The main changes are when the internal of the buffers structure
were adressed directly, we now use routines coming from buf.h
The routine xmlParserInputRead() which wasn't used anywhere is
deprecated too.
Since the whole set of structures was public, the only way
to switch to size_t clean buffer is to introduce an incompatible
API change. Modifying the xmlParserInputBuffer and xmlOutputBuffer
structures is the best place to make this change as those
structures are deep into the parser feeding data, and no public
API suggest to build those manually.
* encoding.c: adds xmlCharEncFirstLineInput, xmlCharEncInput and
xmlCharEncOutput
* enc.h: the functions are not made public but added to this new header
This also add converter functions between xmlBuf and xmlBuffer
* buf.c buf.h: the old xmlBuffer routines but modified for size_t
and using xmlBuf instead of xmlBuffer
* Makefile.am: add the 2 new files
* include/libxml/xmlerror.h: add an entry for the new module
* include/libxml/tree.h: expose the xmlBufPtr type but not the
structure which stay private
Unless explicietely asked for when validating or replacing entities
with their value. Problem pointed out by Tom Lane <tgl@redhat.com>
* parser.c: do not load external parsed entities unless needed
* test/errors/extparsedent.xml result/errors/extparsedent.xml*:
add a regression test to avoid change of the behaviour in the future
tsan reported that rand() is not thread safe, so create
a thread safe wrapper, use rand_r() if available.
Consolidate the function, initialization and cleanup in
dict.c and make sure it is initialized in xmlInitParser()
For https://bugzilla.gnome.org/show_bug.cgi?id=654567
I use xmlTextReader to parse failed that might be incomplete. These files are
the beginning of a well-formed file, but the end is missing so the file as a
whole is not well-formed.
The problem is that xmlTextReader starts returning errors when it encounters
the early EOF, even though I haven't finished reading all of the valid data in
the file. It would be helpful if xmlTextReader kept working until the very
end.
For https://bugzilla.gnome.org/show_bug.cgi?id=622023
when compiled with LDFLAGS="${LDFLAGS} -Wl,-z,-defs -Wl,--no-undefined"
the python module would failed due to the undefined. This add an
explicit reference to python lib.
For https://bugzilla.gnome.org/show_bug.cgi?id=643148
Reported by Bill Clarke <llib@computer.org>, it used a global variable
as a counter for the input id and this was not thread safe. To avoid the
race without adding unneeded locking in the parser path, move the id to
the parser context instead.
The freeing function wasn't called due to a bogus #ifdef surrounding
value. Also switch the code to use the normal libxml2 allocation and
freeing routines.
On Fri, May 11, 2012 at 9:10 AM, Daniel Veillard <veillard@redhat.com> wrote:
> Hi Conrad,
>
> that's interesting ! I was initially afraid of a sudden explosion of
> memory allocations for building a tree since by default buffers tend to
> "waste" memory by using doubling allocations, but that's not the case.
> xmllint --noout doc/libxml2-api.xml
> when compiled with memory debug produce
>
> paphio:~/XML -> cat .memdump
> MEMORY ALLOCATED : 0, MAX was 12756699
>
> and without your patch 12755657, i.e. the increase is minimal.
Heh, I thought that too. Actually you're looking at the result with XML_ALLOC_EXACT! This
is because EXACT adds 10bytes "spare" on each alloc, and that interestingly wastes about the
same amount of space as XML_ALLOC_DOUBLEIT on this example (see below).
So it turns out that the default realloc() on my system actually handles this case really
well — and I guess that all the time in xmlRealloc() was actually in xmlStrlen, not the
underlying realloc() after all (sorry for misleading you). If you replace the realloc()
with a bad one (like valgrind's), then the performance degrades severely.
This patch implements a HYBRID allocator which has the behaviour you describe (it's
like EXACT to start with, though without the spare 10 bytes; and switches to DOUBLEIT
after 4kb) — that gets the memory back down to 12755657, with no noticeable impact on the
performance of the synthetic pathological example under valgrind.
In summary:
max_memory on ./xmllint --noout doc/libxml2-api.xml,
valgrind time on https://gist.github.com/2656940
max_memory valgrind time
before | 12755657 | 29:18.2
EXACT | 12756699 | 2:58.6 <-- this is the state after the first patch.
DOUBLEIT | 12756727 | 0:02.7
HYBRID | 12755754 | 0:02.7 <-- this is the state with both patches.
>
> There is also the cost of creating the buffers all the time.
> I need to read the code and check but I may be interested in an hybrid
> approach where we switch to buffer only when the text node starts to
> become too big (4k would remove nearly all usuall types of "document"
> usage, i.e. not blocks of data)
I tried to avoid too much buffer creation by introducing the xmlBufferDetach function,
which allows re-using one buffer to construct many strings. It's maybe a bit of a "hack"
in API terms though I thought the gains would be worth it.
Conrad
------8<------
To keep memory usage tight in normal conditions it's desirable to only
allocate as much space as is needed. Unfortunately this can lead to
problems when constructing a long string out of small chunks, because
every chunk you add will need to resize the buffer.
To fix this XML_ALLOC_HYBRID will switch (when the buffer is 4kb big)
from using exact allocations to doubling buffer size every time it is
full. This limits the number of buffer resizes to O(log n) (down from
O(n)), and thus greatly increases the performance of constructing very
large strings in this manner.
Hi Veillard and all,
Firstly, thanks for libxml: it's awesome!
I noticed recently that libxml was taking a surprisingly long time to perform some
operations (many minutes instead of milliseconds), and so I did some digging. It turns out
that the problem was caused by the realloc()ing done in xmlNodeAddContentLen() which can
be called many (many) times when assigning some content into a node.
For background, I'm dealing with XML that contains emails, these can have large
attachments (~6MB) which are base-64 encoded, line-wrapped at 78 chars, and each line ends
with . This means that xmlNodeAddContentLen() is being called about 200,000 times,
and so there are 200,000 reallocs of a 6MB string, which takes a while... (I put a synthetic
example of this at https://gist.github.com/2656940)
The attached patch works around that problem by using the existing buffer API to merge the
strings together before even creating the text node, this keeps the number of realloc()s
at a managable level.
I'd love feedback on the patch, and am happy to fix problems with it, or explore other
solutions if you think that this is barking up the wrong tree :).
Thanks,
Conrad
P.S. Should I create a bug for this too?
------8<------
Before this change xmlStringGetNodeList would perform a realloc() of the
entire new content for every XML entity in the assigned text in order to
merge together adjacent text nodes. This had the effect of making
xmlSetNodeContent O(n^2), which led to unexpectedly bad performance on
inputs that contained a large number of XML entities.
After this change the memory management is done by the buffer API,
avoiding the need to continually re-measure and realloc() the string.
For my test data (6MB of 80 character lines, each ending with )
this takes the time to xmlSetNodeContent from about 500 seconds to
around 50ms. I have not profiled smaller cases, though I tried to
minimize the performance impact of my change by avoiding unnecessary
string copying.
Signed-off-by: Conrad Irwin <conrad.irwin@gmail.com>
For https://bugzilla.gnome.org/show_bug.cgi?id=615785
When the <noscript> is found, <head> is closed and a <body> element is created.
The real <body id="xxx"> gets skipped over, so I can't see any of the
body's attributes.
Just don't close <head> when encountering a <noscript>
Add a regression test too
For https://bugzilla.gnome.org/show_bug.cgi?id=609796
Libxml2 fails to validate an instance document against a schema if an element
whose type is a complex extension of some base type with an optional child
element and that child element is not specified in the instance document. For
example, suppose I have some complex type BaseType that is defined to have one
child element in a sequence group that has minOccurs set to 0
For https://bugzilla.gnome.org/show_bug.cgi?id=630682
The python tests were reporting errors, some of it was due to
a small change in case encoding, but the main one was about
htmlSetMetaEncoding(doc, NULL) being broken by not removing
the associated meta tag anymore
For https://bugzilla.gnome.org/show_bug.cgi?id=642916
I just noticed that the HTML_PARSE_NOIMPLIED flag that you can pass to the
HTML-Parser methods doesn't do anything. Its intended purpose is to stop the
HTML-parser from forcibly adding a pair of html/body tags if the stream does
not contain any.
This is highly useful when you don't need this level of strictness.
Unfortunately, specifying it doesn't work, because the option is not
copied into the parsing context.
For https://bugzilla.gnome.org/show_bug.cgi?id=643949
In case of error on an IO creation input the given context
is terminated with the given close function, except if the
error happened in xmlParserInputBufferCreateIO. This can
lead to a resource leak which is fixed by this patch.
For https://bugzilla.gnome.org/show_bug.cgi?id=655218http://www.w3.org/TR/2011/WD-html5-20110525/semantics.html#the-meta-element
"""
The charset attribute specifies the character encoding used by the document.
This is a character encoding declaration. If the attribute is present in an XML
document, its value must be an ASCII case-insensitive match for the string
"UTF-8" (and the document is therefore forced to use UTF-8 as its
encoding).
"""
However, while <meta http-equiv="Content-Type" content="text/html;
charset=utf8"> works, <meta charset="utf8"> does not.
While libxml2 HTML parser is not tuned for HTML5, this is a simple
addition
Also added a testcase
