[5.0] CL delegation order (was: Re: [5][PATCH]Run Watchdog from the jakarta-tomcat-5 build.xml)

[Remy Maucherat <re...@apache.org>]

costinm@covalent.net wrote:
> I love classloaders and reverse delegation...

Speaking of which, I think we should stop using the delegation scheme 
recommended by the servlet spec (ie, look first in own repositories, 
then delegate). The idea is that it is only recommended, it is 
essentially broken and is creating a lot of problems.
I also do not believe that overriding shared libraries is a valid use 
case for the average Tomcat user.

So I propose that:
- the default for the webapp CL is to set "delegate" to "true"
- as before, the value of "delegate" can be changed in server.xml

<ballot>
+1 [ ] Yes, CL delegation should follow the standard model
-1 [ ] No, the webapp CL shouldn't delegate first, keeping the 
recommended model in the spec (which is convenient but broken)
</ballot>

Remy


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>

Re: [5.0] CL delegation order (was: Re: [5][PATCH]Run Watchdog from the jakarta-tomcat-5 build.xml)

[co...@covalent.net]

On Wed, 14 Aug 2002, Remy Maucherat wrote:

> costinm@covalent.net wrote:
> > I agree it is valuable in some cases to not delegate - but we've 
> > seen far too many problems. It's better to let those who 
> > really need this and know what they're doing turn this on.
> 
> The thing is that it looks like a terrific idea, but in the end it's not 
> so good once it is implemented.

Sometimes is more usefull to have fine control.
I think it may be better to just change the order - i.e. the loader
should default to 'delegate' for everything but be able to exclude
specific packages ( right now it is the reverse ). 

What's important is to support the MANIFEST entries - and for that
it may be needed to override some of the stuff from the common
loader. 

Now - if you are going to work on that, it may be a good idea to 
try to move the loader in a separate dir ( commons-style ) and 
merge the 2 loaders ( there's a lot of duplicated code ). I'm not
saying to move it in commons yet, but at least have it useable
independently. I would love to add few hooks and use it in ant...

Costin


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>

Re: [5.0] CL delegation order (was: Re: [5][PATCH]Run Watchdog from the jakarta-tomcat-5 build.xml)

[Remy Maucherat <re...@apache.org>]

costinm@covalent.net wrote:
> I agree it is valuable in some cases to not delegate - but we've 
> seen far too many problems. It's better to let those who 
> really need this and know what they're doing turn this on.

The thing is that it looks like a terrific idea, but in the end it's not 
so good once it is implemented.

(I have to apologize a bit for the tone of the previous email, which was 
mostly caused by months of frustration :-P)

> BTW, ant has also deprecated the 'reverse' loader, due to 
> too many problems.

Remy


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>

Re: [5.0] CL delegation order (was: Re: [5][PATCH]Run Watchdog from the jakarta-tomcat-5 build.xml)

[co...@covalent.net]

On Wed, 14 Aug 2002, Remy Maucherat wrote:

> costinm@covalent.net wrote:
> > I love classloaders and reverse delegation...
> 
> Speaking of which, I think we should stop using the delegation scheme 
> recommended by the servlet spec (ie, look first in own repositories, 
> then delegate). The idea is that it is only recommended, it is 
> essentially broken and is creating a lot of problems.
> I also do not believe that overriding shared libraries is a valid use 
> case for the average Tomcat user.
> 
> So I propose that:
> - the default for the webapp CL is to set "delegate" to "true"
> - as before, the value of "delegate" can be changed in server.xml
> 
> <ballot>
> +1 [X] Yes, CL delegation should follow the standard model
> -1 [ ] No, the webapp CL shouldn't delegate first, keeping the 
> recommended model in the spec (which is convenient but broken)
> </ballot>


I agree it is valuable in some cases to not delegate - but we've 
seen far too many problems. It's better to let those who 
really need this and know what they're doing turn this on.

BTW, ant has also deprecated the 'reverse' loader, due to 
too many problems.

Costin


> 
> Remy
> 
> 
> --
> To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
> For additional commands, e-mail: <ma...@jakarta.apache.org>
> 
> 


--
To unsubscribe, e-mail:   <ma...@jakarta.apache.org>
For additional commands, e-mail: <ma...@jakarta.apache.org>