with safe defaults and hard caps already set**. Its strengths are the ones the measurements show: it is a <!--bench:doos_size-->552 KiB<!--/bench--> single binary, it starts in <!--bench:doos_range-->14 to 65 ms<!--/bench--> [M], it runs with no daemon and no root, it caps memory, processes, CPU and written data by default, and it verifies what it downloads.
+
+Choose by the job:
+
+| If you need | Use |
+|---|---|
+| A quick, isolated shell or script runner for one user, host tools or a distro image, with limits, no daemon | **doos** |
+| OCI images, Dockerfiles or Containerfiles, registries, CI that matches production | **Podman** (rootless, daemonless) or **Docker** |
+| Compose, Kubernetes parity, Docker ecosystem tooling | **Docker** or **Podman** with its Docker compatibility |
+| Sandbox one application with exact control over mounts | **bubblewrap** |
+| A full OS with init and several services, long lived | **Incus / LXC** or **systemd-nspawn** |
+| Shared HPC cluster, MPI, GPUs, signed single-file images | **Apptainer** |
+| A desktop environment tied to your home directory | **Distrobox / Toolbx** |
+| Running code you do not trust at all, or multi-tenant workloads | **Kata / Firecracker** (or gVisor as a middle step), not any shared-kernel container |
+
+Honest limits of this comparison: one machine, 10-run averages, no rootless Docker, no systemd-nspawn, Incus or Apptainer measurements, and several Docker and Podman capabilities listed as general knowledge rather than re-verified. The conclusion about when to use doos rests on what the doos tests and measurements show ([RESEARCH](docs/RESEARCH.md)), not on a claim that it beats mature tools on features. On features it does not.
+
+Back to [README](README.md) for usage, [SPEC](SPEC.md) for the exact behavior, [ARCHITECTURE](docs/ARCHITECTURE.md) for how it works, [AGENTS](AGENTS.md) for the rules, and [CHANGELOG](CHANGELOG.md) for history.
@@ -0,0 +1,35 @@
+# Thin wrapper over the commands used to build, test and update the project. CMake does the building,
+# scripts/ and docs/charts/ do the rest. `make help` lists the targets.
+.DEFAULT_GOAL := help
+BUILD ?= build
+DOOS  := $(BUILD)/doos
+
+.PHONY: help build test check-docs bench charts docs release clean
+
+help: ## list the targets
+	@grep -hE '^[a-z-]+:.*##' $(MAKEFILE_LIST) | sed -E 's/:.*## /\t/' | column -t -s "$$(printf '\t')"
+
+build: ## configure and compile into build/
+	cmake -S . -B $(BUILD) >/dev/null
+	cmake --build $(BUILD)
+
+test: build ## end-to-end suite (real downloads, needs network, pasta, gpg, cgroup v2)
+	tests/run.sh $(DOOS)
+
+check-docs: ## every document links to every other, every relative link resolves
+	scripts/check-docs.sh
+
+bench: build ## re-measure startup, binary sizes and cpu limit into docs/charts/bench.json
+	scripts/bench.sh
+
+charts: ## rebuild docs/charts/*.svg and the generated numbers in COMPARISON.md from bench.json
+	python3 docs/charts/make_charts.py
+
+docs: charts check-docs ## charts plus the docs check
+
+release: ## KIND=patch|minor|major MSG="subject": bump, build, test, commit, tag
+	@test -n "$(KIND)" -a -n "$(MSG)" || { echo 'usage: make release KIND=patch MSG="subject"'; exit 2; }
+	scripts/release.sh $(KIND) "$(MSG)"
+
+clean: ## remove build/
+	rm -rf $(BUILD)
@@ -1,5 +1,7 @@
 # doos

+**Docs:** [README](README.md) · [SPEC](SPEC.md) · [ARCHITECTURE](docs/ARCHITECTURE.md) · [RESEARCH](docs/RESEARCH.md) · [COMPARISON](COMPARISON.md) · [AGENTS](AGENTS.md) · [CHANGELOG](CHANGELOG.md)
+
 User-level containers in C++20: overlayfs, user/mount/pid namespaces, slirp4netns or pasta networking, optional images (Alpine, Ubuntu, Debian, or your own rootfs), memory/cpu/pid/size limits, default hardening (capabilities, seccomp). No root at runtime.

 ```
@@ -11,7 +13,7 @@ doos image list | add [--sha256 HEX] SOURCE [NAME] | delete [--yes] NAME
 doos apparmor        # profile text for this binary (Ubuntu 24.04+)
 ```

-Version: see `VERSION` (currently 1.1.1). Release history: `CHANGELOG.md`.
+Version: see `VERSION` (currently 1.1.3). Release history: `CHANGELOG.md`.

 User-level containers in C++20: overlayfs over the host's `/usr`, `/etc`, `/opt`, user/mount/pid namespaces, networking through slirp4netns. No root at runtime.

@@ -166,12 +168,24 @@ run with Docker's default capability set and a seccomp filter. Memory, pids, cpu
 `--net host` removes network isolation and is opt-in (`--net pasta` is the safer host-like choice).
 What is still not covered is listed at the end of `docs/RESEARCH.md`.

+Is it a security-wise Docker replacement? Same class as rootless Podman by design (no root daemon, user namespaces,
+seccomp, dropped capabilities, cgroup limits), but unaudited, tested on one machine, with a seccomp denylist and no
+per-container AppArmor/SELinux profile. Fine for your own tools; for untrusted code use rootless Podman, gVisor or Kata.
+Details: [COMPARISON](COMPARISON.md) section 6.1.
+
+## Development
+
+`make help` lists the targets: `make build`, `make test` (end-to-end, needs network), `make bench` (re-measure startup, sizes and the cpu limit), `make charts` (rebuild the SVG charts and the generated numbers in the comparison), `make docs`, `make release KIND=patch MSG="..."`. Build and test details are in [AGENTS](AGENTS.md); the charts and how they are generated are in [COMPARISON](COMPARISON.md) section 3 and `docs/charts/`.
+
 ## Project docs

 | File | Content |
 |---|---|
-| `SPEC.md` | Behavior contract and limits |
-| `docs/ARCHITECTURE.md` | Code map and flows |
-| `docs/RESEARCH.md` | What works unprivileged, measurements, gaps |
-| `AGENTS.md` | Rules for AI agents, release workflow |
-| `CHANGELOG.md` | Release history |
+| [`SPEC.md`](SPEC.md) | Behavior contract and limits |
+| [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) | Code map and flows |
+| [`docs/RESEARCH.md`](docs/RESEARCH.md) | What works unprivileged, measurements, remaining limits |
+| [`COMPARISON.md`](COMPARISON.md) | doos next to Docker, Podman, bubblewrap, nspawn, LXC/Incus, Apptainer, Distrobox, gVisor/Kata, with pros, cons, measurements and a conclusion |
+| [`AGENTS.md`](AGENTS.md) | Rules for AI agents, release workflow |
+| [`CHANGELOG.md`](CHANGELOG.md) | Release history |
+
+Not sure whether doos is the right tool? Read the conclusion of [COMPARISON](COMPARISON.md) first.
@@ -1,6 +1,8 @@
 # doos specification

-Authoritative description of behavior and limits. Code constants live in `src/spec.hpp`; change both together.
+**Docs:** [README](README.md) · [SPEC](SPEC.md) · [ARCHITECTURE](docs/ARCHITECTURE.md) · [RESEARCH](docs/RESEARCH.md) · [COMPARISON](COMPARISON.md) · [AGENTS](AGENTS.md) · [CHANGELOG](CHANGELOG.md)
+
+Authoritative description of behavior and limits. Code constants live in `src/spec.hpp`; change both together. How the code implements it: [ARCHITECTURE](docs/ARCHITECTURE.md). Measurements and remaining limits: [RESEARCH](docs/RESEARCH.md). How it differs from other tools: [COMPARISON](COMPARISON.md). Usage: [README](README.md).

 ## CLI

@@ -1 +1 @@
-1.1.1
+1.1.4
@@ -1,6 +1,8 @@
 # Architecture

-How doos works internally. Behavior contract is `SPEC.md`; this explains the code.
+**Docs:** [README](../README.md) · [SPEC](../SPEC.md) · [ARCHITECTURE](ARCHITECTURE.md) · [RESEARCH](RESEARCH.md) · [COMPARISON](../COMPARISON.md) · [AGENTS](../AGENTS.md) · [CHANGELOG](../CHANGELOG.md)
+
+How doos works internally. The behavior contract is [SPEC](../SPEC.md); this explains the code. Why it was built this way and what was measured: [RESEARCH](RESEARCH.md); how it compares with other tools: [COMPARISON](../COMPARISON.md).

 ## File map

@@ -1,5 +1,7 @@
 # Research: user-level containers with overlayfs

+**Docs:** [README](../README.md) · [SPEC](../SPEC.md) · [ARCHITECTURE](ARCHITECTURE.md) · [RESEARCH](RESEARCH.md) · [COMPARISON](../COMPARISON.md) · [AGENTS](../AGENTS.md) · [CHANGELOG](../CHANGELOG.md)
+
 Measured on this machine: Ubuntu, kernel 7.0.0-38, AppArmor 5.0.2, slirp4netns 1.3.3, uid 1000, no root at runtime.

 ## What works unprivileged
@@ -258,6 +260,10 @@ delegated with `cpuset cpu io memory pids`). The end-to-end suite passed 83 chec
 | Size accounting | 80 MB written by container uid 5 in a mode-700 directory (invisible to the host user) against `--size 64m`: frozen; cleanup from a second attach thawed it |
 | Attach time (5 runs) | host or image, slirp: about 40 ms each; pasta: about 65 ms; `--net none`: about 10 ms |

+![Horizontal bar chart of CPU-seconds used by four busy loops in one container, without a limit and with --cpus 0.5. The limited container is highlighted. Values are in the sentence below.](charts/cpu.svg)
+
+The chart above is generated from `docs/charts/bench.json` (`make bench`, then `make charts`); it shows <!--bench:cpu_free-->16.3<!--/bench--> CPU-seconds without a limit and <!--bench:cpu_half-->2.3<!--/bench--> with `--cpus 0.5`; the startup and size charts live in [COMPARISON](../COMPARISON.md).
+
 ### What changed in behavior (and why it is not a loss)

 - Hardening can be switched off per container (`--hardening relaxed`), so nothing that needs the denied calls is lost, it is opt-in now.
@@ -278,3 +284,5 @@ delegated with `cpuset cpu io memory pids`). The end-to-end suite passed 83 chec
 - Without `gpg` installed, signature checks are skipped with a printed note.
 - pasta is not signalled by doos (it drops uid); it exits by itself within about a second after the container's namespace is gone. Measured: no pasta process remained after 1.5 s.
 - seccomp is a denylist. A new kernel syscall is allowed until it is added to the list.
+
+For how doos stacks up against Docker, Podman and others, with these measurements next to theirs, see [COMPARISON](../COMPARISON.md). Behavior contract: [SPEC](../SPEC.md). Code map: [ARCHITECTURE](ARCHITECTURE.md). Usage: [README](../README.md).
@@ -0,0 +1,121 @@
+{
+  "meta": {
+    "date": "2026-10-23",
+    "kernel": "7.0.0-38-generic",
+    "cpus": 12,
+    "runs": 10,
+    "doos": "1.1.2",
+    "podman": "podman version 5.7.0",
+    "docker": "Docker version 29.1.3, build 29.1.3-0ubuntu4.1",
+    "bubblewrap": "bubblewrap 0.11.1",
+    "src_lines": 3445,
+    "shared_libs": 6
+  },
+  "startup": [
+    {
+      "id": "bwrap_none",
+      "label": "bubblewrap (host /usr)",
+      "group": "none",
+      "kind": "other",
+      "ms": 8.4
+    },
+    {
+      "id": "bwrap_net",
+      "label": "bubblewrap (host /usr)",
+      "group": "default",
+      "kind": "other",
+      "ms": 7.4
+    },
+    {
+      "id": "doos_host_none",
+      "label": "doos host mode",
+      "group": "none",
+      "kind": "doos",
+      "ms": 18.7
+    },
+    {
+      "id": "doos_alpine_none",
+      "label": "doos Alpine image",
+      "group": "none",
+      "kind": "doos",
+      "ms": 13.8
+    },
+    {
+      "id": "doos_host_slirp",
+      "label": "doos host mode",
+      "group": "default",
+      "kind": "doos",
+      "ms": 56.9
+    },
+    {
+      "id": "doos_alpine_slirp",
+      "label": "doos Alpine image",
+      "group": "default",
+      "kind": "doos",
+      "ms": 52.1
+    },
+    {
+      "id": "doos_alpine_pasta",
+      "label": "doos Alpine, pasta",
+      "group": "default",
+      "kind": "doos",
+      "ms": 64.6
+    },
+    {
+      "id": "podman_none",
+      "label": "Podman Alpine",
+      "group": "none",
+      "kind": "other",
+      "ms": 207.8
+    },
+    {
+      "id": "podman_net",
+      "label": "Podman Alpine",
+      "group": "default",
+      "kind": "other",
+      "ms": 241.2
+    },
+    {
+      "id": "docker_none",
+      "label": "Docker Alpine (rootful)",
+      "group": "none",
+      "kind": "other",
+      "ms": 343.9
+    },
+    {
+      "id": "docker_net",
+      "label": "Docker Alpine (rootful)",
+      "group": "default",
+      "kind": "other",
+      "ms": 435.6
+    }
+  ],
+  "sizes": [
+    {
+      "id": "bwrap",
+      "bytes": 80424
+    },
+    {
+      "id": "crun",
+      "bytes": 483392
+    },
+    {
+      "id": "doos",
+      "bytes": 565680
+    },
+    {
+      "id": "docker",
+      "bytes": 31374336
+    },
+    {
+      "id": "podman",
+      "bytes": 43180896
+    }
+  ],
+  "cpu": {
+    "wall_seconds": 4.2,
+    "busy_loops": 4,
+    "unlimited": 16.26,
+    "half_cpu": 2.29
+  }
+}
@@ -0,0 +1,48 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 760 200" width="760" role="img" aria-labelledby="t d" style="max-width:100%;height:auto">
+<title id="t">CPU time used by four busy loops</title>
+<desc id="d">Two bars, CPU seconds consumed in 4.2 seconds of wall time. Without a limit 16.3, with --cpus 0.5 2.3.</desc>
+<style>
+  .bg { fill: var(--surface); }
+  .t1 { fill: var(--ink); font: 600 15px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .t2 { fill: var(--ink2); font: 400 12px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .lab { fill: var(--ink); font: 400 13px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .val { fill: var(--ink2); font: 400 12px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .tick { fill: var(--muted); font: 400 11px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .grid { stroke: var(--grid); stroke-width: 1; }
+  .axis { stroke: var(--axis); stroke-width: 1; }
+  .acc { fill: var(--accent); }
+  .oth { fill: var(--other); }
+  :root { --surface:#fcfcfb; --ink:#0b0b0b; --ink2:#52514e; --muted:#898781;
+          --grid:#e1e0d9; --axis:#c3c2b7; --accent:#2a78d6; --other:#898781; }
+  @media (prefers-color-scheme: dark) {
+    :root { --surface:#1a1a19; --ink:#f0efec; --ink2:#c3c2b7; --muted:#898781;
+            --grid:#2c2c2a; --axis:#383835; --accent:#3987e5; --other:#898781; }
+  }
+</style>
+<rect class="bg" x="0" y="0" width="760" height="200" rx="8"/>
+<text class="t1" x="20" y="28">CPU time used by 4 busy loops in one container (CPU-seconds)</text>
+<text class="t2" x="20" y="46">4.2 s of wall time. Half a CPU allows about 2.1 CPU-seconds; measured 2.3.</text>
+<rect class="oth" x="20" y="55" width="12" height="12" rx="2"/>
+<text class="t2" x="38" y="65">no limit</text>
+<rect class="acc" x="116" y="55" width="12" height="12" rx="2"/>
+<text class="t2" x="134" y="65">--cpus 0.5</text>
+<line class="grid" x1="170.0" y1="84" x2="170.0" y2="146"/>
+<text class="tick" x="170.0" y="162" text-anchor="middle">0</text>
+<line class="grid" x1="270.0" y1="84" x2="270.0" y2="146"/>
+<text class="tick" x="270.0" y="162" text-anchor="middle">4</text>
+<line class="grid" x1="370.0" y1="84" x2="370.0" y2="146"/>
+<text class="tick" x="370.0" y="162" text-anchor="middle">8</text>
+<line class="grid" x1="470.0" y1="84" x2="470.0" y2="146"/>
+<text class="tick" x="470.0" y="162" text-anchor="middle">12</text>
+<line class="grid" x1="570.0" y1="84" x2="570.0" y2="146"/>
+<text class="tick" x="570.0" y="162" text-anchor="middle">16</text>
+<line class="grid" x1="670.0" y1="84" x2="670.0" y2="146"/>
+<text class="tick" x="670.0" y="162" text-anchor="middle">20</text>
+<text class="lab" x="160" y="105.0" text-anchor="end">no limit</text>
+<path class="oth" d="M170.0,92.0 h402.5 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-402.5 z"><title>no limit: 16.3 CPU-seconds</title></path>
+<text class="val" x="584.5" y="105.0">16.3 CPU-s</text>
+<text class="lab" x="160" y="135.0" text-anchor="end">--cpus 0.5</text>
+<path class="acc" d="M170.0,122.0 h53.2 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-53.2 z"><title>--cpus 0.5: 2.3 CPU-seconds</title></path>
+<text class="val" x="235.2" y="135.0">2.3 CPU-s</text>
+<line class="axis" x1="170" y1="84" x2="170" y2="146"/>
+</svg>
@@ -0,0 +1,292 @@
+#!/usr/bin/env python3
+"""Builds the SVG charts and the generated numbers in COMPARISON.md / RESEARCH.md from bench.json.
+
+    python3 docs/charts/make_charts.py            (or: make charts)
+
+Input : docs/charts/bench.json   (written by scripts/bench.sh, i.e. make bench)
+Output: docs/charts/startup.svg, size.svg, cpu.svg
+        the blocks between <!-- bench:NAME:begin --> and <!-- bench:NAME:end --> in the markdown documents
+        and inline values written as <!--bench:KEY-->text<!--/bench-->
+
+Design rules (the dataviz method): one form per job, doos in the accent hue and everything else in the neutral
+gray (emphasis), bars at most 24 px with a 4 px rounded data end, direct value labels at the bar tip, a legend,
+hairline solid grid, text in ink tokens (never the series color), light and dark palettes selected separately.
+Colors were checked with the dataviz validator: contrast, CVD separation and the normal-vision floor pass in
+both modes; its "chroma floor" flag on the gray is expected, the gray is the de-emphasis neutral by design.
+Standard library only.
+"""
+import json
+import math
+import re
+import sys
+from pathlib import Path
+from xml.sax.saxutils import escape
+
+HERE = Path(__file__).resolve().parent
+ROOT = HERE.parent.parent
+DATA = HERE / "bench.json"
+
+# ---- palette (dataviz reference instance) ---------------------------------------------------------------
+LIGHT = dict(surface="#fcfcfb", ink="#0b0b0b", ink2="#52514e", muted="#898781", grid="#e1e0d9", axis="#c3c2b7",
+             accent="#2a78d6", other="#898781")
+DARK = dict(surface="#1a1a19", ink="#f0efec", ink2="#c3c2b7", muted="#898781", grid="#2c2c2a", axis="#383835",
+            accent="#3987e5", other="#898781")
+
+STYLE = f"""
+  .bg {{ fill: var(--surface); }}
+  .t1 {{ fill: var(--ink); font: 600 15px system-ui, -apple-system, "Segoe UI", sans-serif; }}
+  .t2 {{ fill: var(--ink2); font: 400 12px system-ui, -apple-system, "Segoe UI", sans-serif; }}
+  .lab {{ fill: var(--ink); font: 400 13px system-ui, -apple-system, "Segoe UI", sans-serif; }}
+  .val {{ fill: var(--ink2); font: 400 12px system-ui, -apple-system, "Segoe UI", sans-serif; }}
+  .tick {{ fill: var(--muted); font: 400 11px system-ui, -apple-system, "Segoe UI", sans-serif; }}
+  .grid {{ stroke: var(--grid); stroke-width: 1; }}
+  .axis {{ stroke: var(--axis); stroke-width: 1; }}
+  .acc {{ fill: var(--accent); }}
+  .oth {{ fill: var(--other); }}
+  :root {{ --surface:{LIGHT['surface']}; --ink:{LIGHT['ink']}; --ink2:{LIGHT['ink2']}; --muted:{LIGHT['muted']};
+          --grid:{LIGHT['grid']}; --axis:{LIGHT['axis']}; --accent:{LIGHT['accent']}; --other:{LIGHT['other']}; }}
+  @media (prefers-color-scheme: dark) {{
+    :root {{ --surface:{DARK['surface']}; --ink:{DARK['ink']}; --ink2:{DARK['ink2']}; --muted:{DARK['muted']};
+            --grid:{DARK['grid']}; --axis:{DARK['axis']}; --accent:{DARK['accent']}; --other:{DARK['other']}; }}
+  }}
+"""
+
+BAR_H = 18          # <= 24 px
+ROW = 30
+RADIUS = 4          # rounded data end only
+
+
+def bar_path(x, y, w, h, r=RADIUS):
+    """Bar growing from x to the right: square at the baseline, rounded at the data end."""
+    w = max(w, 1.0)
+    r = min(r, w, h / 2)
+    return (f"M{x:.1f},{y:.1f} h{w - r:.1f} a{r},{r} 0 0 1 {r},{r} v{h - 2 * r:.1f} "
+            f"a{r},{r} 0 0 1 {-r},{r} h{-(w - r):.1f} z")
+
+
+def svg_open(w, h, title, desc):
+    return (f'<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 {w} {h}" width="{w}" role="img" '
+            f'aria-labelledby="t d" style="max-width:100%;height:auto">\n<title id="t">{escape(title)}</title>\n'
+            f'<desc id="d">{escape(desc)}</desc>\n<style>{STYLE}</style>\n'
+            f'<rect class="bg" x="0" y="0" width="{w}" height="{h}" rx="8"/>\n')
+
+
+def legend(x, y, items):
+    out, cx = [], x
+    for cls, text in items:
+        out.append(f'<rect class="{cls}" x="{cx}" y="{y - 9}" width="12" height="12" rx="2"/>')
+        out.append(f'<text class="t2" x="{cx + 18}" y="{y + 1}">{escape(text)}</text>')
+        cx += 18 + 7 * len(text) + 22
+    return "\n".join(out)
+
+
+def fmt_ms(v):
+    return f"{v:.0f} ms" if v >= 100 else f"{v:.1f} ms".replace(".0 ms", " ms")
+
+
+def fmt_bytes(b):
+    if b < 1 << 20:
+        return f"{b / 1024:.0f} KiB"
+    return f"{b / (1 << 20):.1f} MiB"
+
+
+# ---- chart 1: startup time -----------------------------------------------------------------------------
+def startup_chart(d):
+    rows = [r for r in d["startup"] if r["ms"] is not None]
+    iv = inline_values(d)
+    groups = [("none", "No network"), ("default", "Default network")]
+    left, right, top = 200, 90, 78
+    width = 760
+    plot_w = width - left - right
+    vmax = max(r["ms"] for r in rows)
+    axis_max = math.ceil(vmax / 100.0) * 100
+    step = 100 if axis_max <= 600 else 200
+    scale = plot_w / axis_max
+
+    y = top
+    body = []
+    for key, name in groups:
+        grp = sorted([r for r in rows if r["group"] == key], key=lambda r: r["ms"])
+        if not grp:
+            continue
+        body.append(f'<text class="t1" x="{left - 190 + 12}" y="{y}">{name}</text>')
+        y += 14
+        y0 = y
+        for t in range(0, axis_max + 1, step):
+            x = left + t * scale
+            body.append(f'<line class="grid" x1="{x:.1f}" y1="{y0 - 2}" x2="{x:.1f}" y2="{y0 + len(grp) * ROW}"/>')
+        for r in grp:
+            cls = "acc" if r["kind"] == "doos" else "oth"
+            w = r["ms"] * scale
+            by = y + (ROW - BAR_H) / 2
+            body.append(f'<text class="lab" x="{left - 10}" y="{by + BAR_H / 2 + 4:.1f}" text-anchor="end">{escape(r["label"])}</text>')
+            body.append(f'<path class="{cls}" d="{bar_path(left, by, w, BAR_H)}"><title>{escape(r["label"])}: {fmt_ms(r["ms"])}</title></path>')
+            body.append(f'<text class="val" x="{left + w + 8:.1f}" y="{by + BAR_H / 2 + 4:.1f}">{fmt_ms(r["ms"])}</text>')
+            y += ROW
+        body.append(f'<line class="axis" x1="{left}" y1="{y0 - 2}" x2="{left}" y2="{y}"/>')
+        for t in range(0, axis_max + 1, step):
+            body.append(f'<text class="tick" x="{left + t * scale:.1f}" y="{y + 14}" text-anchor="middle">{t}</text>')
+        y += 38
+    height = int(y - 10)
+    m = d["meta"]
+    out = svg_open(width, height, "Time to start a trivial command",
+                   f"Horizontal bars, milliseconds, lower is better. doos: {iv['doos_range']}; Podman: {iv['podman_range']}; "
+                   f"Docker: {iv['docker_range']}; bubblewrap: {iv['bwrap_range']}. Exact values are in the table in COMPARISON.md.")
+    out += '<text class="t1" x="20" y="28">Time to start a trivial command (ms, lower is better)</text>\n'
+    out += (f'<text class="t2" x="20" y="46">{m["runs"]} runs each, average, warm cache, {m["date"]}, kernel {escape(m["kernel"])}, '
+            f'{m["cpus"]} cores. Docker is a rootful daemon.</text>\n')
+    out += legend(20, 64, [("acc", "doos"), ("oth", "other tools")]) + "\n"
+    out += "\n".join(body) + "\n</svg>\n"
+    return out
+
+
+# ---- chart 2: binary size (log scale) --------------------------------------------------------------------
+def size_chart(d):
+    names = {"bwrap": "bubblewrap", "crun": "crun (OCI runtime)", "doos": "doos", "docker": "docker CLI",
+             "podman": "podman"}
+    rows = sorted([r for r in d["sizes"] if r["bytes"] is not None], key=lambda r: r["bytes"])
+    left, right, top, width = 170, 100, 96, 760
+    plot_w = width - left - right
+    lo, hi = math.log10(50 * 1024), math.log10(100 * (1 << 20))
+    sx = lambda b: (math.log10(b) - lo) / (hi - lo) * plot_w
+    ticks = [(100 * 1024, "100 KiB"), (1 << 20, "1 MiB"), (10 << 20, "10 MiB"), (100 << 20, "100 MiB")]
+    y0 = top + 8
+    body = []
+    for b, lab in ticks:
+        x = left + sx(b)
+        body.append(f'<line class="grid" x1="{x:.1f}" y1="{y0 - 2}" x2="{x:.1f}" y2="{y0 + len(rows) * ROW}"/>')
+        body.append(f'<text class="tick" x="{x:.1f}" y="{y0 + len(rows) * ROW + 16}" text-anchor="middle">{lab}</text>')
+    y = y0
+    for r in rows:
+        cls = "acc" if r["id"] == "doos" else "oth"
+        w = sx(r["bytes"])
+        by = y + (ROW - BAR_H) / 2
+        body.append(f'<text class="lab" x="{left - 10}" y="{by + BAR_H / 2 + 4:.1f}" text-anchor="end">{names.get(r["id"], r["id"])}</text>')
+        body.append(f'<path class="{cls}" d="{bar_path(left, by, w, BAR_H)}"><title>{names.get(r["id"], r["id"])}: {fmt_bytes(r["bytes"])}</title></path>')
+        body.append(f'<text class="val" x="{left + w + 8:.1f}" y="{by + BAR_H / 2 + 4:.1f}">{fmt_bytes(r["bytes"])}</text>')
+        y += ROW
+    body.append(f'<line class="axis" x1="{left}" y1="{y0 - 2}" x2="{left}" y2="{y}"/>')
+    height = int(y + 62)
+    out = svg_open(width, height, "Size of the main executable",
+                   "Horizontal bars on a logarithmic axis: " +
+                   ", ".join(f"{names.get(r['id'], r['id'])} {fmt_bytes(r['bytes'])}" for r in rows) + ".")
+    out += '<text class="t1" x="20" y="28">Size of the main executable (log scale, smaller is smaller)</text>\n'
+    out += '<text class="t2" x="20" y="46">The docker CLI excludes dockerd and containerd.</text>\n'
+    out += '<text class="t2" x="20" y="62">doos excludes the helpers it calls (slirp4netns, pasta, newuidmap, curl, tar).</text>\n'
+    out += legend(20, 82, [("acc", "doos"), ("oth", "other tools")]) + "\n"
+    out += "\n".join(body) + f'\n<text class="t2" x="{left}" y="{y + 38}">Axis is logarithmic: each gridline is 10x the previous one.</text>\n</svg>\n'
+    return out
+
+
+# ---- chart 3: cpu limit ----------------------------------------------------------------------------------
+def cpu_chart(d):
+    c = d["cpu"]
+    if c["unlimited"] is None or c["half_cpu"] is None:
+        return None
+    rows = [("no limit", c["unlimited"], "oth"), ("--cpus 0.5", c["half_cpu"], "acc")]
+    left, right, top, width = 170, 90, 78, 760
+    plot_w = width - left - right
+    axis_max = math.ceil(c["unlimited"] / 4.0) * 4
+    scale = plot_w / axis_max
+    y0 = top + 8
+    body = []
+    for t in range(0, axis_max + 1, 4):
+        x = left + t * scale
+        body.append(f'<line class="grid" x1="{x:.1f}" y1="{y0 - 2}" x2="{x:.1f}" y2="{y0 + len(rows) * ROW}"/>')
+        body.append(f'<text class="tick" x="{x:.1f}" y="{y0 + len(rows) * ROW + 16}" text-anchor="middle">{t}</text>')
+    y = y0
+    for lab, v, cls in rows:
+        w = v * scale
+        by = y + (ROW - BAR_H) / 2
+        body.append(f'<text class="lab" x="{left - 10}" y="{by + BAR_H / 2 + 4:.1f}" text-anchor="end">{lab}</text>')
+        body.append(f'<path class="{cls}" d="{bar_path(left, by, w, BAR_H)}"><title>{lab}: {v:.1f} CPU-seconds</title></path>')
+        body.append(f'<text class="val" x="{left + w + 8:.1f}" y="{by + BAR_H / 2 + 4:.1f}">{v:.1f} CPU-s</text>')
+        y += ROW
+    body.append(f'<line class="axis" x1="{left}" y1="{y0 - 2}" x2="{left}" y2="{y}"/>')
+    ideal = 0.5 * c["wall_seconds"]
+    height = int(y + 54)
+    out = svg_open(width, height, "CPU time used by four busy loops",
+                   f"Two bars, CPU seconds consumed in {c['wall_seconds']} seconds of wall time. Without a limit {c['unlimited']:.1f}, "
+                   f"with --cpus 0.5 {c['half_cpu']:.1f}.")
+    out += '<text class="t1" x="20" y="28">CPU time used by 4 busy loops in one container (CPU-seconds)</text>\n'
+    out += (f'<text class="t2" x="20" y="46">{c["wall_seconds"]} s of wall time. Half a CPU allows about {ideal:.1f} CPU-seconds; '
+            f'measured {c["half_cpu"]:.1f}.</text>\n')
+    out += legend(20, 64, [("oth", "no limit"), ("acc", "--cpus 0.5")]) + "\n"
+    out += "\n".join(body) + "\n</svg>\n"
+    return out
+
+
+# ---- generated markdown ---------------------------------------------------------------------------------
+def startup_table(d):
+    rows = [r for r in d["startup"] if r["ms"] is not None]
+    lines = ["| Setup | Network | Time per run |", "|---|---|---|"]
+    order = {"none": 0, "default": 1}
+    for r in sorted(rows, key=lambda r: (order[r["group"]], r["ms"])):
+        net = "none" if r["group"] == "none" else "default"
+        lines.append(f"| {r['label']} | {net} | {fmt_ms(r['ms'])} |")
+    return "\n".join(lines)
+
+
+def size_table(d):
+    names = {"bwrap": "bubblewrap", "crun": "crun", "doos": "doos", "docker": "docker CLI", "podman": "podman"}
+    lines = ["| Executable | Size |", "|---|---|"]
+    for r in sorted([r for r in d["sizes"] if r["bytes"] is not None], key=lambda r: r["bytes"]):
+        lines.append(f"| {names.get(r['id'], r['id'])} | {fmt_bytes(r['bytes'])} ({r['bytes']:,} bytes) |")
+    return "\n".join(lines)
+
+
+def inline_values(d):
+    ms = lambda ids: [r["ms"] for r in d["startup"] if r["id"].startswith(ids) and r["ms"] is not None]
+    rng = lambda xs: f"{min(xs):.0f} to {max(xs):.0f} ms" if xs else "n/a"
+    size = {r["id"]: r["bytes"] for r in d["sizes"] if r["bytes"] is not None}
+    v = {
+        "doos_range": rng(ms("doos")),
+        "podman_range": rng(ms("podman")),
+        "docker_range": rng(ms("docker")),
+        "bwrap_range": rng(ms("bwrap")),
+        "doos_size": fmt_bytes(size["doos"]) if "doos" in size else "n/a",
+        "docker_size": fmt_bytes(size["docker"]) if "docker" in size else "n/a",
+        "podman_size": fmt_bytes(size["podman"]) if "podman" in size else "n/a",
+        "bench_date": d["meta"]["date"],
+        "src_lines": f'{d["meta"].get("src_lines", 0):,}',
+        "shared_libs": str(d["meta"].get("shared_libs", "n/a")),
+        "runs": str(d["meta"]["runs"]),
+        "versions": f'doos {d["meta"]["doos"]}, {d["meta"]["docker"].replace("Docker version ", "Docker ").split(",")[0]}, '
+                    f'{d["meta"]["podman"].replace("podman version ", "Podman ")}, {d["meta"]["bubblewrap"].replace("bubblewrap ", "bubblewrap ")}',
+        "cpu_free": f'{d["cpu"]["unlimited"]:.1f}' if d["cpu"]["unlimited"] is not None else "n/a",
+        "cpu_half": f'{d["cpu"]["half_cpu"]:.1f}' if d["cpu"]["half_cpu"] is not None else "n/a",
+    }
+    return v
+
+
+def patch_markdown(path, blocks, inline):
+    s = path.read_text()
+    orig = s
+    for name, text in blocks.items():
+        pat = re.compile(rf"(<!-- bench:{name}:begin -->\n).*?(\n<!-- bench:{name}:end -->)", re.S)
+        s = pat.sub(lambda m: m.group(1) + text + m.group(2), s)
+    s = re.sub(r"<!--bench:(\w+)-->.*?<!--/bench-->",
+               lambda m: f"<!--bench:{m.group(1)}-->{inline.get(m.group(1), 'n/a')}<!--/bench-->", s)
+    if s != orig:
+        path.write_text(s)
+        return True
+    return False
+
+
+def main():
+    if not DATA.exists():
+        sys.exit("no bench.json: run `make bench` first (it needs the built binary)")
+    d = json.loads(DATA.read_text())
+    (HERE / "startup.svg").write_text(startup_chart(d))
+    (HERE / "size.svg").write_text(size_chart(d))
+    cpu = cpu_chart(d)
+    if cpu:
+        (HERE / "cpu.svg").write_text(cpu)
+    blocks = {"startup": startup_table(d), "sizes": size_table(d)}
+    inline = inline_values(d)
+    changed = [p.name for p in (ROOT / "COMPARISON.md", ROOT / "docs" / "RESEARCH.md") if patch_markdown(p, blocks, inline)]
+    print("charts: startup.svg size.svg" + (" cpu.svg" if cpu else "") + (f"; updated {', '.join(changed)}" if changed else "; docs already current"))
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,55 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 760 316" width="760" role="img" aria-labelledby="t d" style="max-width:100%;height:auto">
+<title id="t">Size of the main executable</title>
+<desc id="d">Horizontal bars on a logarithmic axis: bubblewrap 79 KiB, crun (OCI runtime) 472 KiB, doos 552 KiB, docker CLI 29.9 MiB, podman 41.2 MiB.</desc>
+<style>
+  .bg { fill: var(--surface); }
+  .t1 { fill: var(--ink); font: 600 15px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .t2 { fill: var(--ink2); font: 400 12px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .lab { fill: var(--ink); font: 400 13px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .val { fill: var(--ink2); font: 400 12px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .tick { fill: var(--muted); font: 400 11px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .grid { stroke: var(--grid); stroke-width: 1; }
+  .axis { stroke: var(--axis); stroke-width: 1; }
+  .acc { fill: var(--accent); }
+  .oth { fill: var(--other); }
+  :root { --surface:#fcfcfb; --ink:#0b0b0b; --ink2:#52514e; --muted:#898781;
+          --grid:#e1e0d9; --axis:#c3c2b7; --accent:#2a78d6; --other:#898781; }
+  @media (prefers-color-scheme: dark) {
+    :root { --surface:#1a1a19; --ink:#f0efec; --ink2:#c3c2b7; --muted:#898781;
+            --grid:#2c2c2a; --axis:#383835; --accent:#3987e5; --other:#898781; }
+  }
+</style>
+<rect class="bg" x="0" y="0" width="760" height="316" rx="8"/>
+<text class="t1" x="20" y="28">Size of the main executable (log scale, smaller is smaller)</text>
+<text class="t2" x="20" y="46">The docker CLI excludes dockerd and containerd.</text>
+<text class="t2" x="20" y="62">doos excludes the helpers it calls (slirp4netns, pasta, newuidmap, curl, tar).</text>
+<rect class="acc" x="20" y="73" width="12" height="12" rx="2"/>
+<text class="t2" x="38" y="83">doos</text>
+<rect class="oth" x="88" y="73" width="12" height="12" rx="2"/>
+<text class="t2" x="106" y="83">other tools</text>
+<line class="grid" x1="214.5" y1="102" x2="214.5" y2="254"/>
+<text class="tick" x="214.5" y="270" text-anchor="middle">100 KiB</text>
+<line class="grid" x1="364.0" y1="102" x2="364.0" y2="254"/>
+<text class="tick" x="364.0" y="270" text-anchor="middle">1 MiB</text>
+<line class="grid" x1="512.0" y1="102" x2="512.0" y2="254"/>
+<text class="tick" x="512.0" y="270" text-anchor="middle">10 MiB</text>
+<line class="grid" x1="660.0" y1="102" x2="660.0" y2="254"/>
+<text class="tick" x="660.0" y="270" text-anchor="middle">100 MiB</text>
+<text class="lab" x="160" y="123.0" text-anchor="end">bubblewrap</text>
+<path class="oth" d="M170.0,110.0 h25.0 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-25.0 z"><title>bubblewrap: 79 KiB</title></path>
+<text class="val" x="207.0" y="123.0">79 KiB</text>
+<text class="lab" x="160" y="153.0" text-anchor="end">crun (OCI runtime)</text>
+<path class="oth" d="M170.0,140.0 h140.3 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-140.3 z"><title>crun (OCI runtime): 472 KiB</title></path>
+<text class="val" x="322.3" y="153.0">472 KiB</text>
+<text class="lab" x="160" y="183.0" text-anchor="end">doos</text>
+<path class="acc" d="M170.0,170.0 h150.4 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-150.4 z"><title>doos: 552 KiB</title></path>
+<text class="val" x="332.4" y="183.0">552 KiB</text>
+<text class="lab" x="160" y="213.0" text-anchor="end">docker CLI</text>
+<path class="oth" d="M170.0,200.0 h408.5 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-408.5 z"><title>docker CLI: 29.9 MiB</title></path>
+<text class="val" x="590.5" y="213.0">29.9 MiB</text>
+<text class="lab" x="160" y="243.0" text-anchor="end">podman</text>
+<path class="oth" d="M170.0,230.0 h429.0 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-429.0 z"><title>podman: 41.2 MiB</title></path>
+<text class="val" x="611.0" y="243.0">41.2 MiB</text>
+<line class="axis" x1="170" y1="102" x2="170" y2="254"/>
+<text class="t2" x="170" y="292">Axis is logarithmic: each gridline is 10x the previous one.</text>
+</svg>
@@ -0,0 +1,90 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 760 502" width="760" role="img" aria-labelledby="t d" style="max-width:100%;height:auto">
+<title id="t">Time to start a trivial command</title>
+<desc id="d">Horizontal bars, milliseconds, lower is better. doos: 14 to 65 ms; Podman: 208 to 241 ms; Docker: 344 to 436 ms; bubblewrap: 7 to 8 ms. Exact values are in the table in COMPARISON.md.</desc>
+<style>
+  .bg { fill: var(--surface); }
+  .t1 { fill: var(--ink); font: 600 15px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .t2 { fill: var(--ink2); font: 400 12px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .lab { fill: var(--ink); font: 400 13px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .val { fill: var(--ink2); font: 400 12px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .tick { fill: var(--muted); font: 400 11px system-ui, -apple-system, "Segoe UI", sans-serif; }
+  .grid { stroke: var(--grid); stroke-width: 1; }
+  .axis { stroke: var(--axis); stroke-width: 1; }
+  .acc { fill: var(--accent); }
+  .oth { fill: var(--other); }
+  :root { --surface:#fcfcfb; --ink:#0b0b0b; --ink2:#52514e; --muted:#898781;
+          --grid:#e1e0d9; --axis:#c3c2b7; --accent:#2a78d6; --other:#898781; }
+  @media (prefers-color-scheme: dark) {
+    :root { --surface:#1a1a19; --ink:#f0efec; --ink2:#c3c2b7; --muted:#898781;
+            --grid:#2c2c2a; --axis:#383835; --accent:#3987e5; --other:#898781; }
+  }
+</style>
+<rect class="bg" x="0" y="0" width="760" height="502" rx="8"/>
+<text class="t1" x="20" y="28">Time to start a trivial command (ms, lower is better)</text>
+<text class="t2" x="20" y="46">10 runs each, average, warm cache, 2026-10-23, kernel 7.0.0-38-generic, 12 cores. Docker is a rootful daemon.</text>
+<rect class="acc" x="20" y="55" width="12" height="12" rx="2"/>
+<text class="t2" x="38" y="65">doos</text>
+<rect class="oth" x="88" y="55" width="12" height="12" rx="2"/>
+<text class="t2" x="106" y="65">other tools</text>
+<text class="t1" x="22" y="78">No network</text>
+<line class="grid" x1="200.0" y1="90" x2="200.0" y2="242"/>
+<line class="grid" x1="294.0" y1="90" x2="294.0" y2="242"/>
+<line class="grid" x1="388.0" y1="90" x2="388.0" y2="242"/>
+<line class="grid" x1="482.0" y1="90" x2="482.0" y2="242"/>
+<line class="grid" x1="576.0" y1="90" x2="576.0" y2="242"/>
+<line class="grid" x1="670.0" y1="90" x2="670.0" y2="242"/>
+<text class="lab" x="190" y="111.0" text-anchor="end">bubblewrap (host /usr)</text>
+<path class="oth" d="M200.0,98.0 h3.9 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-3.9 z"><title>bubblewrap (host /usr): 8.4 ms</title></path>
+<text class="val" x="215.9" y="111.0">8.4 ms</text>
+<text class="lab" x="190" y="141.0" text-anchor="end">doos Alpine image</text>
+<path class="acc" d="M200.0,128.0 h9.0 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-9.0 z"><title>doos Alpine image: 13.8 ms</title></path>
+<text class="val" x="221.0" y="141.0">13.8 ms</text>
+<text class="lab" x="190" y="171.0" text-anchor="end">doos host mode</text>
+<path class="acc" d="M200.0,158.0 h13.6 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-13.6 z"><title>doos host mode: 18.7 ms</title></path>
+<text class="val" x="225.6" y="171.0">18.7 ms</text>
+<text class="lab" x="190" y="201.0" text-anchor="end">Podman Alpine</text>
+<path class="oth" d="M200.0,188.0 h191.3 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-191.3 z"><title>Podman Alpine: 208 ms</title></path>
+<text class="val" x="403.3" y="201.0">208 ms</text>
+<text class="lab" x="190" y="231.0" text-anchor="end">Docker Alpine (rootful)</text>
+<path class="oth" d="M200.0,218.0 h319.3 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-319.3 z"><title>Docker Alpine (rootful): 344 ms</title></path>
+<text class="val" x="531.3" y="231.0">344 ms</text>
+<line class="axis" x1="200" y1="90" x2="200" y2="242"/>
+<text class="tick" x="200.0" y="256" text-anchor="middle">0</text>
+<text class="tick" x="294.0" y="256" text-anchor="middle">100</text>
+<text class="tick" x="388.0" y="256" text-anchor="middle">200</text>
+<text class="tick" x="482.0" y="256" text-anchor="middle">300</text>
+<text class="tick" x="576.0" y="256" text-anchor="middle">400</text>
+<text class="tick" x="670.0" y="256" text-anchor="middle">500</text>
+<text class="t1" x="22" y="280">Default network</text>
+<line class="grid" x1="200.0" y1="292" x2="200.0" y2="474"/>
+<line class="grid" x1="294.0" y1="292" x2="294.0" y2="474"/>
+<line class="grid" x1="388.0" y1="292" x2="388.0" y2="474"/>
+<line class="grid" x1="482.0" y1="292" x2="482.0" y2="474"/>
+<line class="grid" x1="576.0" y1="292" x2="576.0" y2="474"/>
+<line class="grid" x1="670.0" y1="292" x2="670.0" y2="474"/>
+<text class="lab" x="190" y="313.0" text-anchor="end">bubblewrap (host /usr)</text>
+<path class="oth" d="M200.0,300.0 h3.0 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-3.0 z"><title>bubblewrap (host /usr): 7.4 ms</title></path>
+<text class="val" x="215.0" y="313.0">7.4 ms</text>
+<text class="lab" x="190" y="343.0" text-anchor="end">doos Alpine image</text>
+<path class="acc" d="M200.0,330.0 h45.0 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-45.0 z"><title>doos Alpine image: 52.1 ms</title></path>
+<text class="val" x="257.0" y="343.0">52.1 ms</text>
+<text class="lab" x="190" y="373.0" text-anchor="end">doos host mode</text>
+<path class="acc" d="M200.0,360.0 h49.5 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-49.5 z"><title>doos host mode: 56.9 ms</title></path>
+<text class="val" x="261.5" y="373.0">56.9 ms</text>
+<text class="lab" x="190" y="403.0" text-anchor="end">doos Alpine, pasta</text>
+<path class="acc" d="M200.0,390.0 h56.7 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-56.7 z"><title>doos Alpine, pasta: 64.6 ms</title></path>
+<text class="val" x="268.7" y="403.0">64.6 ms</text>
+<text class="lab" x="190" y="433.0" text-anchor="end">Podman Alpine</text>
+<path class="oth" d="M200.0,420.0 h222.7 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-222.7 z"><title>Podman Alpine: 241 ms</title></path>
+<text class="val" x="434.7" y="433.0">241 ms</text>
+<text class="lab" x="190" y="463.0" text-anchor="end">Docker Alpine (rootful)</text>
+<path class="oth" d="M200.0,450.0 h405.5 a4,4 0 0 1 4,4 v10.0 a4,4 0 0 1 -4,4 h-405.5 z"><title>Docker Alpine (rootful): 436 ms</title></path>
+<text class="val" x="617.5" y="463.0">436 ms</text>
+<line class="axis" x1="200" y1="292" x2="200" y2="474"/>
+<text class="tick" x="200.0" y="488" text-anchor="middle">0</text>
+<text class="tick" x="294.0" y="488" text-anchor="middle">100</text>
+<text class="tick" x="388.0" y="488" text-anchor="middle">200</text>
+<text class="tick" x="482.0" y="488" text-anchor="middle">300</text>
+<text class="tick" x="576.0" y="488" text-anchor="middle">400</text>
+<text class="tick" x="670.0" y="488" text-anchor="middle">500</text>
+</svg>
@@ -0,0 +1,94 @@
+#!/usr/bin/env bash
+# Reproducible measurements behind COMPARISON.md and the charts. Writes JSON (default docs/charts/bench.json).
+# Tools that are not installed or not usable are skipped and recorded as null, never guessed.
+# Usage: scripts/bench.sh [output.json]      Environment: BENCH_RUNS (default 10)
+set -u
+cd "$(dirname "$0")/.."
+OUT=${1:-docs/charts/bench.json}
+RUNS=${BENCH_RUNS:-10}
+DOOS=$(readlink -f build/doos)
+[ -x "$DOOS" ] || { echo "bench: build first (make build)" >&2; exit 1; }
+export DOOS_HOME=$(mktemp -d)
+trap '"$DOOS" delete --yes ha >/dev/null 2>&1; rm -rf "$DOOS_HOME"' EXIT
+
+have() { command -v "$1" >/dev/null 2>&1; }
+now() { date +%s%N; }
+# avg_ms LABEL CMD...   -> prints average ms per run, or "null" when the command fails
+avg_ms() {
+  timeout 180 "$@" >/dev/null 2>&1 || { echo null; return; }      # warm-up, may pull an image
+  local s e i; s=$(now)
+  for i in $(seq "$RUNS"); do timeout 60 "$@" >/dev/null 2>&1 || { echo null; return; }; done
+  e=$(now); awk -v s="$s" -v e="$e" -v n="$RUNS" 'BEGIN{printf "%.1f", (e-s)/1e6/n}'
+}
+note() { echo "bench: $*" >&2; }
+
+# ---- setup doos containers (host mode needs nothing, image mode downloads Alpine once)
+"$DOOS" image add alpine >/dev/null 2>&1 || note "alpine image unavailable, image-mode rows will be null"
+"$DOOS" get_or_create --net none hn >/dev/null 2>&1
+"$DOOS" get_or_create hs >/dev/null 2>&1
+"$DOOS" get_or_create --image alpine --net none an >/dev/null 2>&1
+"$DOOS" get_or_create --image alpine as >/dev/null 2>&1
+"$DOOS" get_or_create --image alpine --net pasta ap >/dev/null 2>&1
+
+BW_BASE="--unshare-user --unshare-pid --unshare-ipc --unshare-uts --die-with-parent --ro-bind /usr /usr --symlink usr/bin /bin --symlink usr/lib /lib --symlink usr/lib64 /lib64 --proc /proc --dev /dev"
+rows=()
+add() { rows+=("{\"id\":\"$1\",\"label\":\"$2\",\"group\":\"$3\",\"kind\":\"$4\",\"ms\":$5}"); }
+
+echo "bench: startup, $RUNS runs each" >&2
+if have bwrap; then
+  add bwrap_none "bubblewrap (host /usr)" none other "$(avg_ms bwrap --unshare-all $BW_BASE true)"
+  add bwrap_net  "bubblewrap (host /usr)" default other "$(avg_ms bwrap $BW_BASE true)"
+fi
+add doos_host_none  "doos host mode"    none    doos "$(avg_ms "$DOOS" attach hn -- true)"
+add doos_alpine_none "doos Alpine image" none   doos "$(avg_ms "$DOOS" attach an -- true)"
+add doos_host_slirp "doos host mode"    default doos "$(avg_ms "$DOOS" attach hs -- true)"
+add doos_alpine_slirp "doos Alpine image" default doos "$(avg_ms "$DOOS" attach as -- true)"
+add doos_alpine_pasta "doos Alpine, pasta" default doos "$(avg_ms "$DOOS" attach ap -- true)"
+if have podman; then
+  add podman_none "Podman Alpine" none other "$(avg_ms podman run --rm --network none alpine true)"
+  add podman_net  "Podman Alpine" default other "$(avg_ms podman run --rm alpine true)"
+fi
+if have docker; then
+  add docker_none "Docker Alpine (rootful)" none other "$(avg_ms docker run --rm --network none alpine true)"
+  add docker_net  "Docker Alpine (rootful)" default other "$(avg_ms docker run --rm alpine true)"
+fi
+
+# ---- binary sizes
+size_of() { local p; p=$(command -v "$1" 2>/dev/null) && stat -L -c %s "$p" || echo null; }
+sizes=()
+for t in bwrap crun doos docker podman; do
+  if [ "$t" = doos ]; then b=$(stat -c %s "$DOOS"); else b=$(size_of "$t"); fi
+  sizes+=("{\"id\":\"$t\",\"bytes\":$b}")
+done
+
+# ---- cpu limit: 4 busy loops for 4 seconds, CPU seconds consumed (user+sys)
+cpu_run() {
+  /usr/bin/time -f "%U %S" "$DOOS" attach "$1" -- sh -c 'for i in 1 2 3 4; do (timeout 4 sh -c "while :; do :; done" &); done; sleep 4.2' 2>&1 >/dev/null | tail -1 | awk '{printf "%.2f", $1+$2}'
+}
+echo "bench: cpu limit" >&2
+"$DOOS" get_or_create hc >/dev/null 2>&1; "$DOOS" get_or_create --cpus 0.5 hc5 >/dev/null 2>&1
+cpu_free=$(cpu_run hc); cpu_half=$(cpu_run hc5)
+"$DOOS" delete --yes hc >/dev/null 2>&1; "$DOOS" delete --yes hc5 >/dev/null 2>&1
+
+ver() { "$@" 2>/dev/null | head -1 | tr -d '"\\' ; }
+json_join() { local IFS=,; echo "$*"; }
+cat > "$OUT" <<JSON
+{
+  "meta": {
+    "date": "$(date +%F)",
+    "kernel": "$(uname -r)",
+    "cpus": $(nproc),
+    "runs": $RUNS,
+    "doos": "$(cat VERSION)",
+    "podman": "$(have podman && ver podman --version || echo none)",
+    "docker": "$(have docker && ver docker --version || echo none)",
+    "bubblewrap": "$(have bwrap && ver bwrap --version || echo none)",
+    "src_lines": $(cat src/*.cpp src/*.hpp | wc -l),
+    "shared_libs": $(ldd "$DOOS" | wc -l)
+  },
+  "startup": [$(json_join "${rows[@]}")],
+  "sizes": [$(json_join "${sizes[@]}")],
+  "cpu": {"wall_seconds": 4.2, "busy_loops": 4, "unlimited": ${cpu_free:-null}, "half_cpu": ${cpu_half:-null}}
+}
+JSON
+echo "bench: wrote $OUT" >&2
@@ -0,0 +1,35 @@
+#!/usr/bin/env bash
+# Checks the documentation set: every markdown file links to every other one, and every relative link resolves.
+# Usage: scripts/check-docs.sh   (exit 1 on any problem, lists them all)
+set -u
+cd "$(dirname "$0")/.."
+
+DOCS=(README.md SPEC.md COMPARISON.md AGENTS.md CHANGELOG.md docs/ARCHITECTURE.md docs/RESEARCH.md)
+bad=0
+fail() { echo "docs: $*"; bad=$((bad+1)); }
+
+# no markdown file may exist outside the checked set (so a new doc cannot dodge the rule)
+while IFS= read -r f; do
+  f=${f#./}
+  printf '%s\n' "${DOCS[@]}" | grep -qxF "$f" || fail "$f is not in the DOCS list of scripts/check-docs.sh"
+done < <(find . -name '*.md' -not -path './build/*' -not -path './.git/*' -not -path './prompt.md')  # prompt.md: the original task statement, not documentation
+
+for f in "${DOCS[@]}"; do
+  [ -f "$f" ] || { fail "$f is missing"; continue; }
+  dir=$(dirname "$f")
+  # every other document must be linked
+  for g in "${DOCS[@]}"; do
+    [ "$f" = "$g" ] && continue
+    rel=$(realpath --relative-to="$dir" "$g")
+    grep -qE "\]\(${rel//./\\.}(#[^)]*)?\)" "$f" || fail "$f does not link to $g (expected ]($rel))"
+  done
+  # every relative link must point at something that exists
+  while IFS= read -r target; do
+    case "$target" in http://*|https://*|mailto:*|\#*) continue ;; esac
+    path=${target%%#*}
+    [ -z "$path" ] && continue
+    [ -e "$dir/$path" ] || fail "$f links to missing $target"
+  done < <(grep -oE '\]\([^)]+\)' "$f" | sed -E 's/^\]\(//; s/\)$//')
+done
+
+[ "$bad" -eq 0 ] && echo "docs: ${#DOCS[@]} documents, all cross-linked, all links resolve" || { echo "docs: $bad problem(s)"; exit 1; }
@@ -22,6 +22,8 @@ if grep -rl $'\xe2\x80\x94' --exclude-dir=build --exclude-dir=.git . | grep -q .
  echo "em dash found, remove it first" >&2; exit 1
 fi

+python3 docs/charts/make_charts.py
+scripts/check-docs.sh
 echo "$new" > VERSION
 cmake -S . -B build >/dev/null
 cmake --build build
@@ -12,6 +12,7 @@ out()  { "$DOOS" attach t -- sh -c "$1" 2>&1; }
 trap '"$DOOS" delete --yes t >/dev/null 2>&1; rm -rf "$DOOS_HOME"' EXIT

 check "list empty"                  "$DOOS" list
+check "docs cross-linked, links resolve" "$(dirname "$0")/../scripts/check-docs.sh"
 check "get_or_create creates"       "$DOOS" get_or_create t
 [ "$("$DOOS" get_or_create t)" = "exists t" ] && ok "get_or_create idempotent" || bad "get_or_create idempotent"
 check "invalid name rejected"       bash -c "! '$DOOS' get_or_create 'Bad Name'"
@@ -186,8 +187,15 @@ echo "--- network examples"
 curl -sS -m 2 -o /dev/null http://127.0.0.1:8077/ >/dev/null 2>&1 && bad "slirp: no inbound path" || ok "slirp: no inbound path"
 wait %% 2>/dev/null
 "$DOOS" get_or_create --net pasta pa1 >/dev/null 2>&1; "$DOOS" get_or_create --net pasta pa2 >/dev/null 2>&1
-"$DOOS" attach pa1 -- timeout 12 python3 -m http.server 8066 --bind 0.0.0.0 >/dev/null 2>&1 & sleep 5
-[ "$("$DOOS" attach pa2 -- curl -sS -m 2 -o /dev/null -w '%{http_code}' http://127.0.0.1:8066/ 2>/dev/null)" = "200" ] && ok "pasta: container to container via loopback" || bad "pasta: container to container via loopback"
+"$DOOS" attach pa1 -- timeout 25 python3 -m http.server 8066 --bind 0.0.0.0 >/dev/null 2>&1 &
+# pasta notices a newly bound port after a few seconds: poll instead of sleeping a fixed time
+c2c=000
+for _ in $(seq 1 18); do
+  sleep 1
+  c2c=$("$DOOS" attach pa2 -- curl -sS -m 2 -o /dev/null -w '%{http_code}' http://127.0.0.1:8066/ 2>/dev/null)
+  [ "$c2c" = "200" ] && break
+done
+[ "$c2c" = "200" ] && ok "pasta: container to container via loopback" || bad "pasta: container to container via loopback"
 wait 2>/dev/null
 "$DOOS" delete --yes pa1 >/dev/null; "$DOOS" delete --yes pa2 >/dev/null; "$DOOS" delete --yes nets >/dev/null; "$DOOS" delete --yes iso >/dev/null
Author	SHA1	Message	Date
retoorandClaude Sonnet 5.5	3f4e96937c	Document security verdict versus Docker Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-23 23:26:57 +02:00
retoorandClaude Sonnet 5.5	97fa625a50	Charts and reproducible benchmarks, Makefile, expanded .gitignore Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-23 22:24:56 +02:00
retoorandClaude Sonnet 5.5	4fed40e5cd	Add COMPARISON.md, cross-link all documents, enforce with check-docs Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-23 22:18:42 +02:00
@@ -1 +1 @@
 .1.1
 .1.4