#+title: Reproducibility, Boostrappability and Functional Package Management
#+date: 2026-06-01 Mon
#+author: W. Kosior
#+email: wkosior@agh.edu.pl

* SolarWinds
- "Sunburst"
- 2020
- compromised build infrastructure
- ~18,000 customers affected
- motivated discussion & guidance development

* Chalk Backdoor
- 2025
- "I Got Pwned"
  - developer phished
  - honest about it
- *extremely* popular npm Registry package
  - terminal coloring
  - depended on by almost every project
- backdoor
  - crypto-stealing (typical for NPM registry)
  - millions of downloads in ~2 hours
  - unused "potential" ;)
- + a few other compromised packages

* The Build Process
#+ATTR_ORG: :width 100%
[[./11-build-process-overview.svg]]

* Kinds of Dependencies
- runtime dependencies
- development dependencies
  - build dependencies (compilers, test frameworks, etc.)
  - other tools (IDEs, linters, debuggers, etc.)

* Reproducibility
#+begin_example
hermeti​city         fixed build inputs      ​a bit of care
      |​                     |               ​      |
      |​                     |               ​      |
      +​------+              |              +​------+
       ​      |              |              |
       ​      V              V              V
       ​    bit-to-bit identical result each time
       ​                     |
       ​                     |
       ​                     V
       ​              identical hashes -----> multiparty verification
#+end_example

* Successful Multiparty Verification
#+ATTR_ORG: :width 100%
[[./11-rebuilds-no-contamination-diagram.svg]]

* Unsuccessful Multiparty Verification
#+ATTR_ORG: :width 100%
[[./11-rebuilds-contamination-diagram.svg]]

* Reproducibility & Hermeticity in Guidance
- Software Component Verification Standard
  - OWASP, 2020
  - no direct mention
- Supply Chain Levels for Software Artifacts
  - OSSF, 2023
  - in draft, later removed
  - might be re-introduced in later versions
- Secure Supply Chain Consumption Framework
  - briefly mentioned
  - OSSF, 2022
- Software Supply Chain Security Best Practices
  - CNCF, 2021
  - potentially leverageable when high assurance needed
- Securing the Software Supply Chain: Recommended Practices Guide
  - NSA, ODNI, and CISA, 2022
  - "advanced" mitigations, "additional protection"
  - some text copied from SLSA draft

* Reproducibility in Practice
- 1992-3 (Cygnus, GCC)
  - cross-compilation too!
  - wisdom later lost

* Reproducibility in Practice, Cont.
- 1992-3 (Cygnus, GCC)
- Gitian (2011)
  - VM for builds
  - Bitcoin Core builds until 2021

* Reproducibility in Practice, Cont…
- 1992-3 (Cygnus, GCC)
- Gitian (2011)
- Debian
  - DebConf13  (2013)
  - 24% reproducible in 2013
  - 2018 — APT transport with in-toto
    - not usable as of 2025 :(
  - 97% reproducible in 2025

* Reproducibility in Practice, Cont…
- 1992-3 (Cygnus, GCC)
- Gitian (2011)
- Debian
- other early adopters (2015)
  - Coreboot
  - OpenWrt
  - NetBSD
  - FreeBSD
  - Arch
  - Fedora

* Reproducibility in Practice, Cont…
- 1992-3 (Cygnus, GCC)
- Gitian (2011)
- Debian
- other early adopters (2015)
- reproducible container images
  - /(traditionally a reproducibility-unfriendly area)/
  - base Arch container (2024)
  - GNU Guix containers (2025)

* Reproducibility in Practice, Cont…
- 1992-3 (Cygnus, GCC)
- Gitian (2011)
- Debian
- other early adopters (2015)
- reproducible container images
- more at https://reproducible-builds.org

* Reproducibility CI
- https://reproducible-builds.org/citests/

* Sample Reproducibility Challenges
- timestamps
  - solution: set all to Epoch
- order of filenames
  - directory listings
  - created archives
  - solution: sort names
- paths (debugging information)
  - solutions:
    - exclude paths in DWARF
    - build in the same prefix
- casual *rebuildability* problems
  - build time bombs (e.g., expiring certs in tests)
  - no automated rebuildability for npm Registry, PyPI, etc.

* Diffoscope
#+begin_example
$ diffoscope somearchive.tar.gz otherarchive.tar.gz
--- somearchive.tar.gz
+++ otherarchive.tar.gz
│   --- somearchive.tar
├── +++ otherarchive.tar
│ ├── file list
│ │ @@ -1,2 +1,2 @@
│ │ -drwxr-xr-x   0 urz       (1000) urz       (1000)        0 2026-06-01 05:41:31.000000 somedir/
│ │ --rw-r--r--   0 urz       (1000) urz       (1000)      102 2026-06-01 05:41:32.000000 somedir/lecture-head
│ │ +drwxr-xr-x   0 urz       (1000) urz       (1000)        0 2026-06-01 05:41:10.000000 somedir/
│ │ +-rw-r--r--   0 urz       (1000) urz       (1000)      148 2026-06-01 05:41:10.000000 somedir/lecture-head
│ ├── somedir/lecture-head
│ │ @@ -1,4 +1,4 @@
│ │ -#+title: Software Repositories
│ │ -#+date: 2026-05-25 Mon
│ │ +#+title: Reproducibility, Boostrappability and Functional Package Management
│ │ +#+date: 2026-06-01 Mon
│ │  #+author: W. Kosior
│ │  #+email: wkosior@agh.edu.pl
#+end_example

* Diffoscope, Cont.
#+begin_example
$ diffoscope hello1 hello2
--- ../build-2026-06-01T0759/hello1
+++ hello2
├── readelf --wide --debug-dump=rawline {}
│ @@ -25,15 +25,15 @@
│    Opcode 9 has 1 arg
│    Opcode 10 has 0 args
│    Opcode 11 has 0 args
│    Opcode 12 has 1 arg
│
│   The Directory Table (offset 0x22, lines 2, columns 1):
│    Entry      Name
│ -  0  (line_strp)     (offset: 0x8): /tmp/build-2026-06-01T0759
│ +  0  (line_strp)     (offset: 0x8): /tmp/build-2026-06-01T0800
│    1  (line_strp)     (offset: 0x23): /usr/local/include
#+end_example

* Making Use of Reproducibility
- monitoring software distribution
  - easy
  - Debian, Arch, GNU Guix, Nix, etc.
- verifying software before installing/distributing
  - better guarantees
  - harder
  - Bitcoin Core
- reproducing research

* "Reflections on Trusting Trust"
- Ken Thompson, 1984
- replicating compiler backdoors
- conspiracy theory:
  #+begin_quote
  Proprietary compiler used to build the first GCC versions implanted a
  self-replicating backdoor in them that lives in all GCC binaries to this day.
  #+end_quote
  - now disproved :)
- solutions:
  - Diverse Double-Compiling (Wheeler, 2009)
  - bootstrappable builds

* Bootstrappable Builds
- verify dependencies of dependencies of dependencies, etc.
  - including build dependencies
- 2022 — full source bootstrap
  - hex0 — minimal hex assembler that can assemble its source code
  - hex1 — richer version of hex0 (with, e.g., label jumps)
  - catm — =cat= replacement buildable with hex0
  - hex2 — even richer counterpart of hex1 (with, e.g., absolute addresses)
  - other MesCC Tools (M0, cc_x86, M1, M2, get_machine)
    - can compile simple C
  - GNU Mes (Scheme Lisp interpreter & C compiler)
  - Tiny C Compiler (TCC)
  - GCC (2.95.3)
  - binutils, glibc
  - GCC (4.9.4)
  - more…

* Bootstrappable Builds, Cont.
- verify dependencies of dependencies of dependencies, etc.
- 2022 — full source bootstrap
- thanks to NLnet for funding!

* Bootstrappable Builds, Cont…
- verify dependencies of dependencies of dependencies, etc.
- 2022 — full source bootstrap
- thanks to NLnet for funding!
- what about kernel?
  - live-bootstrap project

* Traditional Package Management / Development
- in the past:
  - surely, 0x1e8266 is the memory address of the framebuffer
- in more modern times:
  - surely =/usr/bin/python= exists and is …
- Filesystem Hierarchy Standard (FHS)
  - =/etc= — configuration files
    =/bin= — statically-linked executables
  - =/usr/share= — architecture-independent files
  - =/usr/bin= — executables
  - =/usr/lib= — libraries
  - =/usr/libexec= — executables for use by programs
  - =/usr/local/bin=, =/usr/local/share= — non-repo software
  - etc.

* Traditional Package Management / Development — Challenges
- software that requires older/newer libraries
- using multiple software versions on a single system
- updating safely (atomically)
- rolling back updates
- ad-hoc installation & usage
- rootless installation
- patching dependencies deep in the tree
- bootstrappability & reproducibility
  - unlike Docker, virtualenv et al :(

* Functional Package Management
- "The Purely Functional Software Deployment Model"
  - Eelco Dolstra, 2006
- Nix package manager
- every program/library in its own directory
  - =/store/008naskq2zc7dq93fpz4ard66qiyzywy-libx11-1.8.10/=
  - =/store/9q279phakibws2s76paciw0g8hvxvl0p-libx11-1.8.12/=
  - =/store/a6kkajhmaymz2rmx7m29bp9aqh9pawrx-libx11-1.8.12/=
  - used as "prefix" (=lib/=, =bin/=, etc. inside)

* Functional Package Management, Cont.
- "The Purely Functional Software Deployment Model"
- Nix package manager
- every program/library in its own directory
- *built software is a function of build inputs*
- store names
  - hash of all sources & dependencies
  - dependency update → new dependee
    - tricks to limit rebuilds & space usage
- dependencies
  - creative use of rpath, etc.
    - e.g., =/store/a6kkajhmaymz2rmx7m29bp9aqh9pawrx-libx11-1.8.12/= points to
      =/store/yj053cys0724p7vs9kir808x7fivz17m-glibc-2.41=
  - no dependence on, e.g., =/usr/bin/python=
  - versions can coexist

* Functional Package Management, Cont…
- "The Purely Functional Software Deployment Model"
- Nix package manager
- every program/library in its own directory
- *built software is a function of build inputs*
- store names
- dependencies
  - creative use of rpath, etc.
    - e.g., =/store/a6kkajhmaymz2rmx7m29bp9aqh9pawrx-libx11-1.8.12/= points to
      =/store/yj053cys0724p7vs9kir808x7fivz17m-glibc-2.41=
  - no dependence on, e.g., =/usr/bin/python=
  - versions can coexist

* Nix
- NixOS
- Nixpkgs
  - stable & rolling release
- DSL for recipes
  - packaging as a code!
  - 1 file — many definitions
    - compare to Debian…
- Nix daemon
- local builds & substitutes
- declarative system configuration

* GNU Guix
- 2014
- Nix' sibling
- Guile Scheme (Lisp family)
  - package definitions
  - system configuration
  - package manager itself
- used for Bitcoin Core releases (since 2021)

* GNU Guix Features
- full source bootstrap merged in
- reproducibility tests
  - =guix challenge=
  - =guix build --rounds=
- hermeticity enforced on builds
- cross-compilation (works in some cases)
- ad-hoc shells
  - & containers
- grafts
- VM images, container images
- =guix deploy= (like Ansible, but for Guix)
- policies: libre software only
- patching packages (code & command line)
  - =guix shell yt-dlp mpv --with-branch=yt-dlp=master=
- time-traveling
- more!

* Functional Package Management — Drawbacks, Problems
- speed (recompilations)
- substitute availability
- limited compatibility (no FHS)
  - but: Guix' container shells with FHS emulation
- live updates
  - e.g., Apache config & declarative service definition
- feature-completeness
  - declarative configuration interfaces — all options of all services
  - secrets deployment (declaratively defined GNU Guix systems)
  - Gentoo =use= flags (equivalent still lacking)
  - package count (Debian > GNU Guix)
  - "cheating"
    - prebuilt npm et al packages in Nix
- no "stable" branch in GNU Guix
- Microsoft bought a GUIX trademark (used for sth else)

* Local Variables
# Local Variables:
# org-image-actual-width: nil
# End: