LismMachine File Reading.

[Keith Price <price%iris.usc.edu@USC.EDU>]

Just for information:
The manual does appear to have a bug, page 144 states clearly that
string-char is the default (depending on when the actual change was made in
the system the times may or may not reflect string-char or character).
The difference, today, in a similar test is roughly 10%, clearly not enough
to change the final conclusion.  (:element-type 'string-char is 10% faster
than :element-type 'character on the previously listed data).

File opening time of 2.8 seconds is irrelevant in a total time of 100+
seconds when we are comparing to other systems with times of 50%.  That is
why I used large files.

More interesting would be the effect of *read-suppress* since the files do
not contain anything that should be affected by it.  (This would be for all
systems.)  Unfortunately, it has other strange interactions and does not
work.

The overall factor of 2 for input is a minor issue in comparing systems (at
least for out applications) and the fact that repeated input is less required
on the Symbolics than a Unix version saves more time than it costs.  The
original complaint was that it was several orders of magnitude, considering
how much faster the hardware in a Sun-4 is, only a factor of 2 is good.
	Keith Price
	price@usc.edu