settings.dist

/*
 * 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 1.71 2007/03/13 01:11:25 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.
 *
 * Alternatives might be:
 *
 *      #define KILL_PROCESS    { int *_int_p = 0L; *_int_p = 1; }
 */
#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>

/*
 * Write the iteration count at the start of every log entry.  This is
 * handy when you are using the DMALLOC_START variable and want to
 * begin the tough debugging at a certain point.  This is also returned
 * by the dmalloc_mark() function.
 */
#define LOG_ITERATION 1

/*
 * Write the pid number at the start of every log entry if the
 * getpid() function is available.  This is handy when you are using
 * dmalloc with a program which forks.  See LOG_REOPEN below.
 */
#define LOG_PID 0

/*
 * If the getpid() function is available, notice when the pid of the
 * process changes and reopen the logfile.  This is handy when you are
 * using dmalloc with a program which forks and you want the separate
 * forked programs to have separate logs.
 *
 * NOTE: This only works if the %p string is in the logfile name
 * otherwise the log might reopen and clobber the existing log.
 */
#define LOG_REOPEN 1

/*
 * Store the number of times a pointer is "seen" being allocated or
 * freed -- it shows up as a s# (for seen) in the logfile.  This is
 * useful for tracking of not-freed memory.  See the documents for more
 * information.
 *
 * NOTE: This creates a certain amount of memory overhead.
 */
#define LOG_PNT_SEEN_COUNT 1

/*
 * Store the iteration count when a pointer is allocated -- it
 * shows up as a i# (for iteration) in the logfile.  This is to give
 * you some idea when during program execution, a pointer was
 * allocated but not freed.
 *
 * NOTE: This creates a certain amount of memory overhead.
 */
#define LOG_PNT_ITERATION 0

/*
 * At the front of each log message, print the output from the ctime()
 * function (not including the \n).  The TIME_NUMBER_TYPE is the type
 * that will store the output of time() and whose address we will pass
 * into ctime.  This requires that the ctime and time functions both
 * be defined.
 */
#if LOG_TIME_NUMBER == 0
#define LOG_CTIME_STRING	1
#define TIME_TYPE		unsigned int
#endif

/*
 * Store the time (in seconds) or timeval (in seconds and
 * microseconds) when a pointer is allocated -- it shows up as a w#
 * (for when) in the logfile.  This is to give you some idea when a
 * pointer was allocated but not freed.  The library will log the
 * starting and the ending time if either of these flags is set.
 * TIMEVAL_INCLUDE is the include file to define struct timeval and
 * GET_TIMEVAL does the actual reading of the current time of day.
 *
 * WARNING: only TIME _or_ TIMEVAL can be defined at one time.
 *
 * NOTE: This creates a certain amount of memory overhead.
 */
#define LOG_PNT_TIME		0
#ifndef TIME_INCLUDE
#define TIME_INCLUDE		<time.h>
#endif
#ifndef TIME_TYPE
#define TIME_TYPE		unsigned int
#endif
#define LOG_PNT_TIMEVAL		0
#define TIMEVAL_INCLUDE		<sys/time.h>
#define TIMEVAL_TYPE		struct timeval
#define GET_TIMEVAL(timeval)	(void)gettimeofday(&(timeval), NULL)

/*
 * In OSF (anyone else?) you can setup __fini_* functions in each
 * module which will be called automagically at shutdown of the
 * program.  If you enable this variable, dmalloc will shut itself
 * down and log statistics when the program closes on its own.  Pretty
 * cool OS feature.
 */
#define FINI_DMALLOC 0

/* If you enable this, you probably want the AUTO_SHUTDOWN flag turned off */
#if FINI_DMALLOC
#undef AUTO_SHUTDOWN
#define AUTO_SHUTDOWN 0
#endif

/*
 * Keep addresses that are freed from recycling back into the used
 * queue for a certain number of memory transactions.  For instance,
 * if this is set to 10 then after you free a pointer, it cannot be
 * reused until after 10 additional calls to malloc, free, realloc,
 * etc..  Define to 0 to disable.  NOTE: setting to 1 does nothing.
 *
 * For more drastic debugging, you can enable the never-reuse flag
 * which will cause the library to never reuse previously allocated
 * memory.  This may significantly expand the memory requirements of
 * your system however.
 */
#define FREED_POINTER_DELAY 20

/*
 * Size of the table of file and line number memory entries.  This
 * memory table records the top locations by file/line or
 * return-address of all pointers allocated.  It also tabulates the
 * freed memory pointers so you can easily locate the large memory
 * leaks.  See the MEMORY_TABLE_TOP_LOG value below to 0 to disable
 * the table.
 *
 * NOTE: The table will only hold the _first_ pointers into the table.
 * If you configure a size of 10 then the 11th pointer allocated will
 * not be accounted for.
 *
 * NOTE: the library will actually allocated 2 times this many entries
 * for speed reasons.
 */
#define MEMORY_TABLE_SIZE 4096

/*
 * This indicates how many of the top entries from the memory table
 * you want to log by default to the log file.
 *
 * NOTE: to display the top entries correctly, your OS must support
 * the quicksort function.
 */
#define MEMORY_TABLE_TOP_LOG 10

/*
 * Define this to 1 to only display the memory table summary of the
 * dumped table pointers.  The default is to display the summary as
 * well as the individual pointers so the individual leaks can be
 * tracked down.
 */
#define DUMP_UNFREED_SUMMARY_ONLY 0

/*
 * If (and _only_ if) your system does not have sbrk(), you can have
 * dmalloc pre-allocate its only heap space.  The default heap size is
 * 1mb but you can set the space to be any size.  For super-small
 * memory applications please understand that dmalloc is in no way
 * optimized for space and so you can easily run out of memory with
 * it.
 *
 * WARNING: this probably should only be used if HAVE_SBRK and
 * HAVE_MMAP are 0.  Please send me email with any problems or
 * comments on this feature.
 */
#if HAVE_SBRK == 0 && HAVE_MMAP == 0
#define INTERNAL_MEMORY_SPACE (1024 * 1024)
#endif

/*
 * The default smallest allowable allocations in bytes.  Any blocks
 * asked for that are smaller will be rounded up to this size.
 */
#define DEFAULT_SMALLEST_ALLOCATION	8

/****************************** thread settings ******************************/

/*
 * The following definition allows use of the library in threaded
 * programs.  The most common package is MIT's pthreads so this is the
 * default.  Please send me mail if these definitions are configurable
 * enough to work with your thread package.
 */

#ifndef LOCK_THREADS
#define LOCK_THREADS 1
#endif

#if LOCK_THREADS

/*
 * Which threads library header to include when needed.  It is assumed
 * that the types and functions in the THREAD_TYPE and THREAD_GET_ID
 * macros below are defined in this include file.  In addition, the
 * thread mutex init, lock, and unlock functions in malloc.c should
 * also be prototyped here.
 */
#define THREAD_INCLUDE			<pthread.h>

/*
 * As we approach the time when we start mutex locking the library, we
 * need to init the mutex variable.  This sets how many times before
 * we start locking should we init the variable taking in account that
 * the init itself might generate a call into the library.  Ugh.
 */
#define THREAD_INIT_LOCK	2

/*
 * For those threaded programs, the following settings allow the
 * library to log the identity of the thread that allocated a specific
 * pointer.  The thread-id will show up as a ``t'' followed by a
 * string identifying the thread.  The LOG_THREAD_ID macro says
 * whether the thread-id is logged at the front of all log messages.
 * The THREAD_TYPE macro defines the variable type of the id.  The
 * THREAD_GET_ID macro is what function to call to get the currently
 * running thread.  The THREAD_ID_TO_STRING defines how the thread-id
 * value is translated to the string necessary to be included with the
 * ``t'' in the logfile.
 */
#define LOG_PNT_THREAD_ID		0
#define THREAD_TYPE			pthread_t
#define THREAD_GET_ID()			pthread_self()
#if HAVE_SNPRINTF
#define THREAD_ID_TO_STRING(buf, buf_size, thread_id)	\
				(void)snprintf((buf), (buf_size), "%#lx", \
					       (long)(thread_id))
#else
#define THREAD_ID_TO_STRING(buf, buf_size, thread_id)	\
				(void)sprintf((buf), "%#lx", (long)(thread_id))
#endif

#endif /* LOCK_THREADS */

#endif /* ! __SETTINGS_H__ */