Some comments inline.
Thanks,
Raymond
----- Original Message -----
From: "ant elder" <[EMAIL PROTECTED]>
To: <[email protected]>
Cc: "Scott Kurz" <[EMAIL PROTECTED]>
Sent: Monday, September 10, 2007 12:07 PM
Subject: Re: Avoiding unnecessary pass-by-value copies
The problem with the flag is that that implementing that also isn't
without
issues, especially at this late stage in our release cycle, so i've been
looking for a simpler way that solves most cases. Comments inline...
...ant
On 9/10/07, Scott Kurz <[EMAIL PROTECTED]> wrote:
Apologies for opening TUSCANY-1678 without noticing the earlier JIRA or
this
discussion.
On 9/7/07, Jean-Sebastien Delfino <[EMAIL PROTECTED]> wrote:
>
>
> I didn't quite get Raymond's comment that we'll have to pass a flag in
> a
> message header. Don't we know what to do w.r.t pass by value just by
> looking at the source and target interfaces? Can't we just decide to
> either insert the interceptor in the invocation chain or not when the
> invocation chain is built?
One problem with deciding when the invocation chain is built is
supporting
the case where a service-side binding
impl accepts requests from either local or remote-JVM clients (the 2nd
case
in TUSCANY-1678). The local client may
require a copy while not the remote one.
Not sure exactly what to do about that case yet, but it does seem like
those
are two separate paths through the binding and so it should be possible to
set some flag somewhere at build time not just at invocation time. Maybe
by
cloning the Interface and flipping the isRemotable flag to trick the copy
code or something like that, i'll have a play around to see if i can find
something but say if you've any ideas.
The ServiceBindingProvider probably know the data is passed over a transport
that uses the pass-by-value semantics. When it delegates the inbound request
to the exposed component, it could help to make the decision. That's why I'm
thinking of a flag or header on the messge to carry such informaiton.
Also are we sure the presence of a DataTransformationInteceptor on the
chain
means a copy has been done? Is that
part of the SPI contract? Say you had a databinding which transformed
Java complex types from one system to another but
did nothing with a simple java.lang.Integer. Is it established that the
databinding impl should do a copy in that case?
Almost all of the data transformation we have so far will result in a copy
of the orginal data. For simple java types that are immutable, no copy is
required.
How about adding a method to the data binding interface to say whether or
not the transformation it does results in a copy occurring?
...ant
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
