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

S&I's idea of EQ?

    Date: 12 Mar 1986  10:11 EST (Wed)
    From: Bill Rozas <JINX%OZ.AI.MIT.EDU at xx.lcs.mit.edu>

    If map is not known, the time to look its value up at runtime is
    comparable to the time which it takes to close the procedure.  In
    particular, in MIT-Scheme (because of first class environments, etc),
    looking up map can take considerably longer than closing the
    lambda-expression, and the latter time is usually negligible.  I think
    that the small time difference which can be gained in this case is not
    very interesting.

This is an empirical question for which I don't have any data, but my
intuition is that the way procedures are used and implemented in T, the
consing overhead here would be unacceptable, probably high enough to
make T want to be incompatible with Scheme in yet one more way, if
Scheme were changed.  The fact that the T implementation "coalesces"
procedures is exploited very heavily in the implementation and, I
suspect, in the way some users write code.  Not all Scheme
implementations have MIT scheme's high variable lookup overhead; consing
and space are not as cheap in most implementations as in MIT scheme; and
not all compilers or users choose to do as much analysis and procedure
integration as you believe is appropriate.