Reproducible Builds: Unexpected Benefits and Problems
summary
Over 1000 reproducibility patches lead to concrete fixes: eliminate timestamps and hostnames, sort directory listings to avoid ext4 hash randomness, fix racey make -j builds and remove -march=native CPU embeds. Tools shown include autoclassify to binary-bisect differing bits, auto_provenance and KVM-based rebuilds, and nagpo to match official artifacts. Examples: PGO in gcc causes nonreproducible builds (turning off PGO costs ~8% speed), shim signing workflow works only when binaries stay identical.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:29Why do reproducible builds matter?
- 2:04What are reproducible builds?
- 3:18Typical problems
- 6:26Surprising problems
- 12:39Surprising benefits
- 16:53Debugging
related talks
47mReproducible builds
23mDemystifying Unreproducible Builds: What, Why, and How?
42mThe Long Road to Reproducible Builds
42mSecurity and Productivity Pick Two with Reproducible Builds
13mDon’t Trust, Verify. Reproducible Builds in Bitcoin
26mReproducible Postgres
21mReproducible Builds and sstate over CDN - Update on AGL's Infra and CIAT
07mReproducible Builds