Invariant Bindings vs. Type Coercer

[Howard Lewis Ship <hl...@gmail.com>]

I've seen a very subtle bug in an application's code.

The application has an interface called FieldSet, that represents a
set of fields on a kind of property sheet.  There's lots of components
that implement FieldSet.  The FieldSet has a "label" property of type
String.  In most cases, the FieldSet's label property is simply the
label property of the TextField inside the FieldSet (some properties
are more complicated).  In rare cases, the FieldSet's label property
may be dynamic, as can the label parameter of the TextField.

There's also a type coercion from FieldSet to String that extracts that label.

Here's where it gets tricky:

  <t:someothercomponent label="component:fieldset"/>

This works perfectly when the FieldSet's label is static.  Because the
"component:" binding prefix is invariant, the TypeCoercion from
FieldSet to String is computed the first time and cached for later.

The problem is that if the FieldSet's label property is actually
dynamic, this is not reflected properly. The first access runs through
the coercion; however the value is cached because "component:" is
invariant and the logic inside Tapestry assumes that any value derived
from an invariant binding is also invariant .. but that's not the
case.

Part of me thinks we can kludge this by simply not caching the results
when a type coercion is necessary.

However, I'm thinking that a lot of the caching associated with
invariant bindings is simply not necessary and would result in
Tapestry operating more predictably without any perceptible
performance change.

What's frustrating the correct solution to this is to build out a
sizable app and do some performance testing with the current state of
the app, and the two solutions above, and see which is best. I'm just
buried under too much urgent billable work to build that out, really
feel like I'm running the red queen's race.


-- 
Howard M. Lewis Ship

Creator of Apache Tapestry

The source for Tapestry training, mentoring and support. Contact me to
learn how I can get you up and productive in Tapestry fast!

(971) 678-5210
http://howardlewisship.com

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tapestry.apache.org
For additional commands, e-mail: dev-help@tapestry.apache.org

Re: Invariant Bindings vs. Type Coercer

[Massimo Lusetti <ml...@gmail.com>]

On Mon, Jan 16, 2012 at 8:49 PM, Howard Lewis Ship <hl...@gmail.com> wrote:

> What's frustrating the correct solution to this is to build out a
> sizable app and do some performance testing with the current state of
> the app, and the two solutions above, and see which is best. I'm just
> buried under too much urgent billable work to build that out, really
> feel like I'm running the red queen's race.

Would you mind elaborating on what a "sizable app" could be?

Cheers
-- 
Massimo
http://meridio.blogspot.com

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tapestry.apache.org
For additional commands, e-mail: dev-help@tapestry.apache.org