python: to_be_file

[nedtwigg]

This is different than to_match_disk. "MatchDisk" has a special file format, garbage-collection, lots of fancy features. to_be_file is only for a single binary facet, it writes the binary to disk unchanged.

The demo that this unlocks is AI image generation, where you can open the snapshot without any extra tooling. Jus ttsay to_be_file("test.jpg") and you can just open the file.

The beginning of it is here:

• selfie/jvm/selfie-lib/src/commonMain/kotlin/com/diffplug/selfie/guts/WriteTracker.kt

  Lines 106 to 144 in c757b1c

  	class ToBeFileWriteTracker : WriteTracker<TypedPath, ToBeFileLazyBytes>() {
  	fun writeToDisk(
  	key: TypedPath,
  	snapshot: ByteArray,
  	call: CallStack,
  	layout: SnapshotFileLayout
  	) {
  	val lazyBytes = ToBeFileLazyBytes(key, layout, snapshot)
  	recordInternal(key, lazyBytes, call, layout)
  	// recordInternal will throw an exception on a duplicate write, so we can safely write to disk
  	lazyBytes.writeToDisk()
  	// and because we are doing duplicate checks, `ToBeFileLazyBytes` can allow its in-memory
  	// data to be garbage collected, because it can safely read from disk in the future
  	}
  	}
  	class ToBeFileLazyBytes(val location: TypedPath, val layout: SnapshotFileLayout, data: ByteArray) {
  	/** When constructed, we always have the data. */
  	var data: ByteArray? = data
  	/**
  	* Shortly after being construted, this data is written to disk, and we can stop holding it in
  	* memory.
  	*/
  	internal fun writeToDisk() {
  	data?.let { layout.fs.fileWriteBinary(location, it) }
  	?: throw IllegalStateException("Data has already been written to disk!")
  	data = null
  	}
  	/**
  	* If we need to read our data, we do it from memory if it's still there, or from disk if it
  	* isn't.
  	*/
  	private fun readData(): ByteArray = data ?: layout.fs.fileReadBinary(location)
  	/** We calculate equality based on this data. */
  	override fun equals(other: Any?): Boolean =
  	if (this === other) true
  	else if (other is ToBeFileLazyBytes) readData().contentEquals(other.readData()) else false
  	override fun hashCode(): Int = readData().contentHashCode()
  	}

[nedtwigg]

• basic to_be_file implementation python to_be_file #366
• needs expect_selfie(actual: bytes) -> BinarySelfie
  • means BinaryFacet
• needs cache_selfie_binary

  • def cache_selfie_binary(to_cache: Cacheable[bytes]) -> CacheSelfieBinary[bytes]: ...
    def cache_selfie_binary(to_cache: Cacheable[T], roundtrip: CacheSelfie[T, bytes]) -> CacheSelfieBinary[T]: ...