On Feb 10, 2010, at 2:08 PM, Emmanuel Lecharny wrote:
On 2/10/10 10:42 PM, Alan D. Cabrera wrote:
On Feb 4, 2010, at 1:48 AM, Emmanuel Lecharny wrote:
Hi,
some more thoughts, as I'd like to define precisely what is an
IoService.
Looking at the existing code, I would define an IoService as a
base Interface for Acceptor and Connector, describing the
relationship between all their components, namely :
- the chain
- the handler
- the configuration
- a state (active/not active, number of sessions, is the service
is disposed, or being disposed...)
- the write log (messages waiting to be written to the client)
I'm not convinced that the write log accessors should be a
separate component. In fact, I would rather see that as a part of
the service's state.
Is that correct? Fo I miss something here ?
Also there is some strange method present in this interface, like
broadcast(). I'm not sure it should be a part of the IoService
interface, but rather moved to IoAcceptor (does it make sense for
a Connector to boradcast a message ?)
I'm not a big fan of tightly coupling the Acceptor and Connector
via a base interface IoService unless there's a concrete and
compelling use case where I would want to interchangeably use either.
The thing is that they are using the exact same mechanisms, the only
big difference is that one initiate the connection, the other one
accept them. But we may see more difference if we look in detail.
I think that I can correctly restate your words as:
The methods in the parent interface, IoService, that would be shared
by the Acceptor and Connector have the same semantics. This is why it
makes sense to me to push the methods into IoService and have them
shared.
If my re-wording is correct then I am going to still hold on to my
point. An interface should not be used as an indicator for
implementation. Which, imo, is the justification you give. To my
mind an Acceptor and Connector live in two different worlds and one
would never pass around both willy nilly using the base interface.
Maybe there's a use case for doing so. I cannot think of one off the
top of my head.
I also think that the modus operandi for the API redesign is to
start w/ apps first, generate interfaces that support those apps,
then fill in the guts. Starting with the existing API tends to
limit the possible ideas and one tends to accidentally bring in
some of the less worthwhile bits.
Here, I would slightly disagree, if we want to stay close t what we
currently have to ease migration.
Imho, the vein of ideas that influenced v1.x and v2.x has run its
course. v2.x was an attempt to "remedy" the issues of v1.x but, imho,
things got way more complicated and bloated. This is why I was
thinking of a radical departure.
If you want to keep backwards compatibility then why not do it in the
2.x branch?
I invite people to look at my HTTP code in my sandbox where I am
sketching out some ideas of a possible mina 3 API. It's a sketch
and there's a ton of bits that have not even been thought off. I
welcome your comments.
Sure, will do.
Many thanks. Please be genteel with your comments. ;)
Regards,
Alan
