Lesson 17 · Persisting the index

Persisting the index

The index has to survive between commands, so Git writes it to a file. Real Git uses a packed binary format; we use a simple line-based one that is easy to read and round-trips cleanly. Today you save and load the index.

The goal

Write the index to disk and read it back into an identical index.

Start here - the target

TO DO

Scenario: The index round-trips through a file

Givenan index holding hello.txt (100644, ce013625030ba8dba906f756967f9e9ca394464a) and README.md (100644, 0805455a24b6c68fbc38d0fa5d121f735984285d)

Whenthe index is written to .mygit/index as one line per entry, mode space id space path, sorted by path, then loaded back

Thenthe file lines are 100644 0805455a24b6c68fbc38d0fa5d121f735984285d README.md then 100644 ce013625030ba8dba906f756967f9e9ca394464a hello.txt

Andthe loaded index has the same two entries as the original

Background

A staging area is only useful if it persists: you run add today, commit tomorrow, and the index must still be there. Git serializes it to .mygit/index. Real Git uses a compact binary format with cached stat data for speed; that format is documented but fiddly, and it is not what this project is about, so we use a plain line-based format instead: one line per entry with the mode, id, and path, sorted by path.

This is the one place we knowingly diverge from real Git’s bytes. It is a safe divergence: the index is a private staging structure, never hashed into an object, so write-tree and every id downstream still match Git exactly. What must round-trip is the information (each path’s mode and id), not the exact file layout. Save it, load it, and confirm you get the same entries back.

Make it work

// one line per entry: "<mode> <id> <path>\n", sorted by path
for _, p := range ix.Paths() {
  e := ix.entries[p]
  fmt.Fprintf(w, "%s %s %s\n", e.Mode, e.Id, p)
}
// load: split each line into mode, id, path and Set it

CheckpointDONE

The index survives across commands. Commit and stop here.