Re: binary interface

[Keisuke Nishida]

At Thu, 01 Mar 2001 23:37:54 -0500,
Keisuke Nishida wrote:
> 
>   Text format:         (1 "foo")            9 bytes
>   Semi-binary format:  (1 s3 foo)          10 bytes
>   Binary format:       [cons][1][#1][cons][#2][()][string][foo\0]  32 bytes
> 
> The semi-binary format is much shorter than the binary format,
> while being very friendly to the reader.  If I/O is much slower
> than CPU, the semi-binary format above should be better than
> the current binary format.  Should we try it?

I guess the current binary format is just bad - too space consuming.
The above semi-binary format is not good enough, either.  We could
implement a more sophisticated and optimized loader, using the same
technique as my VM.  Actually, I could integrate a loading facility
into my VM.  A binary file would be a single bytecode which constructs
objects.  Oh, gee, I didn't need to write a loader at all...