view lib-src/qsort.c @ 5353:38e24b8be4ea

Improve the lexical scoping in #'block, #'return-from. lisp/ChangeLog addition: 2011-02-07 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el: * bytecomp.el (byte-compile-initial-macro-environment): Shadow `block', `return-from' here, we implement them differently when byte-compiling. * bytecomp.el (byte-compile-active-blocks): New. * bytecomp.el (byte-compile-block-1): New. * bytecomp.el (byte-compile-return-from-1): New. * bytecomp.el (return-from-1): New. * bytecomp.el (block-1): New. These are two aliases that exist to have their own associated byte-compile functions, which functions implement `block' and `return-from'. * cl-extra.el (cl-macroexpand-all): Fix a bug here when macros in the environment have been compiled. * cl-macs.el (block): * cl-macs.el (return): * cl-macs.el (return-from): Be more careful about lexical scope in these macros. * cl.el: * cl.el ('cl-block-wrapper): Removed. * cl.el ('cl-block-throw): Removed. These aren't needed in code generated by this XEmacs. They shouldn't be needed in code generated by XEmacs 21.4, but if it turns out the packages do need them, we can put them back. 2011-01-30 Mike Sperber <mike@xemacs.org> * font-lock.el (font-lock-fontify-pending-extents): Don't fail if `font-lock-mode' is unset, which can happen in the middle of `revert-buffer'. 2011-01-23 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (delete): * cl-macs.el (delq): * cl-macs.el (remove): * cl-macs.el (remq): Don't use the compiler macro if these functions were given the wrong number of arguments, as happens in lisp-tests.el. * cl-seq.el (remove, remq): Removed. I added these to subr.el, and forgot to remove them from here. 2011-01-22 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el (byte-compile-setq, byte-compile-set): Remove kludge allowing keywords' values to be set, all the code that does that is gone. * cl-compat.el (elt-satisfies-test-p): * faces.el (set-face-parent): * faces.el (face-doc-string): * gtk-font-menu.el: * gtk-font-menu.el (gtk-reset-device-font-menus): * msw-font-menu.el: * msw-font-menu.el (mswindows-reset-device-font-menus): * package-get.el (package-get-installedp): * select.el (select-convert-from-image-data): * sound.el: * sound.el (load-sound-file): * x-font-menu.el (x-reset-device-font-menus-core): Don't quote keywords, they're self-quoting, and the win from backward-compatibility is sufficiently small now that the style problem overrides it. 2011-01-22 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (block, return-from): Require that NAME be a symbol in these macros, as always documented in the #'block docstring and as required by Common Lisp. * descr-text.el (unidata-initialize-unihan-database): Correct the use of non-symbols in #'block and #'return-from in this function. 2011-01-15 Aidan Kehoe <kehoea@parhasard.net> * cl-extra.el (concatenate): Accept more complicated TYPEs in this function, handing the sequences over to #'coerce if we don't understand them here. * cl-macs.el (inline): Don't proclaim #'concatenate as inline, its compiler macro is more useful than doing that. 2011-01-11 Aidan Kehoe <kehoea@parhasard.net> * subr.el (delete, delq, remove, remq): Move #'remove, #'remq here, they don't belong in cl-seq.el; move #'delete, #'delq here from fns.c, implement them in terms of #'delete*, allowing support for sequences generally. * update-elc.el (do-autoload-commands): Use #'delete*, not #'delq here, now the latter's no longer dumped. * cl-macs.el (delete, delq): Add compiler macros transforming #'delete and #'delq to #'delete* calls. 2011-01-10 Aidan Kehoe <kehoea@parhasard.net> * dialog.el (make-dialog-box): Correct a misplaced parenthesis here, thank you Mats Lidell in 87zkr9gqrh.fsf@mail.contactor.se ! 2011-01-02 Aidan Kehoe <kehoea@parhasard.net> * dialog.el (make-dialog-box): * list-mode.el (display-completion-list): These functions used to use cl-parsing-keywords; change them to use defun* instead, fixing the build. (Not sure what led to me not including this change in d1b17a33450b!) 2011-01-02 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (define-star-compiler-macros): Make sure the form has ITEM and LIST specified before attempting to change to calls with explicit tests; necessary for some tests in lisp-tests.el to compile correctly. (stable-union, stable-intersection): Add compiler macros for these functions, in the same way we do for most of the other functions in cl-seq.el. 2011-01-01 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (dolist, dotimes, do-symbols, macrolet) (symbol-macrolet): Define these macros with defmacro* instead of parsing the argument list by hand, for the sake of style and readability; use backquote where appropriate, instead of calling #'list and and friends, for the same reason. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * x-misc.el (device-x-display): Provide this function, documented in the Lispref for years, but not existing previously. Thank you Julian Bradfield, thank you Jeff Mincy. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * cl-seq.el: Move the heavy lifting from this file to C. Dump the cl-parsing-keywords macro, but don't use defun* for the functions we define that do take keywords, dynamic scope lossage makes that not practical. * subr.el (sort, fillarray): Move these aliases here. (map-plist): #'nsublis is now built-in, but at this point #'eql isn't necessarily available as a test; use #'eq. * obsolete.el (cl-delete-duplicates): Make this available for old compiler macros and old code. (memql): Document that this is equivalent to #'member*, and worse. * cl.el (adjoin, subst): Removed. These are in C. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * simple.el (assoc-ignore-case): Remove a duplicate definition of this function (it's already in subr.el). * iso8859-1.el (char-width): On non-Mule, make this function equivalent to that produced by (constantly 1), but preserve its docstring. * subr.el (subst-char-in-string): Define this in terms of #'substitute, #'nsubstitute. (string-width): Define this using #'reduce and #'char-width. (char-width): Give this a simpler definition, it makes far more sense to check for mule at load time and redefine, as we do in iso8859-1.el. (store-substring): Implement this in terms of #'replace, now #'replace is cheap. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * update-elc.el (lisp-files-needed-for-byte-compilation) (lisp-files-needing-early-byte-compilation): cl-macs belongs in the former, not the latter, it is as fundamental as bytecomp.el. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * cl.el: Provde the Common Lisp program-error, type-error as error symbols. This doesn't nearly go far enough for anyone using the Common Lisp errors. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (delete-duplicates): If the form has an incorrect number of arguments, don't attempt a compiler macroexpansion. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (cl-safe-expr-p): Forms that start with the symbol lambda are also safe. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (= < > <= >=): For these functions' compiler macros, the optimisation is safe even if the first and the last arguments have side effects, since they're only used the once. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (inline-side-effect-free-compiler-macros): Unroll a loop here at macro-expansion time, so these compiler macros are compiled. Use #'eql instead of #'eq in a couple of places for better style. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-extra.el (notany, notevery): Avoid some dynamic scope stupidity with local variable names in these functions, when they weren't prefixed with cl-; go into some more detail in the doc strings. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * byte-optimize.el (side-effect-free-fns): #'remove, #'remq are free of side-effects. (side-effect-and-error-free-fns): Drop dot, dot-marker from the list. 2010-11-17 Aidan Kehoe <kehoea@parhasard.net> * cl-extra.el (coerce): In the argument list, name the first argument OBJECT, not X; the former name was always used in the doc string and is clearer. Handle vector type specifications which include the length of the target sequence, error if there's a mismatch. * cl-macs.el (cl-make-type-test): Handle type specifications starting with the symbol 'eql. 2010-11-14 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (eql): Don't remove the byte-compile property of this symbol. That was necessary to override a bug in bytecomp.el where #'eql was confused with #'eq, which bug we no longer have. If neither expression is constant, don't attempt to handle the expression in this compiler macro, leave it to byte-compile-eql, which produces better code anyway. * bytecomp.el (eq): #'eql is not the function associated with the byte-eq byte code. (byte-compile-eql): Add an explicit compile method for this function, for cases where the cl-macs compiler macro hasn't reduced it to #'eq or #'equal. 2010-10-25 Aidan Kehoe <kehoea@parhasard.net> Add compiler macros and compilation sanity-checking for various functions that take keywords. * byte-optimize.el (side-effect-free-fns): #'symbol-value is side-effect free and not error free. * bytecomp.el (byte-compile-normal-call): Check keyword argument lists for sanity; store information about the positions where keyword arguments start using the new byte-compile-keyword-start property. * cl-macs.el (cl-const-expr-val): Take a new optional argument, cl-not-constant, defaulting to nil, in this function; return it if the expression is not constant. (cl-non-fixnum-number-p): Make this into a separate function, we want to pass it to #'every. (eql): Use it. (define-star-compiler-macros): Use the same code to generate the member*, assoc* and rassoc* compiler macros; special-case some code in #'add-to-list in subr.el. (remove, remq): Add compiler macros for these two functions, in preparation for #'remove being in C. (define-foo-if-compiler-macros): Transform (remove-if-not ...) calls to (remove ... :if-not) at compile time, which will be a real win once the latter is in C. (define-substitute-if-compiler-macros) (define-subst-if-compiler-macros): Similarly for these functions. (delete-duplicates): Change this compiler macro to use #'plists-equal; if we don't have information about the type of SEQUENCE at compile time, don't bother attempting to inline the call, the function will be in C soon enough. (equalp): Remove an old commented-out compiler macro for this, if we want to see it it's in version control. (subst-char-in-string): Transform this to a call to nsubstitute or nsubstitute, if that is appropriate. * cl.el (ldiff): Don't call setf here, this makes for a load-time dependency problem in cl-macs.el 2010-06-14 Stephen J. Turnbull <stephen@xemacs.org> * term/vt100.el: Refer to XEmacs, not GNU Emacs, in permissions. * term/bg-mouse.el: * term/sup-mouse.el: Put copyright notice in canonical "Copyright DATE AUTHOR" form. Refer to XEmacs, not GNU Emacs, in permissions. * site-load.el: Add permission boilerplate. * mule/canna-leim.el: * alist.el: Refer to XEmacs, not APEL/this program, in permissions. * mule/canna-leim.el: Remove my copyright, I've assigned it to the FSF. 2010-06-14 Stephen J. Turnbull <stephen@xemacs.org> * gtk.el: * gtk-widget-accessors.el: * gtk-package.el: * gtk-marshal.el: * gtk-compose.el: * gnome.el: Add copyright notice based on internal evidence. 2010-06-14 Stephen J. Turnbull <stephen@xemacs.org> * easymenu.el: Add reference to COPYING to permission notice. * gutter.el: * gutter-items.el: * menubar-items.el: Fix typo "Xmacs" in permissions notice. 2010-06-14 Stephen J. Turnbull <stephen@xemacs.org> * auto-save.el: * font.el: * fontconfig.el: * mule/kinsoku.el: Add "part of XEmacs" text to permission notice. 2010-10-14 Aidan Kehoe <kehoea@parhasard.net> * byte-optimize.el (side-effect-free-fns): * cl-macs.el (remf, getf): * cl-extra.el (tailp, cl-set-getf, cl-do-remf): * cl.el (ldiff, endp): Tighten up Common Lisp compatibility for #'ldiff, #'endp, #'tailp; add circularity checking for the first two. #'cl-set-getf and #'cl-do-remf were Lisp implementations of #'plist-put and #'plist-remprop; change the names to aliases, changes the macros that use them to using #'plist-put and #'plist-remprop directly. 2010-10-12 Aidan Kehoe <kehoea@parhasard.net> * abbrev.el (fundamental-mode-abbrev-table, global-abbrev-table): Create both these abbrev tables using the usual #'define-abbrev-table calls, rather than attempting to special-case them. * cl-extra.el: Force cl-macs to be loaded here, if cl-extra.el is being loaded interpreted. Previously other, later files would redundantly call (load "cl-macs") when interpreted, it's more reasonable to do it here, once. * cmdloop.el (read-quoted-char-radix): Use defcustom here, we don't have any dump-order dependencies that would prevent that. * custom.el (eval-when-compile): Don't load cl-macs when interpreted or when byte-compiling, rely on cl-extra.el in the former case and the appropriate entry in bytecomp-load-hook in the latter. Get rid of custom-declare-variable-list, we have no dump-time dependencies that would require it. * faces.el (eval-when-compile): Don't load cl-macs when interpreted or when byte-compiling. * packages.el: Remove some inaccurate comments. * post-gc.el (cleanup-simple-finalizers): Use #'delete-if-not here, now the order of preloaded-file-list has been changed to make it available. * subr.el (custom-declare-variable-list): Remove. No need for it. Also remove a stub define-abbrev-table from this file, given the current order of preloaded-file-list there's no need for it. 2010-10-10 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el (byte-compile-constp) Forms quoted with FUNCTION are also constant. (byte-compile-initial-macro-environment): In #'the, if FORM is constant and does not match TYPE, warn at byte-compile time. 2010-10-10 Aidan Kehoe <kehoea@parhasard.net> * backquote.el (bq-vector-contents, bq-list*): Remove; the former is equivalent to (append VECTOR nil), the latter to (list* ...). (bq-process-2): Use (append VECTOR nil) instead of using #'bq-vector-contents to convert to a list. (bq-process-1): Now we use list* instead of bq-list * subr.el (list*): Moved from cl.el, since it is now required to be available the first time a backquoted form is encountered. * cl.el (list*): Move to subr.el. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * test-harness.el (Check-Message): Add an omitted comma here, thank you the buildbot. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * hash-table.el (hash-table-key-list, hash-table-value-list) (hash-table-key-value-alist, hash-table-key-value-plist): Remove some useless #'nreverse calls in these files; our hash tables have no order, it's not helpful to pretend they do. * behavior.el (read-behavior): Do the same in this file, in some code evidently copied from hash-table.el. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * info.el (Info-insert-dir): * format.el (format-deannotate-region): * files.el (cd, save-buffers-kill-emacs): Use #'some, #'every and related functions for applying boolean operations to lists, instead of rolling our own ones that cons and don't short-circuit. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el (byte-compile-initial-macro-environment): * cl-macs.el (the): Rephrase the docstring, make its implementation when compiling files a little nicer. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * descr-text.el (unidata-initialize-unicodedata-database) (unidata-initialize-unihan-database, describe-char-unicode-data) (describe-char-unicode-data): Wrap calls to the database functions with (with-fboundp ...), avoiding byte compile warnings on builds without support for the database functions. (describe-char): (reduce #'max ...), not (apply #'max ...), no need to cons needlessly. (describe-char): Remove a redundant lambda wrapping #'extent-properties. (describe-char-unicode-data): Call #'nsubst when replacing "" with nil in the result of #'split-string, instead of consing inside mapcar. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * x-faces.el (x-available-font-sizes): * specifier.el (let-specifier): * package-ui.el (pui-add-required-packages): * msw-faces.el (mswindows-available-font-sizes): * modeline.el (modeline-minor-mode-menu): * minibuf.el (minibuf-directory-files): Replace the O2N (delq nil (mapcar (lambda (W) (and X Y)) Z)) with the ON (mapcan (lambda (W) (and X (list Y))) Z) in these files. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (= < > <= >=): When these functions are handed more than two arguments, and those arguments have no side effects, transform to a series of two argument calls, avoiding funcall in the byte-compiled code. * mule/mule-cmds.el (finish-set-language-environment): Take advantage of this change in a function called 256 times at startup. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el (byte-compile-function-form, byte-compile-quote) (byte-compile-quote-form): Warn at compile time, and error at runtime, if a (quote ...) or a (function ...) form attempts to quote more than one object. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * byte-optimize.el (byte-optimize-apply): Transform (apply 'nconc (mapcar ...)) to (mapcan ...); warn about use of the first idiom. * update-elc.el (do-autoload-commands): * packages.el (packages-find-package-library-path): * frame.el (frame-list): * extents.el (extent-descendants): * etags.el (buffer-tag-table-files): * dumped-lisp.el (preloaded-file-list): * device.el (device-list): * bytecomp-runtime.el (proclaim-inline, proclaim-notinline) Use #'mapcan, not (apply #'nconc (mapcar ...) in all these files. * bytecomp-runtime.el (eval-when-compile, eval-and-compile): In passing, mention that these macros also evaluate the body when interpreted. tests/ChangeLog addition: 2011-02-07 Aidan Kehoe <kehoea@parhasard.net> * automated/lisp-tests.el: Test lexical scope for `block', `return-from'; add a Known-Bug-Expect-Failure for a contorted example that fails when byte-compiled.
author Aidan Kehoe <kehoea@parhasard.net>
date Mon, 07 Feb 2011 12:01:24 +0000
parents 576fb035e263
children 061f4f90f874
line wrap: on
line source

/* Plug-compatible replacement for UNIX qsort.
   Copyright (C) 1989 Free Software Foundation, Inc.
   Written by Douglas C. Schmidt (schmidt@ics.uci.edu)

This file is part of GNU CC.

GNU QSORT is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2, or (at your option)
any later version.

GNU QSORT is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with GNU QSORT; see the file COPYING.  If not, write to
the Free the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
Boston, MA 02111-1307, USA. */

/* Synched up with: FSF 19.28. */

#ifdef sparc
#include <alloca.h>
#endif

/* Invoke the comparison function, returns either 0, < 0, or > 0. */
#define CMP(A,B) ((*cmp)((A),(B)))

/* Byte-wise swap two items of size SIZE. */
#define SWAP(A,B,SIZE) do {int sz = (SIZE); char *a = (A); char *b = (B); \
    do { char _temp = *a;*a++ = *b;*b++ = _temp;} while (--sz);} while (0)

/* Copy SIZE bytes from item B to item A. */
#define COPY(A,B,SIZE) {int sz = (SIZE); do { *(A)++ = *(B)++; } while (--sz); }

/* This should be replaced by a standard ANSI macro. */
#define BYTES_PER_WORD 8

/* The next 4 #defines implement a very fast in-line stack abstraction. */
#define STACK_SIZE (BYTES_PER_WORD * sizeof (long))
#define PUSH(LOW,HIGH) do {top->lo = LOW;top++->hi = HIGH;} while (0)
#define POP(LOW,HIGH)  do {LOW = (--top)->lo;HIGH = top->hi;} while (0)
#define STACK_NOT_EMPTY (stack < top)                

/* Discontinue quicksort algorithm when partition gets below this size.
   This particular magic number was chosen to work best on a Sun 4/260. */
#define MAX_THRESH 4

/* Stack node declarations used to store unfulfilled partition obligations. */
typedef struct 
{
  char *lo;
  char *hi;
} stack_node;

/* Order size using quicksort.  This implementation incorporates
   four optimizations discussed in Sedgewick:
   
   1. Non-recursive, using an explicit stack of pointer that store the 
      next array partition to sort.  To save time, this maximum amount 
      of space required to store an array of MAX_INT is allocated on the 
      stack.  Assuming a 32-bit integer, this needs only 32 * 
      sizeof (stack_node) == 136 bits.  Pretty cheap, actually.

   2. Choose the pivot element using a median-of-three decision tree.
      This reduces the probability of selecting a bad pivot value and 
      eliminates certain extraneous comparisons.

   3. Only quicksorts TOTAL_ELEMS / MAX_THRESH partitions, leaving
      insertion sort to order the MAX_THRESH items within each partition.  
      This is a big win, since insertion sort is faster for small, mostly
      sorted array segments.
   
   4. The larger of the two sub-partitions is always pushed onto the
      stack first, with the algorithm then concentrating on the
      smaller partition.  This *guarantees* no more than log (n)
      stack size is needed (actually O(1) in this case)! */
      
int 
qsort (base_ptr, total_elems, size, cmp)
     char *base_ptr;
     int total_elems;
     int size;
     int (*cmp)();
{
  /* Allocating SIZE bytes for a pivot buffer facilitates a better 
     algorithm below since we can do comparisons directly on the pivot. */
  char *pivot_buffer = (char *) alloca (size);
  int   max_thresh   = MAX_THRESH * size;

  if (total_elems > MAX_THRESH)
    {
      char       *lo = base_ptr;
      char       *hi = lo + size * (total_elems - 1);
      stack_node stack[STACK_SIZE]; /* Largest size needed for 32-bit int!!! */
      stack_node *top = stack + 1;

      while (STACK_NOT_EMPTY)
        {
          char *left_ptr;
          char *right_ptr;
          {
            char *pivot = pivot_buffer;
            {
              /* Select median value from among LO, MID, and HI. Rearrange
                 LO and HI so the three values are sorted. This lowers the 
                 probability of picking a pathological pivot value and 
                 skips a comparison for both the LEFT_PTR and RIGHT_PTR. */

              char *mid = lo + size * ((hi - lo) / size >> 1);

              if (CMP (mid, lo) < 0)
                SWAP (mid, lo, size);
              if (CMP (hi, mid) < 0)
                SWAP (mid, hi, size);
              else 
                goto jump_over;
              if (CMP (mid, lo) < 0)
                SWAP (mid, lo, size);
            jump_over:
              COPY (pivot, mid, size);
              pivot = pivot_buffer;
            }
            left_ptr  = lo + size;
            right_ptr = hi - size; 

            /* Here's the famous ``collapse the walls'' section of quicksort.  
               Gotta like those tight inner loops!  They are the main reason 
               that this algorithm runs much faster than others. */
            do 
              {
                while (CMP (left_ptr, pivot) < 0)
                  left_ptr += size;

                while (CMP (pivot, right_ptr) < 0)
                  right_ptr -= size;

                if (left_ptr < right_ptr) 
                  {
                    SWAP (left_ptr, right_ptr, size);
                    left_ptr += size;
                    right_ptr -= size;
                  }
                else if (left_ptr == right_ptr) 
                  {
                    left_ptr += size;
                    right_ptr -= size;
                    break;
                  }
              } 
            while (left_ptr <= right_ptr);

          }

          /* Set up pointers for next iteration.  First determine whether
             left and right partitions are below the threshold size. If so, 
             ignore one or both.  Otherwise, push the larger partition's
             bounds on the stack and continue sorting the smaller one. */

          if ((right_ptr - lo) <= max_thresh)
            {
              if ((hi - left_ptr) <= max_thresh) /* Ignore both small partitions. */
                POP (lo, hi); 
              else              /* Ignore small left partition. */  
                lo = left_ptr;
            }
          else if ((hi - left_ptr) <= max_thresh) /* Ignore small right partition. */
            hi = right_ptr;
          else if ((right_ptr - lo) > (hi - left_ptr)) /* Push larger left partition indices. */
            {                   
              PUSH (lo, right_ptr);
              lo = left_ptr;
            }
          else                  /* Push larger right partition indices. */
            {                   
              PUSH (left_ptr, hi);
              hi = right_ptr;
            }
        }
    }

  /* Once the BASE_PTR array is partially sorted by quicksort the rest
     is completely sorted using insertion sort, since this is efficient 
     for partitions below MAX_THRESH size. BASE_PTR points to the beginning 
     of the array to sort, and END_PTR points at the very last element in
     the array (*not* one beyond it!). */

#define MIN(X,Y) ((X) < (Y) ? (X) : (Y))

  {
    char *end_ptr = base_ptr + size * (total_elems - 1);
    char *run_ptr;
    char *tmp_ptr = base_ptr;
    char *thresh  = MIN (end_ptr, base_ptr + max_thresh);

    /* Find smallest element in first threshold and place it at the
       array's beginning.  This is the smallest array element,
       and the operation speeds up insertion sort's inner loop. */

    for (run_ptr = tmp_ptr + size; run_ptr <= thresh; run_ptr += size)
      if (CMP (run_ptr, tmp_ptr) < 0)
        tmp_ptr = run_ptr;

    if (tmp_ptr != base_ptr)
      SWAP (tmp_ptr, base_ptr, size);

    /* Insertion sort, running from left-hand-side up to `right-hand-side.' 
       Pretty much straight out of the original GNU qsort routine. */

    for (run_ptr = base_ptr + size; (tmp_ptr = run_ptr += size) <= end_ptr; )
      {

        while (CMP (run_ptr, tmp_ptr -= size) < 0)
          ;

        if ((tmp_ptr += size) != run_ptr)
          {
            char *trav;

            for (trav = run_ptr + size; --trav >= run_ptr;)
              {
                char c = *trav;
                char *hi, *lo;

                for (hi = lo = trav; (lo -= size) >= tmp_ptr; hi = lo)
                  *hi = *lo;
                *hi = c;
              }
          }

      }
  }
  return 1;
}