actiondfs: shrink cached metadata and simplify inode, parsing, and backing I/O - #30

Merged

dzbarsky merged 20 commits into

mainfrom

codex/actiondfs-deslop-round-eight

Aug 2, 2026

Commits on Aug 2, 2026

actiondfs: check published directory metadata before node origins
dzbarsky
committed
e794ff2
actiondfs: trust staged backing files during write completion
dzbarsky
committed
d6acbd3
actiondfs: splice backing files without redundant credential swaps
dzbarsky
committed
2cb1006
actiondfs: omit file accounting for internal directory blobs
dzbarsky
committed
40da3df
actiondfs: initialize staged inodes once after backing creation
dzbarsky
committed
3a4aa09
actiondfs: trust validated immutable backing inode types
dzbarsky
committed
840759b
actiondfs: publish canonical cached directories without atomic exchange
dzbarsky
committed
131bab3
actiondfs: match identical blob digest pointers before CAS roots
dzbarsky
committed
ed1cdaa
actiondfs: match identical directory digests before CAS roots
dzbarsky
committed
e7eac19
actiondfs: compress cached child records with name offsets
dzbarsky
committed
2070282
actiondfs: allocate directory iteration state only for staged entries
dzbarsky
committed
7de37eb
actiondfs: skip sorting directories with one validated child type
dzbarsky
committed
5948d22
actiondfs: decode single-byte protobuf varints directly
dzbarsky
committed
9233204
actiondfs: avoid owner spinlocks when staged identities already match
dzbarsky
committed
28a49b0
actiondfs: assign immutable inode numbers only once
dzbarsky
committed
e7bd186
actiondfs: initialize only required protobuf child fields
dzbarsky
committed
24ec5a1
actiondfs: accumulate packed filename bytes while parsing children
dzbarsky
committed
34838bc
actiondfs: borrow guaranteed backing files on nonempty input reads
dzbarsky
committed
7d9d5c6
actiondfs: decode directory digests directly into cached children
dzbarsky
committed
964e370
actiondfs: publish immutable inode identity only in the hash initializer
dzbarsky
committed
d1bd7bb

Navigation Menu

actiondfs: shrink cached metadata and simplify inode, parsing, and backing I/O - #30

actiondfs: shrink cached metadata and simplify inode, parsing, and backing I/O#30
dzbarsky merged 20 commits into
mainfrom
codex/actiondfs-deslop-round-eight

Commits on Aug 2, 2026

actiondfs: check published directory metadata before node origins

actiondfs: trust staged backing files during write completion

actiondfs: splice backing files without redundant credential swaps

actiondfs: omit file accounting for internal directory blobs

actiondfs: initialize staged inodes once after backing creation

actiondfs: trust validated immutable backing inode types

actiondfs: publish canonical cached directories without atomic exchange

actiondfs: match identical blob digest pointers before CAS roots

actiondfs: match identical directory digests before CAS roots

actiondfs: compress cached child records with name offsets

actiondfs: allocate directory iteration state only for staged entries

actiondfs: skip sorting directories with one validated child type

actiondfs: decode single-byte protobuf varints directly

actiondfs: avoid owner spinlocks when staged identities already match

actiondfs: assign immutable inode numbers only once

actiondfs: initialize only required protobuf child fields

actiondfs: accumulate packed filename bytes while parsing children

actiondfs: borrow guaranteed backing files on nonempty input reads

actiondfs: decode directory digests directly into cached children

actiondfs: publish immutable inode identity only in the hash initializer