Gothenburg 2026 - Day 3 - Files what are they even good for?

Files - what are they even good for?

• How do we identify a tree of files in a useful way across different RB tools?
  • that said, what should be the goal for file set identity?
  • source trees?
  • build tools?
  • disk images?
  • comparison-accelerator?
  • content-addressed storage?
• There should be a distinction between a file set (as an identity) and a file set (as a transport)
• Existing work:
  • go dirhash
  • git tree hash
  • nar hash
  • bazel REAPI
• some consensus on what these standards support, but not 100% consistency (x bit, xattrs, empty dirs, etc.)
• some movement on order-independent hashing - potential for performance improvements in new standards