Re: proposal for new behavior of compiler when declaring child nodes with same name as attributes
2008-06-02 15:40:05 GMT
Yeah, I only envision this being used in a specific case, where you have some class
that expects to have a certain named child or children, but those children aren't
specified in the class, they are only supplied when a call to create an instance of the
class is used. Like for example
<!-- expects a named child named "content", and turns it upside down -->
<view name="content"> ....
So in that case, the "content" var must be declared somehow to get the class to compile under swf9, so you'd say
<attribute name="content" type="node"/>
in the class declaration.
Aside from this use case, the common case will now be that if you inadvertently shadow a declared
attribute with a named child, you'll get a compiler warning, which should hopefully prevent hard-to-find bugs
Surely you want to be warned if you are tromping on one of the attributes of your superclass?
LZX has always only had one namespace. But it used to have 3 separate lists of 'initial arguments' to constructors:
1) simple initial values
even though these different initial values were all specified in the same place in an attribute.
This was an unending source of bugs, because people were constantly discovering that when they subclassed a class, overriding an initial argument of one kind with another kind never worked the way you expected. (E.g., if your superclass constrained an attribute and you tried to set it to a constant, your superclass would win! The work-around was for you to constrain the value in your subclass to a constant value.) And each of these lists was treated differently when they supplied an initial value for an attribute that had a setter!
Now we have a single initial arguments list, and overriding works as expected. But also, the processing of values for attributes with setters is uniform. This should be a good thing, but revealed a gray area in the language. What did it mean to have a child node with the same name as an attribute? What if that attribute has a setter? Or is processed specially by the class construct method?
To Henry and I, it seems like trying to use the same name for an attribute and a child node ought to be an error. But, if you _really_ want this, we give you an out. You can declare your attribute to be of type node, and when the child is created, it will be stored in the attribute (using the setter if there is one).
On 2008-06-02, at 10:25 EDT, jamesr wrote:It's not that I disagree with the proposal, but in a way I mourn the passing of simpler namespacing. It seems like the orthgonal namespaces increase complexity of the run time, and part of me wants to see that complexity managed by framework authors - i.e. using prefixed names for reserved words so that layout becomes lzLayout. It's not that the proposal is bad from a CS standpoint, but that it steepens the learning curve of how laszlo works. It seems to me that telling people that what names you see from inspecting a node is what populates your namespace (like a typical object), not that some conflict and some don't based on an implementation you don't see.
However this is just philosophy and not pragmatism. If this proposed feature works stably and doesn't cause problems, it's hard to argue against it.
On Jun 2, 2008, at 8:38 AM, P T Withington wrote:I think this is a good proposal, since it should actually be pretty rare that you want to declare an attribute that ends up being one of your child nodes.
On 2008-05-28, at 13:24 EDT, Henry Minsky wroteSorry my message got cut off before I could finish it:
This is a proposal in response to this bug, which came up when someone named
a child view "layout", and got
unexpected results in a view.
The current compiler behavior is that it will allow you to declare a child
with the same name as an attribute,
as long as the attribute has a null default value. This unfortunately
applies to virtually every attribute
declared in the schema, so in practice you can easily shadow all sorts of
important properties in
a view or node, by naming a child view with the same name as a class
attribute, with no warning from the compiler.
This proposal is that we add a new LZX attribute type, "node", which you can
use to declare that
an attribute name can have the same name as child node.
So for example you could have
<attribute name="titleview" type="node"/>
Then elsewhere you could say
And the compiler would not complain.
The existing compiler behavior would be changed so that for any attribute
except "node", if you name a child node
with the same name as that attribute, you get a compiler warning,
whether the attribute was declared with a default value.