- From: Nickolas Kavantzas <nickolas.kavantzas@oracle.com>
- Date: Tue, 3 May 2005 13:33:52 -0700
- To: <public-ws-chor@w3.org>
Forwarding to the public group the email thread regarding CDL and linearity (or absense of it) that happened between me, Kohei, et al. -- Nick ----- Original Message ----- From: "Nickolas Kavantzas" <nickolas.kavantzas@oracle.com> To: "Gary Brown" <gary@enigmatec.net>; "Kohei Honda" <kohei@dcs.qmul.ac.uk>; "Steve Ross-Talbot" <steve@enigmatec.net> Cc: "Nobuko Yoshida" <yoshida@doc.ic.ac.uk>; <kohei@dcs.qmul.ac.uk>; "Marco Carbone" <carbonem@dcs.qmul.ac.uk> Sent: Tuesday, April 12, 2005 11:17 AM Subject: Re: Linearity and Channel Passing. > Hi all, > > For CDL as I've designed it: > > 1) usage="unlimited" means no linearity constraints at all. This is like > normal pi. > > 2) In the case of usage="once", i had it open so one could mean different > types of > linearity as Kohei described below. > > Since, there is another CDL issue that asks for making usage an XML element > instead of > an XML attribute and i have designed the CDL language/schema to allow > extensibility one > could add other things as well, for example usage-output=2, usage-input=1, > etc. > > > Regards, > > -- > Nick > > ----- Original Message ----- > From: "Gary Brown" <gary@enigmatec.net> > To: "Kohei Honda" <kohei@dcs.qmul.ac.uk>; "Nickolas Kavantzas" > <nickolas.kavantzas@oracle.com>; "Steve Ross-Talbot" <steve@enigmatec.net> > Cc: "Nobuko Yoshida" <yoshida@doc.ic.ac.uk>; <kohei@dcs.qmul.ac.uk>; "Marco > Carbone" <carbonem@dcs.qmul.ac.uk> > Sent: Monday, April 11, 2005 12:52 AM > Subject: Re: Linearity and Channel Passing. > > > > Hi Kohei, > > > > Possibly the strongly linear pattern could be employed in the concise CDL > > calculus, as a way to represent the request response interaction in CDL. > > This way it does not need to be explicit in CDL, and therefore it would > not > > be obvious to users- but at the same time it would still be useful for the > > model checking. > > > > Does this mean that a way to model a CDL channel, that is used > continuously > > to send multiple request/response interactions, would be to use a > > persistently linear pattern to model the request channel, and strongly > > linear pattern to model the response channels? Therefore it would mean > that > > the request direction of a channel would be used multiple times, while the > > response direction of a channel would only be used once to respond to a > > particular request. However, even if this is the case, I think we should > be > > able to hide the two types of linearity in the tranformation from CDL to > the > > calculus representation. > > > > Regards > > Gary > > > > ----- Original Message ----- > > From: "Kohei Honda" <kohei_honda@ntlworld.com> > > To: "Gary Brown" <gary@enigmatec.net>; "Kohei Honda" > <kohei@dcs.qmul.ac.uk>; > > "Nickolas Kavantzas" <nickolas.kavantzas@oracle.com>; "Steve Ross-Talbot" > > <steve@enigmatec.net> > > Cc: "Nobuko Yoshida" <yoshida@doc.ic.ac.uk>; <kohei@dcs.qmul.ac.uk>; > "Marco > > Carbone" <carbonem@dcs.qmul.ac.uk> > > Sent: Monday, April 11, 2005 2:28 AM > > Subject: Re: Linearity and Channel Passing. > > > > > > > Hi Gary, > > > > > > Many thanks for your observations. I will reflect some of these points > > > in my post. Also what you implied about replies to multi-cast is quite > > > interesting --- Steve did mention about it, and hearing it twice makes > > > it very clear. Typing such usage (both behaviourally and syntactically) > > > would be fascinating. > > > > > > There are many aspects in types, but one of them is they give us > > > "hooks" using which we can do various static analysis, assuming that > > > a given type is honest --- that a program do obey the constraint via > > > its syntactic structure. As you wrote, the strong linearity is something > > > which has attracted attention among us, but other kinds of channel > > > usage also offer such hooks. > > > > > > One use of strong linearity in pi-processes is when we wish to represent > > > request-reply using only one-way message passing --- the reply channel > > > is usually used as a strongly linear channel (in combination with "new" > > > name generation). I do not know this can be potentially useful in CDL. > > > One possible way to use this would be, for example in compilation into > > > a terser form where we have only request, which may be easier to > > > implement: however CDL may not be used for such purposes. > > > > > > Anyway, I hope we can make CDL 1.0 as robust and well-digested as > > > possible during the remaining period (this is also a preparation for the > > > next stage, I believe). In particular, given its importance, I am happy > > > to participate in further dialogue on channel usage --- how it appears > in > > > real (and perhaps potential) business protocols, and how we can best > > > articulate it. > > > > > > Best wishes, > > > > > > kohei > > > > > > ----- Original Message ----- > > > From: "Gary Brown" <gary@enigmatec.net> > > > To: "Kohei Honda" <kohei@dcs.qmul.ac.uk>; "Nickolas Kavantzas" > > > <nickolas.kavantzas@oracle.com>; "Steve Ross-Talbot" > > > <steve@enigmatec.net> > > > Cc: "Nobuko Yoshida" <yoshida@doc.ic.ac.uk>; <kohei@dcs.qmul.ac.uk>; > > > "Marco Carbone" <carbonem@dcs.qmul.ac.uk> > > > Sent: Sunday, April 10, 2005 11:38 AM > > > Subject: Re: Linearity and Channel Passing. > > > > > > > > >> Hi Kohei > > >> > > >> I found your description of the issues very easy to understand, and I > do > > >> think that LTC is and should be the primary usage pattern for CDL at > the > > >> current time. > > >> > > >> I don't believe that strong linearity, although academically > > >> interesting/useful, adds any value to CDL as I don't believe users will > > >> understand it when defining business protocols. It may be possible that > > >> higher level tools can hide this level of detail, but then they will > > >> probably be presenting the persistently linear style pattern to the > user, > > >> so > > >> makes me wonder whether it is useful. > > >> > > >> The third (server linearity) pattern sounds useful in multi-cast > > >> scenarios. > > >> However, from my understanding of your note, this would have to be a > > >> uni-directional channel - i.e. the exchanges on those channels would > only > > >> consist of requests, with no responses. > > >> > > >> Therefore, in the current version of CDL (where broadcast/multicast is > > >> not > > >> supported/discussed), I think pattern (2) is the only one that is > > >> relevant. > > >> When we tackle (3), it will need to include a description of the > > >> restrictions on the usage of that type of channel. > > >> > > >> Regards > > >> Gary > > >> > > >> ----- Original Message ----- > > >> From: "Kohei Honda" <kohei_honda@ntlworld.com> > > >> To: "Nickolas Kavantzas" <nickolas.kavantzas@oracle.com>; "Steve > > >> Ross-Talbot" <steve@enigmatec.net>; "Gary Brown" <gary@enigmatec.net> > > >> Cc: "Nobuko Yoshida" <yoshida@doc.ic.ac.uk>; <kohei@dcs.qmul.ac.uk>; > > >> "Marco > > >> Carbone" <carbonem@dcs.qmul.ac.uk> > > >> Sent: Thursday, April 07, 2005 12:01 AM > > >> Subject: Linearity and Channel Passing. > > >> > > >> > > >> > Dear Nick, Steve and Gary, > > >> > > > >> > Here is my tentative analysis on linearity. I thank you three to give > > >> > me > > >> > a chance to analyse this point, which would touche many other aspects > > >> > of CDL. > > >> > > > >> > I will be very happy to receive your views on this. I do not post > this > > >> > to > > >> > the list directly since I fear it can be a bit too technical for all > > >> > others --- > > >> > though I try to give as simple illustration as possible. At least I > may > > >> > have > > >> > to add some illustration on process notation I used. > > >> > > > >> > Reflecting your ideas, it may be posted to the public list, since > this > > >> > is > > >> > an > > >> > interesting technical point which may deserve recording as well as > > >> > asking > > >> > for others' opinions. > > >> > > > >> > As I wrote, your views on this analysis are warmly welcome. In > > >> > particular, > > >> > I wonder if what I called LTC is correct or not... > > >> > > > >> > Best wishes, > > >> > > > >> > kohei > > >> > > > >> > > > >> > > >> > > > > > > > > > >
Received on Tuesday, 3 May 2005 20:34:05 UTC