Jaliya Ekanayake wrote:
Hi Reinecke,
Yes, it is a problem that we need to solve. According to the specification
the maximum of the message number that is allowed is 18446744073709551615
which is the max value of the "unsigned long" in some languages.
However when we try to implement in java, we cannot just use some base
types. The problem starts from this kind of situation (If this ever
happens...)
When we read the message number from SOAP initially we have to get it as
String and then get the special type that we need. Here we are in trouble.
We need the functionality of some form of "number" format since we need to
perform increments, checks etc... On this number.
Unfortunately we cannot use double, since it round some parts off from the
number and hence the final value is inconsistent.
So the possibility is to use some form of utility class that either dose
some string processing or string processing + long.
At the moment we are trying our best to pass the interops with other
implementations so if you can derive a solution and send a patch it is
really great.
why not use BigInteger ? no need to reinvent wheel with another
utility class to do arithmetic on big numbers - moreover it is used in
crypto and is very optimized (and i think using native C code):
http://java.sun.com/j2se/1.4.2/docs/api/java/math/BigInteger.html
HTH,
alek
-----Original Message-----
From: Philipp Reinecke [mailto:[EMAIL PROTECTED]]
Sent: Wednesday, April 27, 2005 7:40 PM
To: [email protected]
Subject: Retry: Possibly stupid question regarding msgNumber in Sandesha
Hello,
as I had sent the attached message before, I have been wondering why
I did not receive any replies to it. Only now did I realize that it
did not show up in the archives, either, which seems strange since I
got a copy sent by the list software.
So, can one of the Sandesha developers help me understand/resolve the
issue mentioned below? I am looking at a different way of handling
acknowledgements, and this detail caused some major confusion.
Thanks,
Philipp Reinecke
----- Forwarded message from Philipp Reinecke
<[EMAIL PROTECTED]> -----
[..]
Hello,
I noticed that Sandesha internally uses variables of type `long'
to store the RM message number. According to the JDK API Docu-
mentation these have maximum value of 2^63-1. The WSRM, however,
specifies that message numbers can be in the range from 1 up to
2^64-1, which is quite a bit larger than 2^63-1. Does Sandesha
magically handle the (hypothetical) case of more than 2^63-1
messages in one sequence by utilizing the negative portion of
the long number space, or is this a bug? Judging from the way
MessageNumber.toSOAPEnvelope() works, I lean towards the latter,
since the internal message number is exported without checking
whether it is negative.
Regards,
Philipp Reinecke
----- End forwarded message -----
--
The best way to predict the future is to invent it - Alan Kay
|
