Werner Punz schrieb:
Paul Rivera schrieb:
Interesting solution :)
I agree that the execution time of a compiler level solution will be
better than an interpreted template solution. Perhaps the only
scenario that our interpreted template solution will yeild
significant execution performance gain is when it caches large
amounts of javascript code. Browsing through the components, most of
our large javascript code are already rendered through AddResource
which is already cached. The remaining javascript code embedded into
some renderers are just not significantly large enough.
Do we plan to implement the same convention in myfaces-builder-plugin?
I.e.
An abstract renderer class that c ontains the javascript template
comment
A concrete subclass of the one above generated by
myfaces-builder-plugin that has the template comment from parent
abstract class converted into java code
Actually I would go for the concrete implementation approach. The
maven plugin of the compiler can take care if picking up the correct
files.
It even has package rewriting possibilities (I added such a directive
to the grammar on thursday)
So that people can work on the templated java files and have haven
compile the result into the generated sources (and still can link into
those if needed, due to being in a different package)...
This approach to templating is very interesting..and anything that
improves the current StringBuffer-based approach for javascript
generation is very nice to see! That code is really hard to work on..
What happens with breakpoints etc? This is always a tricky problem with
templates. Generating a subclass does at least mean that breakpoints can
be set in the "real" parent class, and the subclass contains only
generated code (for which breakpoints are not much use). If things get
magically compiled into a different package, then won't breakpoints set
in the original file be ignored?
I'm also somewhat concerned about the debuggability of classes when
templates and "normal" code are mixed in the same method. Does this work
ok? If not, then is it possible to use the convention of creating a
method containing *just* the magic template-comment, with the method
parameters as the data referenced from the template?
Regards,
Simon
