Trying to understand why LZ77 decode is usually sequential

Hi,

I’ve been playing around with an LZ77-like idea and ran into something I don’t fully understand.

I managed to get around ~10 GB/s decode (bit-perfect) on a ~1GB file, while still keeping a global compression context.

What I did (roughly):

• matches are found globally (not per block)
• but I split the output into blocks
• and store offsets so each block knows where its data is (literals / offsets / lengths)

So blocks are not compressed independently, but I try to make them independently decodable.

This seems to allow parallel decode, which is where the speed comes from.

I feel like I’m probably missing something obvious, because I don’t really see this approach used.

So I’m wondering:

• why is LZ77 decode usually kept strictly sequential?
• is this kind of approach already known and I just don’t know about it?
• what would likely break with this in a real implementation?

Happy to share more details if needed.

For reference: the four streams (literals / offsets / lengths / commands) are entropy-coded as single global zstd frames, not per-block.

Thanks