settings.dist

减少内存碎片的malloc分配函数

/*
 * Manual configuration flags
 *
 * Copyright 2000 by Gray Watson
 *
 * This file is part of the dmalloc package.
 *
 * Permission to use, copy, modify, and distribute this software for any
 * purpose and without fee is hereby granted, provided
 * that the above copyright notice and this permission notice appear
 * in all copies, and that the name of Gray Watson not be used in
 * advertising or publicity pertaining to distribution of the document
 * or software without specific, written prior permission.
 *
 * Gray Watson makes no representations about the suitability of the
 * software described herein for any purpose.  It is provided "as is"
 * without express or implied warranty.
 *
 * The author may be contacted via http://dmalloc.com/
 *
 * $Id: settings.dist,v 1.68 2004/09/13 05:14:04 gray Exp $
 */

/*
 * PROGRAMMING NOTE: this file cannot be included before conf.h, so
 * you might as well only include conf.h and never this file.
 */

#ifndef __SETTINGS_H__
#define __SETTINGS_H__

/*
 * Should we allow zero length allocations?  This will generate the
 * smallest possible allocation.
 *
 * FYI: if fence post checking is requested, the top and bottom of the
 * fence post information will be touching.
 */
#define ALLOW_ALLOC_ZERO_SIZE 1

/*
 * Should we allow realloc of a NULL pointer?  If set to one, this
 * will call malloc when 0L is realloc-ed.  This is useful when you
 * are extending an array in a loop and do not want to allocate it
 * specially the first time.
 */
#define ALLOW_REALLOC_NULL 1

/*
 * Should we allow realloc to a 0 size to cause the pointer to be
 * freed?  If set to one, this will call free when a pointer is
 * realloc-ed to a size of 0.  Thanks to Stefan Froehlich
 * for patiently pointing that the realloc in just about every Unix
 * has this functionality.
 */
#define ALLOW_REALLOC_SIZE_ZERO 1

/*
 * I have swayed to public pressure and allowing free(0L) is now the
 * default.  Sigh.  :-)
 *
 * Should we allow the free of a NULL pointer and if it happens,
 * should a message be generated to that effect.  Most (if not all)
 * POSIX C libraries allow you to free() or delete() a NULL pointer
 * and a number of programming reference manuals mention that freeing
 * of NULL is "allowed".  However, I believe that it is bad form,
 * promotes lazy pointer handling, and can often hide program bugs.  I
 * encourage you to do something like the following in your programs:
 * if (pnt != NULL) { free(pnt); }
 *
 * Setting ALLOW_FREE_NULL to 0 will cause an exception whenever a
 * free of NULL is encountered allowing the user to determine where
 * s/he is freeing a possibly random pointer.  I recommend at least
 * logging a message with the ALLOW_FREE_NULL_MESSAGE set to 1.
 *
 * Even with ALLOW_FREE_NULL set to 1, you can enable the
 * 'error-free-null' token at runtime for a specific program to
 * generate an exception when it sees a free(0L).
 */
#define ALLOW_FREE_NULL 1
#define ALLOW_FREE_NULL_MESSAGE 1

/*
 * Should we use the ra-address macros in return.h.  These are system
 * specific macros designed to return the return-address for logging
 * callers (i.e. possible offenders) of malloc routines.
 *
 * Please mail me if you have any questions with this functionality.
 */
#define USE_RETURN_MACROS 1

/*
 * Write this character into memory when it is allocated and not
 * calloc-ed if the alloc-blank token is enabled.  It will also write
 * this into realloc'd memory which you extend or reduce.  You can
 * verify that these sections have not been overwritten with the
 * check-blank token.
 *
 * \332 == 0xda, 0332, decimal 218 (i.e. dmalloc-alloc)
 */
#define ALLOC_BLANK_CHAR	'\332'

/*
 * Write this character into memory when it is freed if the free-blank
 * token is enabled.  You can verify that these sections have not been
 * overwritten with the check-blank token.
 *
 * \337 == 0xdf, 0337, decimal 223 (i.e. dmalloc-free)
 */
#define FREE_BLANK_CHAR		'\337'

/*
 * The following information sets limits on the size of the source
 * file name and line numbers returned by the __FILE__ and __LINE__
 * compiler macros.  You may need to tune then to fit your environment
 * although I would argue that if you have filenames longer than 40
 * characters or files longer than 10,000 lines, you are doing
 * something wrong.
 *
 * MIN_FILE_LENGTH 3 => file "[a-zA-Z].c"
 *
 * Set MAX_FILE_LENGTH or MAX_LINE_NUMBER to be 0 to disable checks.
 *
 * NOTE: if the max is changed then the dmalloc.texi file reference to
 * it will need to be changed too.
 */
#define MIN_FILE_LENGTH		    3
#define MAX_FILE_LENGTH		  100
#define MAX_LINE_NUMBER		30000

/*
 * The largest allowable allocation size.  This is only for
 * verification purposes to control allocations of bizarre sizes.  Any
 * allocation larger than this will generate a ERROR_TOO_BIG error.
 *
 * Set to 0 to disable the test altogether.
 */
#define LARGEST_ALLOCATION	268435456UL		/* 256 mb */

/*
 * Automatically call dmalloc_shutdown if on_exit or atexit is
 * available.  See conf.h for whether configure found on_exit or
 * atexit calls.  If neither is available, your program will have to
 * call dmalloc_shutdown yourself before it exits.  You can also take
 * a look at atexit.c in the contrib directory which may provide this
 * useful functionality for your system.
 *
 * NOTE: If you are having problems with the library going recursive (see
 * LOCK_THREADS below if you are using pthreads), you might want to
 * try setting this to 0.  Because the library makes a call to on_exit
 * or atexit to register itself, it may cause memory transactions by
 * the system causing the dreaded recursive message.  You may then be
 * forced to register dmalloc_shutdown yourself via on_exit or atexit
 * in main() or call dmalloc_shutdown directly before you exit().
 */
#define AUTO_SHUTDOWN 1

/*
 * The ABORT_OKAY is auto-configured but may have to be adjusted by
 * forcing the USE_ABORT to be 1 or 0.  On some OS's, abort calls
 * fclose() which may want to free memory making the library go
 * recursive when it is aborting.  See ABORT_OKAY in the conf.h file
 * for more information.
 *
 * If you need to override it or if ABORT_OKAY is 0, set KILL_PROCESS to
 * the proper way to stop the program.  Killing the current process
 * (id 0) with SIGABRT works on a number of Unix systems.  You may
 * have to define some include file to get the value for the signal
 * that is used.
 */
#if ABORT_OKAY
#define USE_ABORT		1
#else
#define USE_ABORT		0
#endif
#if SIGNAL_OKAY
#define KILL_INCLUDE		<signal.h>
#define KILL_PROCESS		(void)kill(0, SIGABRT)
#endif

/*
 * Define the signals that are to be caught by the catch-signals
 * token.  When caught, these signals will cause an automatic shutdown
 * of the library so that the not-freed memory and other statistics will
 * be displayed.  Thanks Marty.
 */
#if SIGNAL_OKAY
#define SIGNAL1		SIGHUP
#define SIGNAL2		SIGINT
#define SIGNAL3		SIGTERM
#undef SIGNAL4
#undef SIGNAL5
#undef SIGNAL6
#endif

/*
 * Number of bytes to write at the top of allocations (if fence-post
 * checking is enabled).  A larger number means more memory space used
 * up but better protection against fence overruns.  See the manual
 * for more information.
 */
#define FENCE_TOP_SIZE 4

/*
 * Number of bytes to write at the bottom of allocations. See the
 * FENCE_TOP_SIZE setting above or the manual for more information.
 *
 * WARNING: this should changed with caution and probably should only be
 * increased.  If you need to change it, use (ALLOCATION_ALIGNMENT *
 * X) or some such.  For more information see ALLOCATION_ALIGNMENT in
 * conf.h.
 */
#define FENCE_BOTTOM_SIZE ALLOCATION_ALIGNMENT

/*
 * Amount of space that we are to display whenever we need to dump a
 * pointer's contents to a log file or stream.  This should be more
 * than FENCE_BOTTOM_SIZE and FENCE_TOP_SIZE.
 */
#define DUMP_SPACE 20

/*
 * At the front of each log message, print the output from the time()
 * call as a number.  This requires that the time function be defined.
 */
#define LOG_TIME_NUMBER		1
#define TIME_INCLUDE		<time.h>