Release 3.3 Copyright ©1995 by NeXT Computer, Inc. All Rights Reserved.
|3.3 Release Notes:|
This file contains release notes for the 3.2, 3.1, and 3.0 releases of Malloc Debug. Items specific to the 3.1 release are listed first, and the Release 3.0 notes follow. There are no items specific to the 3.3 or 3.2 releases.
Notes Specific to Release 3.1
|The following new features have been added to MallocDebug since Release 3.0.|
|Touched nodes information.|
|MallocDebug can show you which nodes are accessed (read or written) by your application. Knowing which nodes are touched by your application is most useful for tuning the use of different allocation zones, thus improving your program's data locality and minimizing its working set. To learn more about using zones, look in /NextLibrary/Documentation/NextDev/Concepts/Performance.
To record which nodes are touched, MallocDebug must place each allocated node from the relevant zones on its own virtual memory page. Because of this additional memory requirement, you have control over which zones have this per-node monitoring enabled. After you link your application with libMallocDebug.a, you must run the mdbsetup program on your application to enable per-node monitoring for various zones. The command
mdbsetup MyApp.app/Myapp -protectable <zone list>
|enables the viewing of touched nodes within the zones listed in <zone list>. The strings "ALL" or "NONE" may also be specified instead of a specific list of zones.
mdbsetup MyApp.app/Myapp -unprotectable <zone list>
|enables the viewing of touched nodes within all zones except those listed in <zone list>. For examples, specifying "-unprotectable default ObjC" will allow you to see nodes touched in all zones besides the default and Objective-C zones.
mdbsetup MyApp.app/Myapp -print
|shows what zones within the application are enabled for touched node viewing.
After applying mdbsetup to your application, run the application and select it in MallocDebug as described above. To learn what nodes are touched for a given operation of your application, first press the Protect button. Then perform the operation in your application. While you are using the application, MallocDebug records which nodes are touched. To see this list, press the Touched button. To see what nodes have not been touched, press the Untouched button. To stop the recording of touched nodes, press the Unprotect button. Pressing Protect again cleans the slate of recorded nodes.
When touched nodes are being displayed, some new types of nodes are listed. Nodes marked with a `+' were allocated since the Protect button was pressed. Nodes marked with a `-' were allocated and freed since the Protect button was pressed.
3.0 Release Notes:
MallocDebug, a new tool included in Release 2.0, is designed to help you understand and improve the dynamic memory usage of the applications you write. MallocDebug consists of two components:
|a library containing a version of malloc that gathers statistics on memory use|
|an application for examining those statistics|
|The MallocDebug application is located in /NextDeveloper/Apps. Start the application and read the on-line help, which includes information on how to prepare your application for use with MallocDebug.|
|Differences from standard malloc|
|The debugging version of malloc used by MallocDebug is implemented differently from the standard system malloc so that its internal data structures are much less susceptible to being damaged by aberrant programs. This causes several incompatibilities with the standard malloc:
The debugging malloc implements zones by tagging each node with a zone tag rather than actually allocating the node in a different region of memory. Consequently, locality of reference of a program which uses zones cannot be determined by examining the addresses of nodes in MallocDebug.
The debugging malloc does not store information about nodes in the nodes themselves, but rather in an auxiliary data structure which is kept in a remote area of memory. This data is accessed by hashing the address of the node, but this information cannot be accessed from an address which is within the node. For this reason, the NXZoneFromPtr function will work correctly only when passed a pointer to the start of a node. Otherwise it will return NX_NOZONE. For the same reason the leak detector may report a node as a leak even if there is a pointer to some internal part of that node.