redo the entire lazy trees concept #7

New issue

Open

opened 2024-03-07 06:26:31 +00:00 by pennae · 0 comments

pennae commented

2024-03-07 06:26:31 +00:00

Owner

our plan for that is to fully strip back the input accessor infra they've put in place already, standardize on posix fs semantics because that's what nix was built on, decouple libexpr completely, and then start again from there
on a mountpoint-based system that synthesizes store paths for "lazy" sources, and is then taken apart by the external-to-libexpr fs shim
eg you'd get not /foo for a file foo in a zip archive but /nix/store/.mounted/source.zip/foo in the evaluator, and eval don't care

(ie, eat the accesses before hitting the actual fs [h/t jade])

lazy trees are busted. path semantics are unclear and often implemented incorrectly, accessors are semi-invisible and intrusive to the evaluator anyway. lazy trees should be reworked, probably on the basis of something proven: posix semantics and mountpoints our plan for that is to fully strip back the input accessor infra they've put in place already, standardize on posix fs semantics because that's what nix was built on, decouple libexpr completely, and then start again from there on a mountpoint-based system that synthesizes store paths for "lazy" sources, and is then taken apart by the external-to-libexpr fs shim eg you'd get not /foo for a file foo in a zip archive but /nix/store/.mounted/source.zip/foo in the evaluator, and eval don't care (ie, eat the accesses before hitting the actual fs [h/t jade])