Table of Content:
- General overview
- Setting libxml2 set of memory routines
- Cleaning up after using the library
- Debugging routines
- General memory requirements
- Returning memory to the kernel
The module xmlmemory.h
provides the interfaces to the libxml2 memory system:
- libxml2 does not use the libc memory allocator directly but xmlFree(),
xmlMalloc() and xmlRealloc()
- those routines can be reallocated to a specific set of routine, by
default the libc ones i.e. free(), malloc() and realloc()
- the xmlmemory.c module includes a set of debugging routine
It is sometimes useful to not use the default memory allocator, either for
debugging, analysis or to implement a specific behaviour on memory management
(like on embedded systems). Two function calls are available to do so:
- xmlMemGet
() which return the current set of functions in use by the parser
- xmlMemSetup()
which allow to set up a new set of memory allocation functions
Of course a call to xmlMemSetup() should probably be done before calling
any other libxml2 routines (unless you are sure your allocations routines are
compatibles). Libxml2 is not stateless, there is a few set of memory structures needing
allocation before the parser is fully functional (some encoding structures
for example). This also mean that once parsing is finished there is a tiny
amount of memory (a few hundred bytes) which can be recollected if you don't
reuse the library or any document built with it:
- xmlCleanupParser
() is a centralized routine to free the library state and data. Note
that it won't deallocate any produced tree if any (use the xmlFreeDoc()
and related routines for this). This should be called only when the library
is not used anymore.
- xmlInitParser
() is the dual routine allowing to preallocate the parsing state
which can be useful for example to avoid initialization reentrancy
problems when using libxml2 in multithreaded applications
Generally xmlCleanupParser() is safe assuming no parsing is ongoing and
no document is still being used, if needed the state will be rebuild at the
next invocation of parser routines (or by xmlInitParser()), but be careful
of the consequences in multithreaded applications. When configured using --with-mem-debug flag (off by default), libxml2 uses
a set of memory allocation debugging routines keeping track of all allocated
blocks and the location in the code where the routine was called. A couple of
other debugging routines allow to dump the memory allocated infos to a file
or call a specific routine when a given block number is allocated: When developing libxml2 memory debug is enabled, the tests programs call
xmlMemoryDump () and the "make test" regression tests will check for any
memory leak during the full regression test sequence, this helps a lot
ensuring that libxml2 does not leak memory and bullet proof memory
allocations use (some libc implementations are known to be far too permissive
resulting in major portability problems!). If the .memdump reports a leak, it displays the allocation function and
also tries to give some information about the content and structure of the
allocated blocks left. This is sufficient in most cases to find the culprit,
but not always. Assuming the allocation problem is reproducible, it is
possible to find more easily:
- write down the block number xxxx not allocated
- export the environment variable XML_MEM_BREAKPOINT=xxxx , the easiest
when using GDB is to simply give the command
set environment XML_MEM_BREAKPOINT xxxx
before running the program.
- run the program under a debugger and set a breakpoint on
xmlMallocBreakpoint() a specific function called when this precise block
is allocated
- when the breakpoint is reached you can then do a fine analysis of the
allocation an step to see the condition resulting in the missing
deallocation.
I used to use a commercial tool to debug libxml2 memory problems but after
noticing that it was not detecting memory leaks that simple mechanism was
used and proved extremely efficient until now. Lately I have also used valgrind with quite some
success, it is tied to the i386 architecture since it works by emulating the
processor and instruction set, it is slow but extremely efficient, i.e. it
spot memory usage errors in a very precise way. How much libxml2 memory require ? It's hard to tell in average it depends
of a number of things:
- the parser itself should work in a fixed amount of memory, except for
information maintained about the stacks of names and entities locations.
The I/O and encoding handlers will probably account for a few KBytes.
This is true for both the XML and HTML parser (though the HTML parser
need more state).
- If you are generating the DOM tree then memory requirements will grow
nearly linear with the size of the data. In general for a balanced
textual document the internal memory requirement is about 4 times the
size of the UTF8 serialization of this document (example the XML-1.0
recommendation is a bit more of 150KBytes and takes 650KBytes of main
memory when parsed). Validation will add a amount of memory required for
maintaining the external Dtd state which should be linear with the
complexity of the content model defined by the Dtd
- If you need to work with fixed memory requirements or don't need the
full DOM tree then using the xmlReader
interface is probably the best way to proceed, it still allows to
validate or operate on subset of the tree if needed.
- If you don't care about the advanced features of libxml2 like
validation, DOM, XPath or XPointer, don't use entities, need to work with
fixed memory requirements, and try to get the fastest parsing possible
then the SAX interface should be used, but it has known restrictions.
You may encounter that your process using libxml2 does not have a
reduced memory usage although you freed the trees. This is because
libxml2 allocates memory in a number of small chunks. When freeing one
of those chunks, the OS may decide that giving this little memory back
to the kernel will cause too much overhead and delay the operation. As
all chunks are this small, they get actually freed but not returned to
the kernel. On systems using glibc, there is a function call
"malloc_trim" from malloc.h which does this missing operation (note that
it is allowed to fail). Thus, after freeing your tree you may simply try
"malloc_trim(0);" to really get the memory back. If your OS does not
provide malloc_trim, try searching for a similar function. Daniel Veillard | 
