You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@pivot.apache.org by Greg Brown <gk...@mac.com> on 2009/09/17 22:04:57 UTC
Pivot 1.3.1 -> 1.4
Hi all,
I have changed the version number of the next Pivot release from 1.3.1 to 1.4 as a result of the numerous API changes we have made over the past couple of days. While these changes represent some significant improvements, the API has now evolved to the point where we could break backwards compatibility for a lot of applications built on Pivot 1.3, so this can no longer be considered a maintenance release.
1.3.2 will now become 1.4.1, and 1.4 will become 1.5.
Let me know if you have any questions or concerns.
Greg
Re: Pivot 1.3.1 -> 1.4
Posted by Sandro Martini <sa...@gmail.com>.
Ok, let's see in the future what happen, and if someone ask us to fix
something ...
Bye
Re: Pivot 1.3.1 -> 1.4
Posted by Greg Brown <gk...@mac.com>.
> Most of the recent (last month) API changes
> have been due to working through the tutorials, and as such, I
> anticipate
> that they'll stabilize a lot once we finish our push to provide better
> documentation.
Agreed. We stabilized many APIs for 1.3 as we worked through the
tutorials and documentation for those features. Similarly, I expect to
see the APIs for the features we are currently documenting stabilize
for 1.4. API churn has continued to decrease with each release, and I
suspect that it will level off once the documentation sweep is complete.
Re: Pivot 1.3.1 -> 1.4
Posted by Todd Volkert <tv...@gmail.com>.
You bring up a very valid and good scenario. In such cases, we could choose
to branch from the last release's tag and apply fixes to that [stable]
branch as well, so as to release a 1.3.1 while continuing to work on a 1.4.
I think this methodology would serve us well, though probably only once the
codebase stabilizes some more. Most of the recent (last month) API changes
have been due to working through the tutorials, and as such, I anticipate
that they'll stabilize a lot once we finish our push to provide better
documentation.
-T
On Fri, Sep 18, 2009 at 7:18 AM, Sandro Martini <sa...@gmail.com>wrote:
> Hi,
>
> > My 2c is that thus far, we've been releasing minor versions (1.1, 1.2,
> etc.)
> > every 3-4 months, and that that's about right. It's generally good
> practice
> > in open source to release more often than you do in traditional closed
> source software.
> Ok, my experience is mainly on big frameworks, mainly web frameworks,
> where a major release (with some incompatibility in code) is every 12
> / 18 or more months, take for example Struts, Spring, Wicket, etc ...
>
> > That would put a 1.4 release tag around early Dec for a
> > deployment in late Dec or early Jan. I also don't see us releasing
> another
> > major version (2.0, 3.0, etc.) at all until something big changes in the
> > platform -- if you look at the roadmap in JIRA, the 2.0 release has some
> > really grand things, like a whole new look & feel. Until then, I see us
> > releasing 1.10, 1.11, etc. if we have to.
>
> I understand your point of view, and I agree that this is a not-so-big
> project (as a codebase), but take a look at this scenario:
> after some weeks from the pivot release, a developer start to develop
> an application based on Pivot, say on the 1.3 .
> After some other weeks, a (little :-) ) bug is found in the 1.3, and
> the custom application need that fix.
> But in the meantime we are already on the next release (that contains
> incompatibilities with the previous, but probably this is not a major
> like 2.0, but only a medium release 1.4 or another 1.x), and no
> maintenance release for the 1.3 will be released.
> So our user (one or many) can:
> - modify pivot source code, and maybe sending back to us for some fixes
> - use a unstable version, taken from trunk, but in many companies this
> is not wanted (right)
> - wait the next release:
> change the (maybe already working code) to ensure all works good,
> in the case that some non-compatible change has been introduced in the
> meantime
> -- note: the new release could have old bugs fixed, but usually
> has new bugs, so what is better: a old or a new bug ? Probably none,
> but in this case what choice ...
> - other ?
>
>
> I think in our Release Policy (if any) we should handle these situations.
> So at least one maintenance sub-release could be enough for most cases.
>
>
> Staying on the same field of RIAs, for example what are doing Adobe
> (Flex), and Microsoft (Silverlight) ?
> Major releases every 12 / 18 or more months, and many mid releases,
> any with their maintenance sub-releases.
>
>
> My effort in this project is limited compared to that of Greg and
> Todd, so for me there aren't problems, but i think that we should
> improve cases like this.
>
> Comments ?
> Mentors, in your experience on Apache Projects what can you suggest to us ?
>
>
> Sandro
>
Re: Pivot 1.3.1 -> 1.4
Posted by Greg Brown <gk...@mac.com>.
> styles are are exposed through a layer of indirection - not
> by API. The fact that they're backed by real bean methods would be
> something that ideally wouldn't be exposed to callers.
I actually think that the way styles are implemented is fairly
elegant. From the component's perspective, they are simply key/value
pairs, which decouples the component from the skin nicely. On the
other hand, developers can easily read the Javadoc to see what styles
are supported and what the expected types and values are.
G
Re: Pivot 1.3.1 -> 1.4
Posted by Todd Volkert <tv...@gmail.com>.
> have no problem with this, as Wicket is doing the same thing in reality.
> One
> additional problem for both Pivot and Wicket is that the API is not very
> clear, and in fact many classes are "polluted" by properties and
view/model interaction, and view styles are also APIs, even though the
> methods are intended to be hidden. This is the nature of the "accepted
> approach" and perhaps not too much can be done about it.
>
The API being clear I'd agree with, insomuch as our documentation is very
lacking (and currently being addressed). But I think it's very consistent,
which makes it very clear once you work with it. I'm curious what instances
you found of the view talking directly to the model, since the view layer is
the skin, and it should only be talking to the component, which is the
controller. The styles are are exposed through a layer of indirection - not
by API. The fact that they're backed by real bean methods would be
something that ideally wouldn't be exposed to callers. In an ideal world,
we wouldn't provide Javadoc for the skins on our documentation site, but
instead "style docs" that were generated using either annotations or
introspection.
> In the long term, I expect a stable 2.0 emerge that follows the general
> comaptibility guidelines and the cruft has been removed. Once that happen,
> I
> think we should be much more stringent than now.
>
Yup - I agree.
Re: Pivot 1.3.1 -> 1.4
Posted by Niclas Hedhman <ni...@hedhman.org>.
Btw, all this comes across like 'bitching' and that is not my
intention. Sorry about that.
I think all-in-all, Pivot is not by any means worse than anything
comparable, and pushing ahead the way you have done is really
appreciated. Perfection is not always a goal in itself.
Keep up the good work.
Cheers
On Sat, Sep 19, 2009 at 3:50 PM, Niclas Hedhman <ni...@hedhman.org> wrote:
> On Fri, Sep 18, 2009 at 10:21 PM, Greg Brown <gk...@mac.com> wrote:
>>> many classes are "polluted" by properties and view/model interaction, and
>>> view styles are also APIs, even though the methods are intended to be
>>> hidden. This is the nature of the "accepted approach" and perhaps not too
>>> much can be done about it.
>>
>> I'm not sure exactly what you mean by "polluted" - can you clarify? I
>> actually think the APIs are very clear, especially now that we have been
>> over them 2 or 3 times.
>
> :-) I am a bit of API zealot, so I don't expect you to agree with me,
> and I don't know where to start. To sample the obvious, Component has
> for instance the following methods;
>
> public Point mapPointFromAncestor(Container ancestor, int x, int y)
> public Point mapPointToAncestor(Container ancestor, int x, int y)
> public Display getDisplay()
> public Graphics2D getGraphics()
>
> I can't for a second figure out why these are 'public' and should
> probably be 'protected'.
>
> There is a method
>
> public Skin getSkin()
>
> Is it part of the API (or not) that I call getSkin(), cast it to the
> implementation and invokes methods on it directly? If not, it
> shouldn't be public. If it is, you just added all the public methods
> of all the view classes to be part of the API.
>
> All containers have a method;
>
> public final int add(Component component)
>
> yet, many containers are not using it as its primary populator, off my
> head I recall Accordion.getPanels().add() or something like that. If
> only a small number of Containers are actually using the add(), it
> should exist at that level, IMHO.
>
>
> Then there is stuff I am less sure about;
>
> Component.validate() sounds like an internal method between a
> Container and its children.
> Component.paint( Graphics2D graphics ) --> same thing.
>
> Component.getAttributes() --> is 'protected' but none of the existing
> sub-components calls it. Instead it is called from 'package private'
> visibility of Accordion, TabPane, TablePane and Form. So that is
> indication that scope visibility is incorrect (I could not have
> implemented Accordion outside the package). Also, Attributes sound
> like something that should be properly typed and not a 'guess my type'
> Object.
>
> And this is just scratching on the surface. A lot of the issues could
> have been solved with using interfaces instead of classes, where the
> implementation details in the class can cast to its known types for
> access. Some issues should perhaps be solved via an injection
> mechanism.
>
>
> So, nah, you won't get an A for API Design... ;-)
>
>
> Cheers
> --
> Niclas Hedhman, Software Developer
> http://www.qi4j.org - New Energy for Java
>
> I live here; http://tinyurl.com/2qq9er
> I work here; http://tinyurl.com/2ymelc
> I relax here; http://tinyurl.com/2cgsug
>
--
Niclas Hedhman, Software Developer
http://www.qi4j.org - New Energy for Java
I live here; http://tinyurl.com/2qq9er
I work here; http://tinyurl.com/2ymelc
I relax here; http://tinyurl.com/2cgsug
Re: Pivot 1.3.1 -> 1.4
Posted by Greg Brown <gk...@mac.com>.
> :-) I am a bit of API zealot, so I don't expect you to agree with me,
> and I don't know where to start. To sample the obvious, Component has
> for instance the following methods;
>
> public Point mapPointFromAncestor(Container ancestor, int x, int y)
> public Point mapPointToAncestor(Container ancestor, int x, int y)
These are convenience methods for translating between coordinate
systems. You might use them, for example, if you had registered a
mouse listener on the display but needed the mouse location in a
particular component's coordinate space. Other toolkits have similar
methods.
> public Display getDisplay()
A Display represents the root of a component hierarchy. Originally, we
had a single Display singleton, but that doesn't work when multiple
applets from the same origin are hosted in a single page (since all
the applets run in the same JVM instance). So, we support multiple
Display instances in the same JVM so the applets don't stomp on each
other.
> public Graphics2D getGraphics()
This returns a graphics context that can be used to draw on a
component directly. AWT has a similar method. However, it is currently
only used by ScrollPane, and may no longer be necessary if we change
the way scroll panes optimize scrolling. Todd can comment better on
this.
> There is a method
>
> public Skin getSkin()
Are you talking about Component#getSkin()? If so, that method is
protected. Maybe you assumed it was public because it appears in the
Javadoc?
> All containers have a method;
>
> public final int add(Component component)
>
> yet, many containers are not using it as its primary populator, off my
> head I recall Accordion.getPanels().add() or something like that. If
> only a small number of Containers are actually using the add(), it
> should exist at that level, IMHO.
Skins often use this method to add other components to the container.
For example, TerraTabPaneSkin adds tab button components. The "tabs"
collection allows the caller (and the skin) to distinguish between
components that are actually tabs and those that are not.
> Then there is stuff I am less sure about;
>
> Component.validate() sounds like an internal method between a
> Container and its children.
It is, but it needs to be public, primarily so skins can call it.
Also, there's no reason for it not to be public - it isn't harmful.
> Component.paint( Graphics2D graphics ) --> same thing.
Again, there is no reason for this method not to be public - and,
there are valid use cases for making it public (for example, taking a
screen shot of a component). Note that the paint() method is also
public in AWT.
> Component.getAttributes() --> is 'protected' but none of the existing
> sub-components calls it.
Making it protected allows 3rd party components to also define
attributes.
> Also, Attributes sound
> like something that should be properly typed and not a 'guess my type'
> Object.
The definition of the type is up to the subclass. Attributes are
strongly typed within containers that use them.
> So, nah, you won't get an A for API Design... ;-)
Conversely, if you were taking my API design class, you might not
score a particularly high grade yourself. :-)
Re: Pivot 1.3.1 -> 1.4
Posted by Niclas Hedhman <ni...@hedhman.org>.
On Fri, Sep 18, 2009 at 10:21 PM, Greg Brown <gk...@mac.com> wrote:
>> many classes are "polluted" by properties and view/model interaction, and
>> view styles are also APIs, even though the methods are intended to be
>> hidden. This is the nature of the "accepted approach" and perhaps not too
>> much can be done about it.
>
> I'm not sure exactly what you mean by "polluted" - can you clarify? I
> actually think the APIs are very clear, especially now that we have been
> over them 2 or 3 times.
:-) I am a bit of API zealot, so I don't expect you to agree with me,
and I don't know where to start. To sample the obvious, Component has
for instance the following methods;
public Point mapPointFromAncestor(Container ancestor, int x, int y)
public Point mapPointToAncestor(Container ancestor, int x, int y)
public Display getDisplay()
public Graphics2D getGraphics()
I can't for a second figure out why these are 'public' and should
probably be 'protected'.
There is a method
public Skin getSkin()
Is it part of the API (or not) that I call getSkin(), cast it to the
implementation and invokes methods on it directly? If not, it
shouldn't be public. If it is, you just added all the public methods
of all the view classes to be part of the API.
All containers have a method;
public final int add(Component component)
yet, many containers are not using it as its primary populator, off my
head I recall Accordion.getPanels().add() or something like that. If
only a small number of Containers are actually using the add(), it
should exist at that level, IMHO.
Then there is stuff I am less sure about;
Component.validate() sounds like an internal method between a
Container and its children.
Component.paint( Graphics2D graphics ) --> same thing.
Component.getAttributes() --> is 'protected' but none of the existing
sub-components calls it. Instead it is called from 'package private'
visibility of Accordion, TabPane, TablePane and Form. So that is
indication that scope visibility is incorrect (I could not have
implemented Accordion outside the package). Also, Attributes sound
like something that should be properly typed and not a 'guess my type'
Object.
And this is just scratching on the surface. A lot of the issues could
have been solved with using interfaces instead of classes, where the
implementation details in the class can cast to its known types for
access. Some issues should perhaps be solved via an injection
mechanism.
So, nah, you won't get an A for API Design... ;-)
Cheers
--
Niclas Hedhman, Software Developer
http://www.qi4j.org - New Energy for Java
I live here; http://tinyurl.com/2qq9er
I work here; http://tinyurl.com/2ymelc
I relax here; http://tinyurl.com/2cgsug
Re: Pivot 1.3.1 -> 1.4
Posted by Greg Brown <gk...@mac.com>.
> many classes are "polluted" by properties and view/model
> interaction, and
> view styles are also APIs, even though the methods are intended to be
> hidden. This is the nature of the "accepted approach" and perhaps
> not too
> much can be done about it.
I'm not sure exactly what you mean by "polluted" - can you clarify? I
actually think the APIs are very clear, especially now that we have
been over them 2 or 3 times.
> In the long term, I expect a stable 2.0 emerge that follows the
> general
> comaptibility guidelines and the cruft has been removed. Once that
> happen, I
> think we should be much more stringent than now.
I agree. I also agree that Pivot is currently following what would
more traditionally be used as a "0.x" release model, but I think the
"1.x" scheme is OK because the platform is much more complete than a
"0.x" scheme would imply.
Re: Pivot 1.3.1 -> 1.4
Posted by Niclas Hedhman <ni...@hedhman.org>.
<touched send by mistake>
many classes are "polluted" by properties and view/model interaction, and
view styles are also APIs, even though the methods are intended to be
hidden. This is the nature of the "accepted approach" and perhaps not too
much can be done about it.
In the long term, I expect a stable 2.0 emerge that follows the general
comaptibility guidelines and the cruft has been removed. Once that happen, I
think we should be much more stringent than now.
-- Niclas
On Sep 18, 2009 9:38 PM, "Niclas Hedhman" <ni...@hedhman.org> wrote:
Well, all this is a lot of personal choice.
I'm pretty concious about these things myself, and wouldn't do a 1.0 release
until I feel the APIs are fairly stable from a compatibility point of view.
In a 0.x series you can change as much as you like.
For example, we are currently about to release Qi4j 1.0, but that represents
4th generation of complete rewrites and API changes, plus numerous
incompatible smaller changes along the way.
Now, for Pivot, I think the current 1.x is really a 0.x series, as I think
that there should not be backward incompatible changes within 1.x. And I
have no problem with this, as Wicket is doing the same thing in reality. One
additional problem for both Pivot and Wicket is that the API is not very
clear, and in fact many classes
> > On Sep 18, 2009 7:18 PM, "Sandro Martini" <sa...@gmail.com>
wrote: > > Hi,
> My 2c is that thus far, we've been releasing minor versions (1.1, 1.2,
etc.) > every 3-4 months, ...
> > Ok, my experience is mainly on big frameworks, mainly web frameworks, >
where a major release (...
> That would put a 1.4 release tag around early Dec for a > deployment in
late Dec or early Jan. I...
> > I understand your point of view, and I agree that this is a not-so-big >
project (as a codebase...
Re: Pivot 1.3.1 -> 1.4
Posted by Niclas Hedhman <ni...@hedhman.org>.
Well, all this is a lot of personal choice.
I'm pretty concious about these things myself, and wouldn't do a 1.0 release
until I feel the APIs are fairly stable from a compatibility point of view.
In a 0.x series you can change as much as you like.
For example, we are currently about to release Qi4j 1.0, but that represents
4th generation of complete rewrites and API changes, plus numerous
incompatible smaller changes along the way.
Now, for Pivot, I think the current 1.x is really a 0.x series, as I think
that there should not be backward incompatible changes within 1.x. And I
have no problem with this, as Wicket is doing the same thing in reality. One
additional problem for both Pivot and Wicket is that the API is not very
clear, and in fact many classes
On Sep 18, 2009 7:18 PM, "Sandro Martini" <sa...@gmail.com> wrote:
Hi,
> My 2c is that thus far, we've been releasing minor versions (1.1, 1.2,
etc.) > every 3-4 months, ...
Ok, my experience is mainly on big frameworks, mainly web frameworks,
where a major release (with some incompatibility in code) is every 12
/ 18 or more months, take for example Struts, Spring, Wicket, etc ...
> That would put a 1.4 release tag around early Dec for a > deployment in
late Dec or early Jan. I...
I understand your point of view, and I agree that this is a not-so-big
project (as a codebase), but take a look at this scenario:
after some weeks from the pivot release, a developer start to develop
an application based on Pivot, say on the 1.3 .
After some other weeks, a (little :-) ) bug is found in the 1.3, and
the custom application need that fix.
But in the meantime we are already on the next release (that contains
incompatibilities with the previous, but probably this is not a major
like 2.0, but only a medium release 1.4 or another 1.x), and no
maintenance release for the 1.3 will be released.
So our user (one or many) can:
- modify pivot source code, and maybe sending back to us for some fixes
- use a unstable version, taken from trunk, but in many companies this
is not wanted (right)
- wait the next release:
change the (maybe already working code) to ensure all works good,
in the case that some non-compatible change has been introduced in the
meantime
-- note: the new release could have old bugs fixed, but usually
has new bugs, so what is better: a old or a new bug ? Probably none,
but in this case what choice ...
- other ?
I think in our Release Policy (if any) we should handle these situations.
So at least one maintenance sub-release could be enough for most cases.
Staying on the same field of RIAs, for example what are doing Adobe
(Flex), and Microsoft (Silverlight) ?
Major releases every 12 / 18 or more months, and many mid releases,
any with their maintenance sub-releases.
My effort in this project is limited compared to that of Greg and
Todd, so for me there aren't problems, but i think that we should
improve cases like this.
Comments ?
Mentors, in your experience on Apache Projects what can you suggest to us ?
Sandro
Re: Pivot 1.3.1 -> 1.4
Posted by Sandro Martini <sa...@gmail.com>.
Hi,
> My 2c is that thus far, we've been releasing minor versions (1.1, 1.2, etc.)
> every 3-4 months, and that that's about right. It's generally good practice
> in open source to release more often than you do in traditional closed source software.
Ok, my experience is mainly on big frameworks, mainly web frameworks,
where a major release (with some incompatibility in code) is every 12
/ 18 or more months, take for example Struts, Spring, Wicket, etc ...
> That would put a 1.4 release tag around early Dec for a
> deployment in late Dec or early Jan. I also don't see us releasing another
> major version (2.0, 3.0, etc.) at all until something big changes in the
> platform -- if you look at the roadmap in JIRA, the 2.0 release has some
> really grand things, like a whole new look & feel. Until then, I see us
> releasing 1.10, 1.11, etc. if we have to.
I understand your point of view, and I agree that this is a not-so-big
project (as a codebase), but take a look at this scenario:
after some weeks from the pivot release, a developer start to develop
an application based on Pivot, say on the 1.3 .
After some other weeks, a (little :-) ) bug is found in the 1.3, and
the custom application need that fix.
But in the meantime we are already on the next release (that contains
incompatibilities with the previous, but probably this is not a major
like 2.0, but only a medium release 1.4 or another 1.x), and no
maintenance release for the 1.3 will be released.
So our user (one or many) can:
- modify pivot source code, and maybe sending back to us for some fixes
- use a unstable version, taken from trunk, but in many companies this
is not wanted (right)
- wait the next release:
change the (maybe already working code) to ensure all works good,
in the case that some non-compatible change has been introduced in the
meantime
-- note: the new release could have old bugs fixed, but usually
has new bugs, so what is better: a old or a new bug ? Probably none,
but in this case what choice ...
- other ?
I think in our Release Policy (if any) we should handle these situations.
So at least one maintenance sub-release could be enough for most cases.
Staying on the same field of RIAs, for example what are doing Adobe
(Flex), and Microsoft (Silverlight) ?
Major releases every 12 / 18 or more months, and many mid releases,
any with their maintenance sub-releases.
My effort in this project is limited compared to that of Greg and
Todd, so for me there aren't problems, but i think that we should
improve cases like this.
Comments ?
Mentors, in your experience on Apache Projects what can you suggest to us ?
Sandro
Re: Pivot 1.3.1 -> 1.4
Posted by Todd Volkert <tv...@gmail.com>.
My 2c is that thus far, we've been releasing minor versions (1.1, 1.2, etc.)
every 3-4 months, and that that's about right. It's generally good practice
in open source to release more often than you do in traditional closed
source software. That would put a 1.4 release tag around early Dec for a
deployment in late Dec or early Jan. I also don't see us releasing another
major version (2.0, 3.0, etc.) at all until something big changes in the
platform -- if you look at the roadmap in JIRA, the 2.0 release has some
really grand things, like a whole new look & feel. Until then, I see us
releasing 1.10, 1.11, etc. if we have to.
-T
On Fri, Sep 18, 2009 at 5:56 AM, Sandro Martini <sa...@gmail.com>wrote:
> Hi Greg,
> for me it's Ok, but my only trouble is to ensure that this new release
> will not be delivered in a short time immediately after the 1.3,
> because in this case the general perception could be that the project
> is still evolving too fast (and this is true) because it's not enough
> mature (and this is not true).
> And if we want to spread to more users this could not be the best ...
>
> As a best practice we could apply for at least a minor release after a
> major, as in most projects.
>
> What do you think ?
>
> Sandro
>
Re: Pivot 1.3.1 -> 1.4
Posted by Sandro Martini <sa...@gmail.com>.
Hi Greg,
for me it's Ok, but my only trouble is to ensure that this new release
will not be delivered in a short time immediately after the 1.3,
because in this case the general perception could be that the project
is still evolving too fast (and this is true) because it's not enough
mature (and this is not true).
And if we want to spread to more users this could not be the best ...
As a best practice we could apply for at least a minor release after a
major, as in most projects.
What do you think ?
Sandro