Looking at the first few bytes of the archive will help us because I plan to move the metadata to the back of the archive file, and it would be expensive to seek to the end of the file just to determine the file type. I figure a 64-byte header is enough to hold some extra data that identifies the file type and points the unstarch and starchcat tools to the right byte in the file where the metadata is kept:
This image describes what I’m planning for Starch v2. Basically, we reserve a 64-byte header at the front of the archive that contains the following four items:
After these 64 bytes, the compressed, per-chromosome streams start, and we then wrap up with the metadata at the end of the file.
Initially, the header will be the magic number and all zeros. At the conclusion of creating the archive, once all the streams are prepared and the metadata is ready to hash, parts of the header are written over with calculated values (offset and hash).
To assist with picking the right magic number, Ned Batchelder put together a Python script to generate all the hex words possible from a small subset of the English dictionary — words like deadbeef, dec0ded and 0ddba11 which can be used as magic numbers to identify a file type. Thanks, Ned! So far, I’m liking the magic word ca5cade5 as it has a nice biological flavor to it.