1 Mar
2012
1 Mar
'12
6:20 p.m.
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
1 Mar
1 Mar
9:54 p.m.
+1
Best regards,
Andreas
2012-03-01 18:20, Thomas Mortagne skrev:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
10:19 p.m.
On Thu, Mar 1, 2012 at 18:20, Thomas Mortagne <thomas.mortagne(a)xwiki.com>wrote;wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
+1, Sounds good and it is effectively the right time to do so.
(by the way, a very clear explanation of the issue, great mail, pleasant to
read)
--
Thomas Mortagne
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs
--
Denis Gervalle
SOFTEC sa - CEO
eGuilde sarl - CTO
10:33 p.m.
On Mar 1, 2012, at 6:20 PM, Thomas Mortagne wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
+1 too
So we could summarize all this in JIRA by creating an issue that says "Add support
for generics in Component injection".
Thanks
-Vincent
2 Mar
2 Mar
12:19 p.m.
+1
Thanks,
Marius
On Thu, Mar 1, 2012 at 7:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs
5 Mar
5 Mar
5:32 p.m.
+1
Thanks,
Eduard
On Fri, Mar 2, 2012 at 1:19 PM, Marius Dumitru Florea <
mariusdumitru.florea(a)xwiki.com> wrote:
+1
Thanks,
Marius
On Thu, Mar 1, 2012 at 7:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs
6:53 p.m.
Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
--
Thomas Mortagne
6 Mar
6 Mar
2:51 p.m.
It's committed.
Would be nice to have comments on the new APIs (suggest better naming,
etc...). It's a very important API and we really need to have
something we are sure of before 4.0 final.
On Mon, Mar 5, 2012 at 6:53 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
--
Thomas Mortagne
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
--
Thomas Mortagne
11 Mar
11 Mar
9:05 p.m.
Hi,
We must consider how we define the equality of RoleHints. Different
implementations of the interface ParameterizedType do not need to
implement equals and hashValue consistently. For instance, this lookup
doesn't work at the moment:
Utils.getComponent(new ParameterizedType() {
@Override
public Type [] getActualTypeArguments() {
Type [] args = {
String.class
};
return args;
}
@Override
public Type getOwnerType() {
return null;
}
@Override
public Type getRawType() {
return DocumentReferenceResolver.class;
}
});
At the same time, there is the backwards compliancy issue. This might be
ok for RoleHint equality:
/**
* Defines the equality of types.
*
* @param t1 operand for comparison.
* @param t2 operand for comparison.
* @return {\code true} if the roles should be considered equal.
*/
private boolean typeEquals(Type t1, Type t2) {
if (t1 instanceof ParameterizedType && t2 instanceof
ParameterizedType) {
ParameterizedType pt1 = (ParameterizedType) t1;
ParameterizedType pt2 = (ParameterizedType) t2;
Type [] args1 = pt1.getActualTypeArguments();
Type [] args2 = pt2.getActualTypeArguments();
if (args1.length != args2.length) {
return false;
}
for (int i = 0; i < args1.length; i++) {
if (!typeEquals(args1[i], args2[i])) {
return false;
}
}
} else if ((t1 instanceof Class && !(t2 instanceof Class))
|| (t2 instanceof Class && !(t1 instanceof Class))) {
/*
* For compliancy with legacy code.
*
* Looking up components using raw classes only
* matches components registered as raw classes, and a
* component registered as raw class can only be
* looked up using a raw class.
*/
return false;
}
Class c1 = ReflectionUtils.getTypeClass(t1);
Class c2 = ReflectionUtils.getTypeClass(t2);
return c1.equals(c2);
}
@Override
@SuppressWarnings("unchecked")
public boolean equals(Object obj)
{
if (this == obj) {
return true;
}
if ((obj == null) || (obj.getClass() != this.getClass())) {
return false;
}
RoleHint<T> rolehint = (RoleHint<T>) obj;
return typeEquals(getRoleType(), rolehint.getRoleType())
&& (getHint() == rolehint.getHint() || (getHint() != null &&
getHint().equals(rolehint.getHint())));
}
@Override
public int hashCode()
{
int hash = 8;
hash = 31 * hash + (null == getRoleType() ? 0 :
ReflectionUtils.getTypeClass(getRoleType()).hashCode());
hash = 31 * hash + (null == getHint() ? 0 : getHint().hashCode());
return hash;
}
/Andreas
2012-03-06 14:51, Thomas Mortagne skrev:
It's committed.
Would be nice to have comments on the new APIs (suggest better naming,
etc...). It's a very important API and we really need to have
something we are sure of before 4.0 final.
On Mon, Mar 5, 2012 at 6:53 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
--
Thomas Mortagne
9:20 p.m.
Maybe we should also think a little bit more if 'Type' really is the
right interface to use. Both Guice and CDI defines 'TypeLiteral'
classes. Maybe we should do the same?
Best Regards,
/Andreas
2012-03-11 21:05, Andreas Jonsson skrev:
Hi,
We must consider how we define the equality of RoleHints. Different
implementations of the interface ParameterizedType do not need to
implement equals and hashValue consistently. For instance, this lookup
doesn't work at the moment:
Utils.getComponent(new ParameterizedType() {
@Override
public Type [] getActualTypeArguments() {
Type [] args = {
String.class
};
return args;
}
@Override
public Type getOwnerType() {
return null;
}
@Override
public Type getRawType() {
return DocumentReferenceResolver.class;
}
});
At the same time, there is the backwards compliancy issue. This might be
ok for RoleHint equality:
/**
* Defines the equality of types.
*
* @param t1 operand for comparison.
* @param t2 operand for comparison.
* @return {\code true} if the roles should be considered equal.
*/
private boolean typeEquals(Type t1, Type t2) {
if (t1 instanceof ParameterizedType && t2 instanceof
ParameterizedType) {
ParameterizedType pt1 = (ParameterizedType) t1;
ParameterizedType pt2 = (ParameterizedType) t2;
Type [] args1 = pt1.getActualTypeArguments();
Type [] args2 = pt2.getActualTypeArguments();
if (args1.length != args2.length) {
return false;
}
for (int i = 0; i < args1.length; i++) {
if (!typeEquals(args1[i], args2[i])) {
return false;
}
}
} else if ((t1 instanceof Class && !(t2 instanceof Class))
|| (t2 instanceof Class && !(t1 instanceof Class))) {
/*
* For compliancy with legacy code.
*
* Looking up components using raw classes only
* matches components registered as raw classes, and a
* component registered as raw class can only be
* looked up using a raw class.
*/
return false;
}
Class c1 = ReflectionUtils.getTypeClass(t1);
Class c2 = ReflectionUtils.getTypeClass(t2);
return c1.equals(c2);
}
@Override
@SuppressWarnings("unchecked")
public boolean equals(Object obj)
{
if (this == obj) {
return true;
}
if ((obj == null) || (obj.getClass() != this.getClass())) {
return false;
}
RoleHint<T> rolehint = (RoleHint<T>) obj;
return typeEquals(getRoleType(), rolehint.getRoleType())
&& (getHint() == rolehint.getHint() || (getHint() != null &&
getHint().equals(rolehint.getHint())));
}
@Override
public int hashCode()
{
int hash = 8;
hash = 31 * hash + (null == getRoleType() ? 0 :
ReflectionUtils.getTypeClass(getRoleType()).hashCode());
hash = 31 * hash + (null == getHint() ? 0 : getHint().hashCode());
return hash;
}
/Andreas
2012-03-06 14:51, Thomas Mortagne skrev:
It's committed.
Would be nice to have comments on the new APIs (suggest better naming,
etc...). It's a very important API and we really need to have
something we are sure of before 4.0 final.
On Mon, Mar 5, 2012 at 6:53 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
--
Thomas Mortagne
12 Mar
12 Mar
8:59 a.m.
For now I provided a org.xwiki.component.util.DefaultParameterizedType
which is implemented like the one in the oracle JVM and which is what
I seen other libraries that need custom ParameterizedType doing. Works
pretty well so far with both orache JVM and OpenJDK. At worst if
someone report an issue with another JVM I have a backup plan which is
to change any ParameterizedType provided to EmbeddableComponentManager
in our DefaultParameterizedType but I would prefer to avoid that if
not absolutely necessary.
Also note that the issue is only when you lookup a component, you
can't have this kind of issue with injection. We don't have that much
ParameterizedType, only reference resolver/serializer plus the
Provider so it's not a very commons task either.
On Sun, Mar 11, 2012 at 9:05 PM, Andreas Jonsson <aj(a)member.fsf.org> wrote:
Hi,
We must consider how we define the equality of RoleHints. Different
implementations of the interface ParameterizedType do not need to
implement equals and hashValue consistently. For instance, this lookup
doesn't work at the moment:
Actually they do need to implement equals the same way since that's
part of the specs but there is no spec for hashCode yes.
Utils.getComponent(new ParameterizedType() {
@Override
public Type [] getActualTypeArguments() {
Type [] args = {
String.class
};
return args;
}
@Override
public Type getOwnerType() {
return null;
}
@Override
public Type getRawType() {
return DocumentReferenceResolver.class;
}
});
You don't need to do that, DocumentReferenceResolver provide to
statics Type to make easier to lookup it.
Utils.getComponent(DocumentReferenceResolver.TYPE_STRING)
At the same time, there is the backwards compliancy
issue. This might be
ok for RoleHint equality:
/**
* Defines the equality of types.
*
* @param t1 operand for comparison.
* @param t2 operand for comparison.
* @return {\code true} if the roles should be considered equal.
*/
private boolean typeEquals(Type t1, Type t2) {
if (t1 instanceof ParameterizedType && t2 instanceof
ParameterizedType) {
ParameterizedType pt1 = (ParameterizedType) t1;
ParameterizedType pt2 = (ParameterizedType) t2;
Type [] args1 = pt1.getActualTypeArguments();
Type [] args2 = pt2.getActualTypeArguments();
if (args1.length != args2.length) {
return false;
}
for (int i = 0; i < args1.length; i++) {
if (!typeEquals(args1[i], args2[i])) {
return false;
}
}
} else if ((t1 instanceof Class && !(t2 instanceof Class))
|| (t2 instanceof Class && !(t1 instanceof Class))) {
/*
* For compliancy with legacy code.
*
* Looking up components using raw classes only
* matches components registered as raw classes, and a
* component registered as raw class can only be
* looked up using a raw class.
*/
return false;
}
Class c1 = ReflectionUtils.getTypeClass(t1);
Class c2 = ReflectionUtils.getTypeClass(t2);
return c1.equals(c2);
}
@Override
@SuppressWarnings("unchecked")
public boolean equals(Object obj)
{
if (this == obj) {
return true;
}
if ((obj == null) || (obj.getClass() != this.getClass())) {
return false;
}
RoleHint<T> rolehint = (RoleHint<T>) obj;
return typeEquals(getRoleType(), rolehint.getRoleType())
&& (getHint() == rolehint.getHint() || (getHint() != null &&
getHint().equals(rolehint.getHint())));
}
@Override
public int hashCode()
{
int hash = 8;
hash = 31 * hash + (null == getRoleType() ? 0 :
ReflectionUtils.getTypeClass(getRoleType()).hashCode());
hash = 31 * hash + (null == getHint() ? 0 : getHint().hashCode());
return hash;
}
/Andreas
2012-03-06 14:51, Thomas Mortagne skrev:
--
Thomas Mortagne
It's committed.
Would be nice to have comments on the new APIs (suggest better naming,
etc...). It's a very important API and we really need to have
something we are sure of before 4.0 final.
On Mon, Mar 5, 2012 at 6:53 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
--
Thomas Mortagne
4 Apr
4 Apr
8 p.m.
Hi Thomas/All,
All looks good except CM.lookupComponent() which Ï don't like a lot for the following
reasons:
* It's a bit long. It's longer than before and it would be nicer if it were as
short as before or even shorter ;)
* It's not symmetrical with other lookups like lookupList and lookupMap, which should
theoretically be lookupComponentList() and lookupComponentMap() which are even longer
Thus I'd prefer that the new API for lookup be named differently.
I propose to use "resolve":
CM.resolve(Type)
CM.resolveList(Type)
CM.resolveMap(Type)
Alternative:
CM.getInstance(Type)
CM.getInstanceList(Type)
CM.getInstanceMap(Type)
or even the short:
CM.get(Type)
CM.getList(Type)
CM.getMap(Type)
For information Guice uses getInstance(), Picocontainer uses getComponentInstance().
WDYT?
My preference goes to getInsta nce():
* users are tempted by default to start typing "get" to get an instance and then
they'll see getInstance in autocompletion and choose it.
* it's the same as Guice
Thanks
-Vincent
On Mar 6, 2012, at 2:51 PM, Thomas Mortagne wrote:
It's committed.
Would be nice to have comments on the new APIs (suggest better naming,
etc...). It's a very important API and we really need to have
something we are sure of before 4.0 final.
On Mon, Mar 5, 2012 at 6:53 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
--
Thomas Mortagne
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
--
Thomas Mortagne
5 Apr
5 Apr
2:28 a.m.
On Wed, Apr 4, 2012 at 20:00, Vincent Massol <vincent(a)massol.net> wrote:
Hi Thomas/All,
All looks good except CM.lookupComponent() which Ï don't like a lot for
the following reasons:
* It's a bit long. It's longer than before and it would be nicer if it
were as short as before or even shorter ;)
* It's not symmetrical with other lookups like lookupList and lookupMap,
which should theoretically be lookupComponentList() and
lookupComponentMap() which are even longer
Thus I'd prefer that the new API for lookup be named differently.
I propose to use "resolve":
CM.resolve(Type)
CM.resolveList(Type)
CM.resolveMap(Type)
-0
Alternative:
CM.getInstance(Type)
CM.getInstanceList(Type)
CM.getInstanceMap(Type)
+1
or even the short:
CM.get(Type)
CM.getList(Type)
CM.getMap(Type)
+0
For information Guice uses getInstance(), Picocontainer uses
getComponentInstance().
WDYT?
My preference goes to getInsta nce():
* users are tempted by default to start typing "get" to get an instance
and then they'll see getInstance in autocompletion and choose it.
* it's the same as Guice
Thanks
-Vincent
On Mar 6, 2012, at 2:51 PM, Thomas Mortagne wrote:
--
Denis Gervalle
SOFTEC sa - CEO
eGuilde sarl - CTO
It's committed.
Would be nice to have comments on the new APIs (suggest better naming,
etc...). It's a very important API and we really need to have
something we are sure of before 4.0 final.
On Mon, Mar 5, 2012 at 6:53 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs
Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
--
Thomas Mortagne
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
--
Thomas Mortagne 
2:45 a.m.
On 04/04/2012 02:00 PM, Vincent Massol wrote:
Hi Thomas/All,
All looks good except CM.lookupComponent() which Ï don't like a lot for the following
reasons:
* It's a bit long. It's longer than before and it would be nicer if it were as
short as before or even shorter ;)
* It's not symmetrical with other lookups like lookupList and lookupMap, which should
theoretically be lookupComponentList() and lookupComponentMap() which are even longer
Thus I'd prefer that the new API for lookup be named differently.
I propose to use "resolve":
CM.resolve(Type)
CM.resolveList(Type)
CM.resolveMap(Type)
Alternative:
CM.getInstance(Type)
CM.getInstanceList(Type)
CM.getInstanceMap(Type)
or even the short:
CM.get(Type)
CM.getList(Type)
CM.getMap(Type)
For information Guice uses getInstance(), Picocontainer uses getComponentInstance().
WDYT?
My preference goes to getInstance():
* users are tempted by default to start typing "get" to get an instance and
then they'll see getInstance in autocompletion and choose it.
* it's the same as Guice
Same, +1 for getInstance.
--
Sergiu Dumitriu
http://purl.org/net/sergiu/
8:35 a.m.
On Wed, Apr 4, 2012 at 8:00 PM, Vincent Massol <vincent(a)massol.net> wrote:
Hi Thomas/All,
All looks good except CM.lookupComponent() which Ï don't like a lot for the following
reasons:
* It's a bit long. It's longer than before and it would be nicer if it were as
short as before or even shorter ;)
* It's not symmetrical with other lookups like lookupList and lookupMap, which should
theoretically be lookupComponentList() and lookupComponentMap() which are even longer
Thus I'd prefer that the new API for lookup be named differently.
I propose to use "resolve":
CM.resolve(Type)
CM.resolveList(Type)
CM.resolveMap(Type)
Alternative:
CM.getInstance(Type)
CM.getInstanceList(Type)
CM.getInstanceMap(Type)
or even the short:
CM.get(Type)
CM.getList(Type)
CM.getMap(Type)
For information Guice uses getInstance(), Picocontainer uses getComponentInstance().
WDYT?
"get" sounds a bit lightweight to be for something supposed to also
initialize what you ask but if Guice use that I guess it means it's
OK.
I'm OK with any of theses suggestions so I will apply the voted one.
My preference goes to getInsta nce():
* users are tempted by default to start typing "get" to get an instance and
then they'll see getInstance in autocompletion and choose it.
* it's the same as Guice
"getInstance" is longer than "lookup" ;)
Thanks
-Vincent
On Mar 6, 2012, at 2:51 PM, Thomas Mortagne wrote:
--
Thomas Mortagne
It's committed.
Would be nice to have comments on the new APIs (suggest better naming,
etc...). It's a very important API and we really need to have
something we are sure of before 4.0 final.
On Mon, Mar 5, 2012 at 6:53 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
--
Thomas Mortagne
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
--
Thomas Mortagne
--
Thomas Mortagne
8:45 a.m.
On Apr 5, 2012, at 8:35 AM, Thomas Mortagne wrote:
On Wed, Apr 4, 2012 at 8:00 PM, Vincent Massol
<vincent(a)massol.net> wrote:
Yes…. But I couldn't find a short one that would get everyone's agreement ;)
It's still shorter than lookupComponent.
There's also getComponent, getComponentList and getComponentMap which are not that bad
but getInstance() seems to win everyone's opinnion so far and I think it's good
that it's the same API than Guice.
Thanks
-Vincent
Hi Thomas/All,
All looks good except CM.lookupComponent() which Ï don't like a lot for the following
reasons:
* It's a bit long. It's longer than before and it would be nicer if it were as
short as before or even shorter ;)
* It's not symmetrical with other lookups like lookupList and lookupMap, which should
theoretically be lookupComponentList() and lookupComponentMap() which are even longer
Thus I'd prefer that the new API for lookup be named differently.
I propose to use "resolve":
CM.resolve(Type)
CM.resolveList(Type)
CM.resolveMap(Type)
Alternative:
CM.getInstance(Type)
CM.getInstanceList(Type)
CM.getInstanceMap(Type)
or even the short:
CM.get(Type)
CM.getList(Type)
CM.getMap(Type)
For information Guice uses getInstance(), Picocontainer uses getComponentInstance().
WDYT?
"get" sounds a bit lightweight to be for something supposed to also
initialize what you ask but if Guice use that I guess it means it's
OK.
I'm OK with any of theses suggestions so I will apply the voted one.
My preference goes to getInsta nce():
* users are tempted by default to start typing "get" to get an instance and
then they'll see getInstance in autocompletion and choose it.
* it's the same as Guice
"getInstance" is longer than "lookup" ;) Thanks
-Vincent
On Mar 6, 2012, at 2:51 PM, Thomas Mortagne wrote:
--
Thomas Mortagne
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs It's committed.
Would be nice to have comments on the new APIs (suggest better naming,
etc...). It's a very important API and we really need to have
something we are sure of before 4.0 final.
On Mon, Mar 5, 2012 at 6:53 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs Hi guys,
I'm almost done but there is one little detail left: for easier
retro-compatibility I would like to introduce a new @Role annotation
and deprecate the @ComponentRole annotation. The idea is that Role
will "officially" take into account generic parameters while
ComponentRole does not. That way no existing component role behavior
will be broken or require crappy retro-compatibility like registering
automatically both generic and non generic roles (what I started to do
at first).
On Thu, Mar 1, 2012 at 6:20 PM, Thomas Mortagne
<thomas.mortagne(a)xwiki.com> wrote:
> Hi devs,
>
> I recently modified component manager internals and changed the role
> used as key in the map from Class to Type so that it properly make a
> difference between Provider<Toto> and Provider<Titi> without the need
> to generate a special hint.
>
> For the one not aware of what Type is lets say it's what Class is
> extending. Among other things Type is also extended by
> ParameterizedType which is interesting here because it basically
> contain a Class with it's generic parameters (it also indicate if the
> type in included in another one but we don't care in our case ;)).
>
> To summarize:
> * Provider is a Class
> * Provider<String> is ParameterizedType
> and both are Type
>
> That said I would like to change the whole ComponentManager and
> related descriptors APIs to use Type as role instead of Class.
>
> = What it bring
>
> What it means is that it makes possible to cleanly represent a
> Provider in which the generic type is meaningful right in the
> ComponentDescriptor (right now the generic type is extract from the
> implementation class when the provider is registered which is pretty
> crappy) and simplify a lot the current implementation.
>
> But it also mean we could use this new "feature" for other components.
> For example right now we have things like:
>
> @Component
> public class DefaultStringDocumentReferenceResolver implements
> DocumentReferenceResolver<String>
>
> and
>
> @Component
> @Named("default/reference")
> public class DefaultReferenceDocumentReferenceResolver implements
> DocumentReferenceResolver<EntityReference>
>
> and injected by
>
> @Inject
> DocumentReferenceResolver<String> resolver;
>
> and
>
> @Inject
> @Named("default/reference")
> DocumentReferenceResolver<EntityReference> resolver;
>
> Having Type used as role means that we could have:
>
> @Inject
> DocumentReferenceResolver<String> resolver;
>
> and
>
> @Inject
> DocumentReferenceResolver<EntityReference> resolver;
>
> = What it breaks
>
> So what's the catch ?
>
> Basically it would break any code that currently programmatically
> register or unregister Provider since It would remove the current hack
> to discover the Provider parameter (assuming that the
> ComponentDescriptor now provide the complete role). But that's OK IMO
> given that Provider support is very recent and I doubt it's a common
> use case.
>
> We would obviously keep and deprecate all the existing APIs based on
> Class which are pretty easy to implement since Class extends Type.
> Even if removing them would not cause any issue when you build it's
> not binary compatible since it's not the same method signature.
>
> So WDYT ?
>
> Here is my +1 and since we are starting a new major version it sounds
> a good occasion so introduce such change.
>
> --
> Thomas Mortagne
--
Thomas Mortagne
--
Thomas Mortagne
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs
5 Mar
5 Mar
7:28 p.m.
On 03/01/2012 12:20 PM, Thomas Mortagne wrote:
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReferenceResolver implements
DocumentReferenceResolver<String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentReferenceResolver implements
DocumentReferenceResolver<EntityReference>
and injected by
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<String> resolver;
and
@Inject
DocumentReferenceResolver<EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
+1.
--
Sergiu Dumitriu
http://purl.org/net/sergiu/
7:29 p.m.
+1
On Mon, Mar 5, 2012 at 19:28, Sergiu Dumitriu <sergiu(a)xwiki.com> wrote:
On 03/01/2012 12:20 PM, Thomas Mortagne wrote:
--
Denis Gervalle
SOFTEC sa - CEO
eGuilde sarl - CTO
Hi devs,
I recently modified component manager internals and changed the role
used as key in the map from Class to Type so that it properly make a
difference between Provider<Toto> and Provider<Titi> without the need
to generate a special hint.
For the one not aware of what Type is lets say it's what Class is
extending. Among other things Type is also extended by
ParameterizedType which is interesting here because it basically
contain a Class with it's generic parameters (it also indicate if the
type in included in another one but we don't care in our case ;)).
To summarize:
* Provider is a Class
* Provider<String> is ParameterizedType
and both are Type
That said I would like to change the whole ComponentManager and
related descriptors APIs to use Type as role instead of Class.
= What it bring
What it means is that it makes possible to cleanly represent a
Provider in which the generic type is meaningful right in the
ComponentDescriptor (right now the generic type is extract from the
implementation class when the provider is registered which is pretty
crappy) and simplify a lot the current implementation.
But it also mean we could use this new "feature" for other components.
For example right now we have things like:
@Component
public class DefaultStringDocumentReference**Resolver implements
DocumentReferenceResolver<**String>
and
@Component
@Named("default/reference")
public class DefaultReferenceDocumentRefere**nceResolver implements
DocumentReferenceResolver<**EntityReference>
and injected by
@Inject
DocumentReferenceResolver<**String> resolver;
and
@Inject
@Named("default/reference")
DocumentReferenceResolver<**EntityReference> resolver;
Having Type used as role means that we could have:
@Inject
DocumentReferenceResolver<**String> resolver;
and
@Inject
DocumentReferenceResolver<**EntityReference> resolver;
= What it breaks
So what's the catch ?
Basically it would break any code that currently programmatically
register or unregister Provider since It would remove the current hack
to discover the Provider parameter (assuming that the
ComponentDescriptor now provide the complete role). But that's OK IMO
given that Provider support is very recent and I doubt it's a common
use case.
We would obviously keep and deprecate all the existing APIs based on
Class which are pretty easy to implement since Class extends Type.
Even if removing them would not cause any issue when you build it's
not binary compatible since it's not the same method signature.
So WDYT ?
Here is my +1 and since we are starting a new major version it sounds
a good occasion so introduce such change.
+1.
--
Sergiu Dumitriu
http://purl.org/net/sergiu/
______________________________**_________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/**mailman/listinfo/devs<http://lists.xwiki.org/ma…
4547
days inactive
4582
days old
17 comments
7 participants
participants (7)
-
Andreas Jonsson -
Denis Gervalle -
Eduard Moraru -
Marius Dumitru Florea -
Sergiu Dumitriu -
Thomas Mortagne -
Vincent Massol