On Tuesday 01 January 2002 01:53 pm, Dan Sugalski wrote: > At 08:46 AM 1/1/2002 -1000, David & Lisa Jacobs wrote: > >Actually, I started down this road as well, but thought it would make > > more sense to make the opcode_t unsigned. This would allow us to keep > > it a consistent size across platforms if we want. > > I'd rather opcode_t stays signed--that gives us two magic returns, 0 and > -1, that we can cheaply check for to signal errors or something. > > I think a variant on this patch would be in order.
This only really causes heartache on CRI, which would have to manually sign-extend a negative 32-bit-packed opcode_t. (Which may or may not be a valid value within a packfile. That's a separate issue.) It could also cause potential problems for opcode dispatch that is table based. (Of course, having any number with bit 32 set, signed or not, will cause problems with dispatch that is table based. :-) Given it's current use, I'd expect that the tests would have to be for either 0 or for any negative number. Either way, it's a two operation test. -- Bryan C. Warnock [EMAIL PROTECTED]
