view lib-src/qsort.c @ 826:6728e641994e

[xemacs-hg @ 2002-05-05 11:30:15 by ben] syntax cache, 8-bit-format, lots of code cleanup README.packages: Update info about --package-path. i.c: Create an inheritable event and pass it on to XEmacs, so that ^C can be handled properly. Intercept ^C and signal the event. "Stop Build" in VC++ now works. bytecomp-runtime.el: Doc string changes. compat.el: Some attempts to redo this to make it truly useful and fix the "multiple versions interacting with each other" problem. Not yet done. Currently doesn't work. files.el: Use with-obsolete-variable to avoid warnings in new revert-buffer code. xemacs.mak: Split up CFLAGS into a version without flags specifying the C library. The problem seems to be that minitar depends on zlib, which depends specifically on libc.lib, not on any of the other C libraries. Unless you compile with libc.lib, you get errors -- specifically, no _errno in the other libraries, which must make it something other than an int. (#### But this doesn't seem to obtain in XEmacs, which also uses zlib, and can be linked with any of the C libraries. Maybe zlib is used differently and doesn't need errno, or maybe XEmacs provides an int errno; ... I don't understand. Makefile.in.in: Fix so that packages are around when testing. abbrev.c, alloc.c, buffer.c, buffer.h, bytecode.c, callint.c, casefiddle.c, casetab.c, casetab.h, charset.h, chartab.c, chartab.h, cmds.c, console-msw.h, console-stream.c, console-x.c, console.c, console.h, data.c, device-msw.c, device.c, device.h, dialog-msw.c, dialog-x.c, dired-msw.c, dired.c, doc.c, doprnt.c, dumper.c, editfns.c, elhash.c, emacs.c, eval.c, event-Xt.c, event-gtk.c, event-msw.c, event-stream.c, events.c, events.h, extents.c, extents.h, faces.c, file-coding.c, file-coding.h, fileio.c, fns.c, font-lock.c, frame-gtk.c, frame-msw.c, frame-x.c, frame.c, frame.h, glade.c, glyphs-gtk.c, glyphs-msw.c, glyphs-msw.h, glyphs-x.c, glyphs.c, glyphs.h, gui-msw.c, gui-x.c, gui.h, gutter.h, hash.h, indent.c, insdel.c, intl-win32.c, intl.c, keymap.c, lisp-disunion.h, lisp-union.h, lisp.h, lread.c, lrecord.h, lstream.c, lstream.h, marker.c, menubar-gtk.c, menubar-msw.c, menubar-x.c, menubar.c, minibuf.c, mule-ccl.c, mule-charset.c, mule-coding.c, mule-wnnfns.c, nas.c, objects-msw.c, objects-x.c, opaque.c, postgresql.c, print.c, process-nt.c, process-unix.c, process.c, process.h, profile.c, rangetab.c, redisplay-gtk.c, redisplay-msw.c, redisplay-output.c, redisplay-x.c, redisplay.c, redisplay.h, regex.c, regex.h, scrollbar-msw.c, search.c, select-x.c, specifier.c, specifier.h, symbols.c, symsinit.h, syntax.c, syntax.h, syswindows.h, tests.c, text.c, text.h, tooltalk.c, ui-byhand.c, ui-gtk.c, unicode.c, win32.c, window.c: Another big Ben patch. -- FUNCTIONALITY CHANGES: add partial support for 8-bit-fixed, 16-bit-fixed, and 32-bit-fixed formats. not quite done yet. (in particular, needs functions to actually convert the buffer.) NOTE: lots of changes to regex.c here. also, many new *_fmt() inline funs that take an Internal_Format argument. redo syntax cache code. make the cache per-buffer; keep the cache valid across calls to functions that use it. also keep it valid across insertions/deletions and extent changes, as much as is possible. eliminate the junky regex-reentrancy code by passing in the relevant lisp info to the regex routines as local vars. add general mechanism in extents code for signalling extent changes. fix numerous problems with the case-table implementation; yoshiki never properly transferred many algorithms from old-style to new-style case tables. redo char tables to support a default argument, so that mapping only occurs over changed args. change many chartab functions to accept Lisp_Object instead of Lisp_Char_Table *. comment out the code in font-lock.c by default, because font-lock.el no longer uses it. we should consider eliminating it entirely. Don't output bell as ^G in console-stream when not a TTY. add -mswindows-termination-handle to interface with i.c, so we can properly kill a build. add more error-checking to buffer/string macros. add some additional buffer_or_string_() funs. -- INTERFACE CHANGES AFFECTING MORE CODE: switch the arguments of write_c_string and friends to be consistent with write_fmt_string, which must have printcharfun first. change BI_* macros to BYTE_* for increased clarity; similarly for bi_* local vars. change VOID_TO_LISP to be a one-argument function. eliminate no-longer-needed CVOID_TO_LISP. -- char/string macro changes: rename MAKE_CHAR() to make_emchar() for slightly less confusion with make_char(). (The former generates an Emchar, the latter a Lisp object. Conceivably we should rename make_char() -> wrap_char() and similarly for make_int(), make_float().) Similar changes for other *CHAR* macros -- we now consistently use names with `emchar' whenever we are working with Emchars. Any remaining name with just `char' always refers to a Lisp object. rename macros with XSTRING_* to string_* except for those that reference actual fields in the Lisp_String object, following conventions used elsewhere. rename set_string_{data,length} macros (the only ones to work with a Lisp_String_* instead of a Lisp_Object) to set_lispstringp_* to make the difference clear. try to be consistent about caps vs. lowercase in macro/inline-fun names for chars and such, which wasn't the case before. we now reserve caps either for XFOO_ macros that reference object fields (e.g. XSTRING_DATA) or for things that have non-function semantics, e.g. directly modifying an arg (BREAKUP_EMCHAR) or evaluating an arg (any arg) more than once. otherwise, use lowercase. here is a summary of most of the macros/inline funs changed by all of the above changes: BYTE_*_P -> byte_*_p XSTRING_BYTE -> string_byte set_string_data/length -> set_lispstringp_data/length XSTRING_CHAR_LENGTH -> string_char_length XSTRING_CHAR -> string_emchar INTBYTE_FIRST_BYTE_P -> intbyte_first_byte_p INTBYTE_LEADING_BYTE_P -> intbyte_leading_byte_p charptr_copy_char -> charptr_copy_emchar LEADING_BYTE_* -> leading_byte_* CHAR_* -> EMCHAR_* *_CHAR_* -> *_EMCHAR_* *_CHAR -> *_EMCHAR CHARSET_BY_ -> charset_by_* BYTE_SHIFT_JIS* -> byte_shift_jis* BYTE_BIG5* -> byte_big5* REP_BYTES_BY_FIRST_BYTE -> rep_bytes_by_first_byte char_to_unicode -> emchar_to_unicode valid_char_p -> valid_emchar_p Change intbyte_strcmp -> qxestrcmp_c (duplicated functionality). -- INTERFACE CHANGES AFFECTING LESS CODE: use DECLARE_INLINE_HEADER in various places. remove '#ifdef emacs' from XEmacs-only files. eliminate CHAR_TABLE_VALUE(), which duplicated the functionality of get_char_table(). add BUFFER_TEXT_LOOP to simplify iterations over buffer text. define typedefs for signed and unsigned types of fixed sizes (INT_32_BIT, UINT_32_BIT, etc.). create ALIGN_FOR_TYPE as a higher-level interface onto ALIGN_SIZE; fix code to use it. add charptr_emchar_len to return the text length of the character pointed to by a ptr; use it in place of charcount_to_bytecount(..., 1). add emchar_len to return the text length of a given character. add types Bytexpos and Charxpos to generalize Bytebpos/Bytecount and Charbpos/Charcount, in code (particularly, the extents code and redisplay code) that works with either kind of index. rename redisplay struct params with names such as `charbpos' to e.g. `charpos' when they are e.g. a Charxpos, not a Charbpos. eliminate xxDEFUN in place of DEFUN; no longer necessary with changes awhile back to doc.c. split up big ugly combined list of EXFUNs in lisp.h on a file-by-file basis, since other prototypes are similarly split. rewrite some "*_UNSAFE" macros as inline funs and eliminate the _UNSAFE suffix. move most string code from lisp.h to text.h; the string code and text.h code is now intertwined in such a fashion that they need to be in the same place and partially interleaved. (you can't create forward references for inline funs) automated/lisp-tests.el, automated/symbol-tests.el, automated/test-harness.el: Fix test harness to output FAIL messages to stderr when in batch mode. Fix up some problems in lisp-tests/symbol-tests that were causing spurious failures.
author ben
date Sun, 05 May 2002 11:33:57 +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;
}