Re: [PATCH] Font antialiasing v2

[Szymon Janc]

On Mon, 12 Apr 2010 07:24:09 -0700 Colin D Bennett <address@hidden> wrote

> > This issue should be handled at compress time by choosing to compress
> > by blocks of desired size. This way font layer doesn't need to care
> > anymore.
> 
> But can you randomly seek to an block transparently and read it that
> way?

Maybe not fully randomly, but jumping by blocks will be quite fast even if file
was seeked backward when one must start from begining of file(at least with
xz). I don't know if jumping block backward is available in gzip, probably not.

If block size would be chosen wisely it should be possible to buffer block
(uncompressed) data and there should be quite a big chance that data locality
will minimize need for jumping (and decompression). Everything fully
transparent for caller. If You add compressed data buffering (sth like
compressed_data->bufio->decompressor->decompressed_block_buffer->caller) this
should work quite nicely.

As I've mentioned in my post about xz compression, I'll try to propose some
sollution fot that in the near future.


-- 
Szymon Janc