GitLab

at 2023-08-21T08:10:45-07:00

Fix type-init problem and set min length to 150 for subsetp

I was wrong about `type-init` having hashtables available.  The issue
was `*min-list-length-for-hashtable*` not being defined during
`type-init`.  Also, some timing tests indicate that we need a list
length of about 150 for the hashtable implementation to be faster than
the list implementation.

So, add a new variable `*min-list-length-for-subsetp-hashtable*` with
a default value of 150.  A `cold-init-form` is added so it's available
during `type-init`.

The variable `*allow-hashtable-for-set-functions*` is removed because
we don't need it anymore.

Also, as a test, I set `*min-list-length-for-subsetp-hashtable*` to
10, and we were still able to build lisp without problems.

@@ -347,9 +347,6 @@
    #+gengc (setf conditions::*handler-clusters* nil)
    (setq intl::*default-domain* "cmucl")
    (setq intl::*locale* "C")
 -  ;; During init, we can't use hashtables to speed up the set
 -  ;; functions.  In particular, subsetp is used in type-init.
 -  (setq lisp::*allow-hashtable-for-set-functions* nil)
    ;; Many top-level forms call INFO, (SETF INFO).
    (print-and-call c::globaldb-init)

@@ -989,8 +989,6 @@
  	      (rplacd splicex (cdr x)))
  	  (setq splicex x)))))
 -(defvar *allow-hashtable-for-set-functions* t)
+-
  (declaim (start-block shorter-list-to-hashtable subsetp))
  (defun shorter-list-to-hashtable (list1 list2 key test test-not)
@@ -1013,7 +1011,7 @@
               (lst2 list2 (cdr lst2)))
              ((or (null lst1) (null lst2))
               (values len (if (null lst1) list1 list2))))
 -      (when (< min-length *min-list-length-for-hashtable*)
 +      (when (< min-length kernel::*min-list-length-for-subsetp-hashtable*)
          (return-from shorter-list-to-hashtable nil))
        (let ((hashtable (make-hash-table :test hash-test :size min-length)))
          (dolist (item shorter-list)
@@ -1031,7 +1029,7 @@
    ;; take care to disable this for the kernel.core.  SAVE will set
    ;; this to true when it's safe to use hash tables for SUBSETP.
    (multiple-value-bind (hashtable shorter-list)
 -      (when *allow-hashtable-for-set-functions*
 +      (when t
          (shorter-list-to-hashtable list1 list2 key test test-not))
      (cond (hashtable
             (cond ((eq shorter-list list1)

@@ -320,8 +320,6 @@
  	     (intl::setlocale)
  	     (ext::process-command-strings process-command-line)
  	     (setf *editor-lisp-p* nil)
 -	     ;; Allow using hashtables to speed up the set functions
 -	     (setf lisp::*allow-hashtable-for-set-functions* t)
  	     (macrolet ((find-switch (name)
  			  `(find ,name *command-line-switches*
  				 :key #'cmd-switch-name

@@ -377,6 +377,14 @@
  (cold-load-init (setq *use-implementation-types* t))
  (declaim (type boolean *use-implementation-types*))
 +(defvar *min-list-length-for-subsetp-hashtable* 150
 +  "The minimum length of either list argument for subsetp where a
 +  hashtable is used to speed up processing instead of using a basic
 +  list implementation.  This value was determined by experimentation.")
++
 +(cold-load-init (setq *min-list-length-for-subsetp-hashtable* 150))
 +(declaim (type fixnum *min-list-length-for-subsetp-hashtable*))
++
  ;;; DELEGATE-COMPLEX-{SUBTYPEP-ARG2,INTERSECTION}  --  Interface
  ;;;
  ;;;    These functions are used as method for types which need a complex

Raymond Toy pushed to branch issue-240-subsetp-with-hash-table at cmucl / cmucl

Commits:

4 changed files:

Changes: