syncany-team team mailing list archive

[Philipp Heckel <philipp.heckel@xxxxxxxxx>]

Hi Fabrice,

> The ChunkEntry is only the database representation, right?

Yes, ChunkEntry is only database representation, so the actual data is
missing, unlike the object in the "chunk" package.

>
> Hum, but if the ChunkEntry and the Chunk both contains the checksum
> which is the Id of the Chunk, why do we need a ChunkEntryId is the first
> place? 

The ChunkEntryId only exists because this is needed for the "quick
access" caches in the Database and DatabaseVersion classes. We need a
ChunkEntryId, because "byte[]" cannot be used as key in a Map:

    // Quick access cache
    private Map<ChunkEntryId, MultiChunkEntry> chunkMultiChunkCache; 

> Or to put it differently, why not replacing the byte[] in
> ChunkEntry and Chunk by a ChunkId?

That I don't really know. We could do that, but only if it's clear from
the code that for ChunkEntry objects the identifier is a checksum. Maybe
the Id for chunks should be named ChunkChecksum to make this clearer
(and the ID scheme more complex :-)).

ChunkEntry and Chunk should not have the same IDs because they are in
different packages. These packages should not cross-reference each other.

> I'm asking because if you want to keep getChunkInputStream(byte[])
> (which totally makes sense), then the ChunkEntryId must leek its
> internal representation (as a byte[]). So either it's a getCheckSum
> specific to ChunkEntryId, so far so good, or it's a general method of
> ObjectId. I would refrain from using the latter solution, because it
> will be inefficient for tight memory representation of short ids. Not a
> big deal, again. Also if you go for Gregor's solution (or a variant of
> it), it would mean putting getByteArray in the ObjectId which again is
> not so beautiful.

I don't see any reason for the "org.syncany.chunk.*" classes to
introduce an ID concept, so having a getChunkInputString(ObjectId)
method would not be sensible. Following the argumentation of above, it
is hence necessary to leak the internal representation to call the
getChunkInputString(byte[]) method.

Hope this makes sense :-)

Best
Philipp