Vincent Jacques wrote:
Hello,
I'm trying to use the SAX2 API to implement a match function for
boost::asio::async_read_until. Please don't stop reading if you don't
know this library, I'll try to explain :-)
This boost::asio::async_read_until function reads some bytes on a tcp/ip
socket, then calls the user-provided match function to check if it has
read enough data. If yes, it triggers a user-provided callback function.
If no, it performs a new read on the socket, and tries again.
What I want is that the callback is called when I receive each complete
element which is a direct son of the root element:
<?xml version="1.0">
<root>
<element><foo><bar/></foo></element> <!-- CALLBACK HERE -->
<bee><gee/></bee> <!-- CALLBACK HERE -->
</root>
So, I've designed a SAX2 content handler which tells me if the parsing
is at such a place, and I loop on SAX2XMLReader::parseNext, asking my
handler if it's ok to call the callback, each time it returns.
I've also designed a BinInputStream and a InputSource that transfers the
data given by Boost.Asio to Xerces.
My problem is that my match function is called in a context where I'm
not sure I'll be able to give anything to Xerces if it calls readBytes.
It's possible for example, that I've already given
"<?xml version="1.0"><root><element><foo><bar/></foo></elem"
to Xerces, because I've received only this. In this case, I must reply
'false' to Boost, to wait for some more bytes.
But at this point, Xerces calls readBytes, and I don't know what to do:
-if I give it 0 bytes, it throws an exception telling it has reached the
end of file, and I cannot resume the parse when I've received more data.
-if I throw an exception from readBytes, It's no better.
*To make it short*: I need to cleanly exit parseNext when it calls
readBytes and I cannot give anything, while keeping the possibility to
call parseNext later.
Any idea ?
Unfortunately, parseFirst() and parseNext() weren't designed to support
speculative reading of the input source. They assume the input is
always available, or that a call to readBytes() will block until
something is available. Furthermore, parseNext() won't return until
it's actually parsed something, and what that unit of work is is not
well-specified.
My gut feeling is it would be difficult to implement this behavior with
parseFirst() and parseNext(), because they are implemented using the
same core code as parse().
Dave
