Skip to content

Latest commit

 

History

History
65 lines (40 loc) · 2.22 KB

dev.md

File metadata and controls

65 lines (40 loc) · 2.22 KB
Raw

.dgit structure

index: point to blobs and path of file (or shortcut access), contains staging changes
objects: blobs, stored as hash, and cut head for faster lookup
branches: points to current commit HEAD: points to branches

cli design

git init creates a .dgit file (like .git) git add: stage changes: write blobs to disk -> update object tree -> write index file based on tree
todo: delete write blobs to disk step, as it confusing, to write_files of tree_dir.rs todo: does not track changes yet, so may add both file despite identical blob

git commit: commit changes: create entire new tree from index file -> write files -> write commit based on that tree -> clear index file
-> adjust branches to point to it write commit: write commit as a blob
write files: write content of staged files as blob

git log: keep tracing

Questions

more trees? AVL, red black

i dont know why i made a btree lol, maybe just separate reading and writing back

Todo:

  • diff: current commit tree, vs current working directory tree, compare of the 2, diferent hash means different. account for add

we can put it like path-hash map though, iterate once, and check in the other

git diff

branch: show branch branch : new branch checkout : switch branch (must commit)

Jon

Jon git

  • has decompress/encompress -> open file reader, not read string and pass whole string to compress

  • read blob header -> read it's size -> allocate exact buffer size -> read blob to buffer
    -> instead of allocating exact buffer, we can have a limit reader, which should not read more than X bytes.
    (guarded read)

  • not use println cuz it print string -> use stdout.write_all, in case it a binary (images) -> stills render out

  • hash writer: we need to write with zlib encoded and then hashed the thing -> Wrap the encoder+hasher with HashWriter, them impl Write trait for HashWriter -> allow using macro write!, which encode then hash when called

  • commit write whole tree instead of reading from index file???
    -> same as you recursively hash nested file first, then get the overall result and hash the outer

  • Object instead of separate?: this is good, for blob, tree, commit

anyhow, thiserror crate