Re: Emacs Lisp JIT Compiler

[Paul Eggert]

Eli Zaretskii wrote:
> we should choose our JIT library so that it doesn't
> suffer from such issues

Yes, of course that would be preferable. However, performance variation is just 
one factor, and other factors in JIT libraries are just as important, if not 
more so. We shouldn't have this one factor overriding everything else.

Besides, none of the JIT approaches we're talking about are free from the issues 
Daniel raised: they all have significant performance differences on different 
platforms. If we're going to reject a JIT approach merely because a 3x 
performance improvement on one platform isn't equally shared by other platforms, 
we might as well just be up front about it and say we're going to reject all JIT 
approaches.

(If performance variation is really a problem, we could have Emacs default to 
running slowly on all platforms, and speed up only when the user asks it to. :-)