Re: [MIT-Scheme-devel] For your amusement

[Taylor R Campbell]

   Date: Fri, 17 Jun 2011 10:34:21 -0700
   From: Joe Marshall <address@hidden>

   I remember the main reason I didn't want to make this the default.
   I had wanted to use the Y operator to find the fixed-point and I
   expected Liar to `tame' it.  Unfortunately, Liar isn't compiling the
   Y operator correctly, so I tried self-application instead.

What goes wrong?  I presume it's a different problem from the obscure
closure analysis bug I fixed a year and a half ago?

   On Wed, Jun 15, 2011 at 5:01 PM, Taylor R Campbell <address@hidden> wrote:
   > Extra time spent in flow analysis and closure analysis?  (Perhaps
   > another reason to avoid using it by default is the scary comment in
   > compiler/fgopt/closan.scm...)  Or are you noticing time spent running
   > particular code that uses named LET, not building the whole system
   > overall?

   Not sure.  I'm going to try measuring this more seriously to find out
   what the problem is.

A quick start might be to run

cd mit-scheme/src && scheme --batch-mode <<EOF
(begin
  (load "etc/compile")
  (fluid-let ((named-let-strategy X))
    (with-stack-sampling 10 compile-everything)))
EOF

for each strategy X and compare the output.  (Even if that doesn't
explain the difference, it might reveal bottlenecks in the build
anyway...)