Mercurial > hg > xemacs-beta
view lib-src/qsort.c @ 853:2b6fa2618f76
[xemacs-hg @ 2002-05-28 08:44:22 by ben]
merge my stderr-proc ws
make-docfile.c: Fix places where we forget to check for EOF.
code-init.el: Don't use CRLF conversion by default on process output. CMD.EXE and
friends work both ways but Cygwin programs don't like the CRs.
code-process.el, multicast.el, process.el: Removed.
Improvements to call-process-internal:
-- allows a buffer to be specified for input and stderr output
-- use it on all systems
-- implement C-g as documented
-- clean up and comment
call-process-region uses new call-process facilities; no temp file.
remove duplicate funs in process.el.
comment exactly how coding systems work and fix various problems.
open-multicast-group now does similar coding-system frobbing to
open-network-stream.
dumped-lisp.el, faces.el, msw-faces.el: Fix some hidden errors due to code not being defined at the right time.
xemacs.mak: Add -DSTRICT.
================================================================
ALLOW SEPARATION OF STDOUT AND STDERR IN PROCESSES
================================================================
Standard output and standard error can be processed separately in
a process. Each can have its own buffer, its own mark in that buffer,
and its filter function. You can specify a separate buffer for stderr
in `start-process' to get things started, or use the new primitives:
set-process-stderr-buffer
process-stderr-buffer
process-stderr-mark
set-process-stderr-filter
process-stderr-filter
Also, process-send-region takes a 4th optional arg, a buffer.
Currently always uses a pipe() under Unix to read the error output.
(#### Would a PTY be better?)
sysdep.h, sysproc.h, unexfreebsd.c, unexsunos4.c, nt.c, emacs.c, callproc.c, symsinit.h, sysdep.c, Makefile.in.in, process-unix.c: Delete callproc.c. Move child_setup() to process-unix.c.
wait_for_termination() now only needed on a few really old systems.
console-msw.h, event-Xt.c, event-msw.c, event-stream.c, event-tty.c, event-unixoid.c, events.h, process-nt.c, process-unix.c, process.c, process.h, procimpl.h: Rewrite the process methods to handle a separate channel for
error input. Create Lstreams for reading in the error channel.
Many process methods need change. In general the changes are
fairly clear as they involve duplicating what's used for reading
the normal stdout and changing for stderr -- although tedious,
as such changes are required throughout the entire process code.
Rewrote the code that reads process output to do two loops, one
for stdout and one for stderr.
gpmevent.c, tooltalk.c: set_process_filter takes an argument for stderr.
================================================================
NEW ERROR-TRAPPING MECHANISM
================================================================
Totally rewrite error trapping code to be unified and support more
features. Basic function is call_trapping_problems(), which lets
you specify, by means of flags, what sorts of problems you want
trapped. these can include
-- quit
-- errors
-- throws past the function
-- creation of "display objects" (e.g. buffers)
-- deletion of already-existing "display objects" (e.g. buffers)
-- modification of already-existing buffers
-- entering the debugger
-- gc
-- errors->warnings (ala suspended errors)
etc. All other error funs rewritten in terms of this one.
Various older mechanisms removed or rewritten.
window.c, insdel.c, console.c, buffer.c, device.c, frame.c: When creating a display object, added call to
note_object_created(), for use with trapping_problems mechanism.
When deleting, call check_allowed_operation() and note_object
deleted().
The trapping-problems code records the objects created since the
call-trapping-problems began. Those objects can be deleted, but
none others (i.e. previously existing ones).
bytecode.c, cmdloop.c: internal_catch takes another arg.
eval.c: Add long comments describing the "five lists" used to maintain
state (backtrace, gcpro, specbind, etc.) in the Lisp engine.
backtrace.h, eval.c: Implement trapping-problems mechanism, eliminate old mechanisms or
redo in terms of new one.
frame.c, gutter.c: Flush out the concept of "critical display section", defined by
the in_display() var. Use an internal_bind() to get it reset,
rather than just doing it at end, because there may be a non-local
exit.
event-msw.c, event-stream.c, console-msw.h, device.c, dialog-msw.c, frame.c, frame.h, intl.c, toolbar.c, menubar-msw.c, redisplay.c, alloc.c, menubar-x.c: Make use of new trapping-errors stuff and rewrite code based on
old mechanisms.
glyphs-widget.c, redisplay.h: Protect calling Lisp in redisplay.
insdel.c: Protect hooks against deleting existing buffers.
frame-msw.c: Use EQ, not EQUAL in hash tables whose keys are just numbers.
Otherwise we run into stickiness in redisplay because
internal_equal() can QUIT.
================================================================
SIGNAL, C-G CHANGES
================================================================
Here we change the way that C-g interacts with event reading. The
idea is that a C-g occurring while we're reading a user event
should be read as C-g, but elsewhere should be a QUIT. The former
code did all sorts of bizarreness -- requiring that no QUIT occurs
anywhere in event-reading code (impossible to enforce given the
stuff called or Lisp code invoked), and having some weird system
involving enqueue/dequeue of a C-g and interaction with Vquit_flag
-- and it didn't work.
Now, we simply enclose all code where we want C-g read as an event
with {begin/end}_dont_check_for_quit(). This completely turns off
the mechanism that checks (and may remove or alter) C-g in the
read-ahead queues, so we just get the C-g normal.
Signal.c documents this very carefully.
cmdloop.c: Correct use of dont_check_for_quit to new scheme, remove old
out-of-date comments.
event-stream.c: Fix C-g handling to actually work.
device-x.c: Disable quit checking when err out.
signal.c: Cleanup. Add large descriptive comment.
process-unix.c, process-nt.c, sysdep.c: Use QUIT instead of REALLY_QUIT.
It's not necessary to use REALLY_QUIT and just confuses the issue.
lisp.h: Comment quit handlers.
================================================================
CONS CHANGES
================================================================
free_cons() now takes a Lisp_Object not the result of XCONS().
car and cdr have been renamed so that they don't get used directly;
go through XCAR(), XCDR() instead.
alloc.c, dired.c, editfns.c, emodules.c, fns.c, glyphs-msw.c, glyphs-x.c, glyphs.c, keymap.c, minibuf.c, search.c, eval.c, lread.c, lisp.h: Correct free_cons calling convention: now takes Lisp_Object,
not Lisp_Cons
chartab.c: Eliminate direct use of ->car, ->cdr, should be black box.
callint.c: Rewrote using EXTERNAL_LIST_LOOP to avoid use of Lisp_Cons.
================================================================
USE INTERNAL-BIND-*
================================================================
eval.c: Cleanups of these funs.
alloc.c, fileio.c, undo.c, specifier.c, text.c, profile.c, lread.c, redisplay.c, menubar-x.c, macros.c: Rewrote to use internal_bind_int() and internal_bind_lisp_object()
in place of whatever varied and cumbersome mechanisms were
formerly there.
================================================================
SPECBIND SANITY
================================================================
backtrace.h: - Improved comments
backtrace.h, bytecode.c, eval.c: Add new mechanism check_specbind_stack_sanity() for sanity
checking code each time the catchlist or specbind stack change.
Removed older prototype of same mechanism.
================================================================
MISC
================================================================
lisp.h, insdel.c, window.c, device.c, console.c, buffer.c: Fleshed out authorship.
device-msw.c: Correct bad Unicode-ization.
print.c: Be more careful when not initialized or in fatal error handling.
search.c: Eliminate running_asynch_code, an FSF holdover.
alloc.c: Added comments about gc-cons-threshold.
dialog-x.c: Use begin_gc_forbidden() around code to build up a widget value
tree, like in menubar-x.c.
gui.c: Use Qunbound not Qnil as the default for
gethash.
lisp-disunion.h, lisp-union.h: Added warnings on use of VOID_TO_LISP().
lisp.h: Use ERROR_CHECK_STRUCTURES to turn on
ERROR_CHECK_TRAPPING_PROBLEMS and ERROR_CHECK_TYPECHECK
lisp.h: Add assert_with_message.
lisp.h: Add macros for gcproing entire arrays. (You could do this before
but it required manual twiddling the gcpro structure.)
lisp.h: Add prototypes for new functions defined elsewhere.
author | ben |
---|---|
date | Tue, 28 May 2002 08:45:36 +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; }