You are viewing a plain text version of this content. The canonical link for it is here.
Posted to server-dev@james.apache.org by Vincenzo Gianferrari Pini <vi...@praxis.it> on 2003/05/27 17:54:14 UTC

[no virus] New Redirect & sons becoming available

Noel,

I'm attaching to this message the code of the new Redirect and NotifySender mailets; the others (NotifyPostmaster, Forward and the new Bounce) will follow in the next days, and should be straightforward.

I have tested thoroughly (at least I put my best efforts), and they seem to work quite well.

They should also be well (?) documented using javadoc: it is my intention though to add a special "notes for implementors" section with guidelines on how to extend Redirect and override its protected methods.

A few notes here:

---------------------

1) I have followed (at least I hope so) the guidelines discussed among us (all the getX and setX stuff).

1.1 The service(Mail) code begins creating a new Mail object (called all around "newMail") calling duplicate(getMailetContext().getId()) on the original Mail object, called all around "originalMail".

1.2 Then, based on the inline type being UNALTERED or not, and based on passThrough being false or true, the newMail object will either share the MimeMessage with originalMail, or embed a copy of it, or embed a brand new MimeMessage to build. The detailed code here is taken (and modified) from the NotifySender code patched by you for James version 2.1.3.

1.3 After that, the various setX methods are called to modify newMail using the outcome of getX(originalMail) .

1.4 Then a newMail.getMessage().saveChanges() is issued and afterwards, if the inline type is UNALTERED, the message id of newMail is reset to the message id of the originalMail, otherwise it gets a new message id.

1.5 Finally, as newMail is ready, a getMailetContext().sendMail(newMail) is issued, and passThrough processing is done. Note here that, as sendMail(Mail) is not defined in the MailetContext interface, although is implemented in the James class, I had to modify MailetContext: I didn't like to code ((James) getMailetContext()).sendMail(newMail), and supposed that it was simply forgotten. Also the getId() method needed above (see 1.1) was missing in MailetContext, so I added it too. If it is not OK let me know what to do.

2) A slight departure from the agreed "protocol": some setX methods (setSender when sender is null, setSubjectPrefix and the new setIsReply) need to access also some other information from originalMail (it is safer than getting them from newMail, because the order could matter), so I had to change setX(Mail, Tx) to setX(Mail, Tx, Mail), the first Mail being the newMail object to modify, and the second one being originalMail to consult if and when needed. So the calls are:

setX(newMail, getX(originalMail), originalMail);

As I foresee such need for other subclasses, I used this third parameter in all setX methods. It simply doesn't hurt.

3) To define IDENTITY constant MailAddress-es, that I use to "late bind" the "magic" values, I inserted them in the *MailAddress class* as follows:
    public static final MailAddress SENDER      = new MailAddress();
    ...

and are referenceable as MailAddress.SENDER, MailAddress.UNALTERED etc. My choice to put such constants in the MailAddress class was to make them available for future usage, and to overcome a "throws ParseException" declared in the various constructors of MailAddress. To do that I defined a new private constructor MailAddress() (see the code). If you don't like this solution let me know.

4) I carefully wrote Redirect in order to not (hopefully) break any subclass that may have been already written; at the same time I adopted a rule of behaviour that, when followed by implementors of new subclasses (as I did for NotifySender), would allow to set always the static parameter to true, and always return isStatic() as true, and having the performance advantages of static but with the flexibility of dynamic.

4.1 As discussed in previous emails in this list, the rule is to separate the roles of getX() and getX(Mail) *exactly* based on the first being static and the second being dynamic. So, following it, I can set static always to true, even working with "magic" values both in Redirect and in new subclasses (see new NotifySender). Any existing subclass of Redirect, if has used a getX() dynamically, just needs, as before, to set static to false, and will continue to work as before.

5) I also carefully wrote Redirect in order to (i) continue to behave as before (even looking at the text generated) when defined in an existing config.xml, so it is not breaking existing usages, and to (ii) allow the same for existing mailets that will become a subclass of it, like NotifySender.

5.1 At the same time I've tried to have things done in order to allow Redirect to be anytime a substitution of any of it subclasses, at least logically although not generating the same text in the messages, just by coding appropriately its init parameters in config.xml. To do that I added, and documented, some new parameters and "magic" values (and related getX() methods) that will default to the old behaviour: <returnPath> and <isReply>. I added also new values to existing parameters, like <to>unaltered</to> asked by Hontvari Jozsef. I think that this goal should always be pursued, and I will add new parameters etc. to Redirect if the need arises while coding the new subclasses.


Nothing else comes to my mind now.

-------------------

As said, I tested the code and I think it is OK, but obviously more people should do it before releasing.

I'm also expecting comments from you and the others, that hopefully will not force me to recode too much :-)

In the meantime I will code the other subclasses, hoping to have them ready in a few days. I need though ASAP an OK from you and others to proceed.

Thanks

Vincenzo

RE: [no virus] New Redirect & sons becoming available

Posted by "Noel J. Bergman" <no...@devtech.com>.
Vincenzo,

I've started to look over the code.  Looks good so far.  A few apparently
redundant methods, and some typos, but all in all it looks good.

With the exception of the NULL address, I don't think that the rest belong
in the general MailAddress class.  They are specific to the processing done
by Redirect and subclasses.  That's why at one point I'd had something like:

class Redirect {
   ...
   class MailAddress extends org.apache.mailet.MailAddress {
   ...
   }
   ...
}

Also, you called it EMPTY.  I thought that we had agreed to call it NULL,
since it is referred to as a null sender in the RFC documents.

As I said, generally the work looks really good so far.

	--- Noel

-----Original Message-----
From: Vincenzo Gianferrari Pini
[mailto:vincenzo.gianferraripini@praxis.it]
Sent: Tuesday, May 27, 2003 11:54
To: List James-Dev
Subject: [no virus] New Redirect & sons becoming available


Noel,

I'm attaching to this message the code of the new Redirect and NotifySender
mailets; the others (NotifyPostmaster, Forward and the new Bounce) will
follow in the next days, and should be straightforward.

I have tested thoroughly (at least I put my best efforts), and they seem to
work quite well.

They should also be well (?) documented using javadoc: it is my intention
though to add a special "notes for implementors" section with guidelines on
how to extend Redirect and override its protected methods.

A few notes here:

---------------------

1) I have followed (at least I hope so) the guidelines discussed among us
(all the getX and setX stuff).

1.1 The service(Mail) code begins creating a new Mail object (called all
around "newMail") calling duplicate(getMailetContext().getId()) on the
original Mail object, called all around "originalMail".

1.2 Then, based on the inline type being UNALTERED or not, and based on
passThrough being false or true, the newMail object will either share the
MimeMessage with originalMail, or embed a copy of it, or embed a brand new
MimeMessage to build. The detailed code here is taken (and modified) from
the NotifySender code patched by you for James version 2.1.3.

1.3 After that, the various setX methods are called to modify newMail using
the outcome of getX(originalMail) .

1.4 Then a newMail.getMessage().saveChanges() is issued and afterwards, if
the inline type is UNALTERED, the message id of newMail is reset to the
message id of the originalMail, otherwise it gets a new message id.

1.5 Finally, as newMail is ready, a getMailetContext().sendMail(newMail) is
issued, and passThrough processing is done. Note here that, as
sendMail(Mail) is not defined in the MailetContext interface, although is
implemented in the James class, I had to modify MailetContext: I didn't like
to code ((James) getMailetContext()).sendMail(newMail), and supposed that it
was simply forgotten. Also the getId() method needed above (see 1.1) was
missing in MailetContext, so I added it too. If it is not OK let me know
what to do.

2) A slight departure from the agreed "protocol": some setX methods
(setSender when sender is null, setSubjectPrefix and the new setIsReply)
need to access also some other information from originalMail (it is safer
than getting them from newMail, because the order could matter), so I had to
change setX(Mail, Tx) to setX(Mail, Tx, Mail), the first Mail being the
newMail object to modify, and the second one being originalMail to consult
if and when needed. So the calls are:

setX(newMail, getX(originalMail), originalMail);

As I foresee such need for other subclasses, I used this third parameter in
all setX methods. It simply doesn't hurt.

3) To define IDENTITY constant MailAddress-es, that I use to "late bind" the
"magic" values, I inserted them in the *MailAddress class* as follows:
    public static final MailAddress SENDER      = new MailAddress();
    ...

and are referenceable as MailAddress.SENDER, MailAddress.UNALTERED etc. My
choice to put such constants in the MailAddress class was to make them
available for future usage, and to overcome a "throws ParseException"
declared in the various constructors of MailAddress. To do that I defined a
new private constructor MailAddress() (see the code). If you don't like this
solution let me know.

4) I carefully wrote Redirect in order to not (hopefully) break any subclass
that may have been already written; at the same time I adopted a rule of
behaviour that, when followed by implementors of new subclasses (as I did
for NotifySender), would allow to set always the static parameter to true,
and always return isStatic() as true, and having the performance advantages
of static but with the flexibility of dynamic.

4.1 As discussed in previous emails in this list, the rule is to separate
the roles of getX() and getX(Mail) *exactly* based on the first being static
and the second being dynamic. So, following it, I can set static always to
true, even working with "magic" values both in Redirect and in new
subclasses (see new NotifySender). Any existing subclass of Redirect, if has
used a getX() dynamically, just needs, as before, to set static to false,
and will continue to work as before.

5) I also carefully wrote Redirect in order to (i) continue to behave as
before (even looking at the text generated) when defined in an existing
config.xml, so it is not breaking existing usages, and to (ii) allow the
same for existing mailets that will become a subclass of it, like
NotifySender.

5.1 At the same time I've tried to have things done in order to allow
Redirect to be anytime a substitution of any of it subclasses, at least
logically although not generating the same text in the messages, just by
coding appropriately its init parameters in config.xml. To do that I added,
and documented, some new parameters and "magic" values (and related getX()
methods) that will default to the old behaviour: <returnPath> and <isReply>.
I added also new values to existing parameters, like <to>unaltered</to>
asked by Hontvari Jozsef. I think that this goal should always be pursued,
and I will add new parameters etc. to Redirect if the need arises while
coding the new subclasses.


Nothing else comes to my mind now.

-------------------

As said, I tested the code and I think it is OK, but obviously more people
should do it before releasing.

I'm also expecting comments from you and the others, that hopefully will not
force me to recode too much :-)

In the meantime I will code the other subclasses, hoping to have them ready
in a few days. I need though ASAP an OK from you and others to proceed.

Thanks

Vincenzo


---------------------------------------------------------------------
To unsubscribe, e-mail: james-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: james-dev-help@jakarta.apache.org