[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Robert S. Boyer <boyer@CLI.COM>: Fast IO]



Symbolics Genera is even slower at read-char.  But I don't think a
declaration is the answer.  I believe that if we wanted to make
read-char fast on streams where it can be fast (simple buffered
streams), a simple-stream declaration would not add any significant
additional speed.  It might save about 2 microseconds for an indirect
call.  This is assuming that compatibility with past implementations
of streams was not an issue (and I don't think it's reasonable in
designing a new language to assume that such would be an issue).

I know that Symbolics Genera read-char is slow because we haven't
made any effort to make it fast, not because non-simple streams make
it impossible to make it fast.  I can't speak for Lucid but I wouldn't
be surprised if the story is the same for them.  So I'd oppose adding
anything to the language in this area.  Instead, customers should signal
to implementors their discontent with the current quality of implementation
of read-char.  Implementors can then raise the priority of that and
possibly shift resources from other things.