[Tinymux] TinyMUX 2.4.0.19 Alpha

[Stephen Dennis]

I could also detect 3DNow, EMT64, etc., but since the Intel compiler
does not yet produce any specialized code paths for those extensions,
once detected, I wouldn't be able to map them to anything.

I wrote one MMX instruction, one SSE instruction, etc. This was about
coaxing the Intel run-time to use code paths it already had when an AMD
chip is present.

Sometimes, we inadvertently break it, but the code is already 64-bit
clean.

Areas of future interest:

 - The only way to fix 'sub(0.1,0.09)' --> '0.01000000000000001'
   without sacrificing accuracy generally is to stop using base-2
   floating point. I'm not sure whether I should wait on base-10
   floating-point hardware to hit (the standards for it are forming
   now), or use an existing package.

 - SQL connector.

 - Implement proper telnet protocol.

 - Make a C++ version of an LBUF and separate the color information.
   The 'ESC[' sequences would be generated at the network. At the
   moment, NOANSI strips these sequences out. In the future, NOANSI
   would prevent them from being generated. I'm not yet sure how this
   is represented in the database, yet.

 - Unicode would be nice. It's hard and nearly every line of code would
   be affected, but I'd love to see a Chinese-language MU. This would
   require a lot of dedicated work to pull off properly.

 - There is still a long list of Work Items and bug fixes. For example,
   @chownall behaves unexpected when used on a non-player object. It
   works on the object's owner to divest them of all their stuff
   instead of refusing to do anything. Things like that are always
   being reported.

 - Typed LBUFS. If I'm already passing around C++ LBUFS, then it isn't
   that much of a stretch for them to be VARIANTS. That is, add()
   produces a number, and internally, there is a string form, and a
   floating-point form, and an integer form. The conversions become
   lazy. 'mul(add())' would sees add() pass the floating-point or
   integer form to mul() without doing converting to and from a string.


The most reasonable order is: SQL, Telnet, C++ LBUF, Unicode, and then
typed LBUFS. I think if I can bury all ANSI/String stuff into a C++
class, the Unicode changes would be primarily changes to that class.

Honestly? Think 'years' not 'months'.


Brazil