Lesson 05 · Writing an object to the store

Writing an object to the store

Now we actually persist an object: compress its serialized bytes with zlib and write them to the fan-out path. This one function stores blobs today and every other object type later, so we make it generic from the start.

The goal

Compress and store a serialized object, returning its id.

Start here - the target

TO DO

Scenario: An object is compressed and written under its id

Givenan initialised repository and the content hello and a newline

WhenWriteObject(blob, content) stores it

Thenit returns ce013625030ba8dba906f756967f9e9ca394464a and a file exists at objects/ce/013625030ba8dba906f756967f9e9ca394464a

Andreading that file and zlib-inflating it yields exactly the serialized bytes blob 6 NUL hello newline, and writing the same content again is a harmless no-op that returns the same id

Background

Git compresses every loose object with zlib before writing it, so the file on disk is the deflated form of the serialized bytes. We store it at the fan-out path from the last lesson. Because the id is a hash of the content, an object is immutable: if a file already exists at that path it holds identical bytes, so re-writing is a no-op. That is why staging the same file twice costs nothing.

This is also where a small repository handle earns its place: a value holding the repository root so that reads and writes know where the object database is. Introduce it now (an Open(root) that returns it, say) and thread it through every store operation from here on. Notice too that we made WriteObject take a type rather than hard-coding blob. Trees and commits are stored by this exact same function later; only the type word and content differ. We use the standard library’s zlib here as plumbing, the same way we lean on SHA-1 - the lesson is the store, not the compression. Do not pin the exact compressed bytes: zlib’s output depends on the compression level, so we assert on the bytes we get back after inflating, which are always the serialized object.

Make it work

// deflate the serialized bytes with zlib, write to the fan-out path
func (r *Repo) WriteObject(typ string, content []byte) (string, error) {
  data := Serialize(typ, content)
  id := HashObject(typ, content)
  // zlib.NewWriter -> write data -> close; MkdirAll the dir; write file
  // skip the write if the file already exists (same id => same bytes)
  return id, nil
}

CheckpointDONE

You can store any object in the database and read its bytes back. Commit and stop here.