Thilina, First of all thanks for your input. Actually in the current logic we are creating a new stream from pure xml and giving it to the xml_parser. The only difference is we keep on parsing for mime even after finding the SOAP. I hope we can modify it to stop further parsing at that point and do on demand parsing. But the problem with this approach is when the SOAP part is pretty large the whole message (I mean SOAP) is in memory. Is the same thing happens with Axis2/JAVA ?
Thanks, -Manjula. On Tue, 2008-07-15 at 03:26 -0700, Thilina Gunarathne wrote: > Hi Manjula, > One thing you guys can try to do is to break up the stream at the > MIME() parser level. Then the MIME parser only gives the SOAP Envelope > part to the XML parser and nothing beyond. This can be achieved by > faking a input stream. MIME parser gets the original input stream and > then gives the XML parser a faked inputstream, something which > provides the same interface and functionality as the original > inputstream, but the MIME parser has control over what goes in to > it. > > For an example, if reading an InputStream is reading a memory pointer > (please excuse my very bad knowledge about C I/O) which gets filled by > the transport layer, then the MIME parser can also write the data in > to some memory location which will act as the input stream location > for XMLReader. > > If the above is possible, then the MIME parser can stream the actual > input stream directly to the faked inputstream till the SOAP part > boundary is reached or till the user requests another MIME part. > Streaming MIME parsing will happen throughout the process to limit the > FAKED input stream to the contents of the SOAP part. If the user > requests a MIME part while the XML reader is only half way in to > parsing the SOAP envelope, then the MIME parser should read ahead the > SOAP part, store it in memory and should provide it as needed to the > FAKED input stream. One other way to visualize this is to think MIME > parsing happens at a level below the XML parsing, just like transport > level. > > I'm not sure how hard to do this in Axis2/C (or whether it's > impossible)... Let me know if I'm not clear with the idea. > thanks, > Thilina > > > On Tue, Jul 15, 2008 at 3:04 AM, Supun Kamburugamuva > <[EMAIL PROTECTED]> wrote: > > But then again the problem will occur when xml_parser > give the control > back to mime_parser. Because xml_parser will read > beyond the soap part > when parsing. So the parser should be modified to > return the the > unwanted data it read beyond the soap part while > parsing. I think > Guththila can be modified for this requirement, but > does any one know > how to do this with libxml2. > > I assume by "read beyond the soap part" you mean consuming the > stream beyond the soap part. But I think this is unavoidable > in any parser. > > 1. Parser don't know where the xml ends. > 2. The parser doesn't read character by chareacter from the > stream (if it does it'll be very inefficient) and it reads > chunks of data from the stream. > > Becuase of the above reasons the parser will consume the > stream beyond the soap part. > > Supun.. > > > > > > On Sat, 2008-07-12 at 21:21 +0530, Manjula Peiris > wrote: > > On Sat, 2008-07-12 at 19:31 +0530, Milinda Pathirage > wrote: > > > As Thilina mentioned in above thread Axis2/Java > keeps a instance of > > > mime parser in it's xml builder. So I think it's > ok to follow that > > > method. AFAIK, I think if we enable MTOM caching > we have to have this > > > feature or is anybody have ideas about how to > solve this without going > > > to this method?. Manjula can you point out the > changes we need in the > > > parser. Then we can discuss the problem and > available solutions. This > > > will help us to avoid any problems occur in the > future because of the > > > changes. > > > If we have two methods for xml reader creation, > how we can use both of > > > them in the Axis2/C engine. Are we going to check > the MIME headers and > > > create the required xml reader? > > > > In the MTOM case the serialized message is most > probably in the > > following format. > > > > --MimeBoundary > > mime_header1:xxxxx > > mime_header2:xxxxx > > > > <soapenv:Envelope>.......</soapenv:Envelope> > > --MimeBoundary > > mime_header1:xxxx > > mime_header2:xxxx > > > > yyyyyyyyybinarydatayyyyyyyyyyyy > > --MimeBoundary-- > > > > We have the information form http_headers whether > this is a multipart > > message or not. So will call the soap_builder > straightaway with a flag > > telling this is a multipart message. Since the > modified soap_builder > > will have a reference to mime_parser instance when > seeing this flag it > > will call the mime_parser. mime_parser will search > for \r\n\r\n sequence > > between SOAP part and first mime_headers set. When > searching for this it > > may read some part of the SOAP. So after this search > we can't > > straightaway call either xml_reader_create_for_io or > > xml_reader_create_for_memory, because none of them > will not get correct > > xml. So what we need is first read from a buffer and > then when that > > buffer is finished read from the network stream. > This will cause to > > define and implement the new xml_reader function and > corresponding > > modifications at parser level as I mentioned in my > previous mail. > > > > -Manjula. > > > > > > > > > > Thanks, > > > Milinda... > > > > > > On Sat, Jul 12, 2008 at 6:53 PM, Manjula Peiris > <[EMAIL PROTECTED]> > > > wrote: > > > Hi all, > > > > > > MTOM caching stuff is finished in both > send and receiving > > > sides.It is > > > available here [1]. But due to this [2] > problem, it cannot be > > > integrated > > > to Axis2/C trunk. > > > > > > In order to solve that problem we need to > keep a reference of > > > mime_parser instance in the soap_builder > similarly > > > stax_builder. So when > > > parsing the soap part the xml_reader will > not just read from > > > the stream, > > > but also need to read from the buffer > which will be available > > > after > > > parsing mime_headrs. So we need to create > a parser instance > > > which will > > > first read from the buffer and when that > buffer is not > > > containing any > > > more data, it will read from the stream. > So we may need to add > > > a new > > > function say, > > > xml_reader_create_for_memory_and_io and > change some of the > > > logic in > > > parser. > > > Your suggestions are highly appreciated. > > > > > > > > [1]https://svn.apache.org/repos/asf/webservices/axis2/branches/c/post_1_4_mtom/c > > > > > > > > [2]http://marc.info/?l=axis-c-dev&amp;m=121320959601769&amp;w=2 > > > > > > Thanks, > > > -Manjula. > > > > > > > > > -- > > > Manjula Peiris: > http://manjula-peiris.blogspot.com/ > > > > > > > > > > > --------------------------------------------------------------------- > > > To unsubscribe, e-mail: > [EMAIL PROTECTED] > > > For additional commands, e-mail: > [EMAIL PROTECTED] > > > > > > > > > > > > > > > -- > > > http://mpathirage.com > > > http://wso2.org "Oxygen for Web Service > Developers" > > > http://wsaxc.blogspot.com "Web Services With > Axis2/C" > > > > > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: > [EMAIL PROTECTED] > > For additional commands, e-mail: > [EMAIL PROTECTED] > > > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: > [EMAIL PROTECTED] > For additional commands, e-mail: > [EMAIL PROTECTED] > > > > > > > > -- > Thilina Gunarathne - http://thilinag.blogspot.com --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
