undefined.cache.js when starting GWT app in web mode?

[Ed]

When I start a compiled GWT web app, it's not starting as the startup js
script (the one included in the index.html), is complaining that it can't
find undefined.cache.js file.
Please some advice/pointers how to solve this?
Notes:
- In dev mode it works fine.
- In quess it has something to do with the extra formFactor gwt property I
introduced to create app's for different devices. But can't detect any
mistakes here, also compared it with how mgwt does it, which is almost the
same. It works in dev mode also.

I find it hard to search for the problem as I don't really know were to
start.

[Ed]

I am debugging the bootstrap js script to find out what is going wrong and
I am a bit confused, please some help?

It seems that the bootstrap script throws an exception that is silently
ignored, such that the strongName variable is undefined, such that it tries
to load undefined.cache.js..
BTW: I noticed this issue: 8135<https://code.google.com/p/google-web-toolkit/issues/detail?id=8135>,
that also had problems solving a bootstrap issue as the same exception is
silently ignored. Please improve this exception handling.
Anyway: I have defined a property called "browser.engine" (with the help of
@Thomas) as specified below (in it's own gwt xml file). Because it's a
derived property, I haven't specified a property provider. However, as far
as I understand, an exception occurs as in the bootstrap script as it can't
find a property provider.
In the boostrap script that is included in the index.html file the
following line determines the strong name (that fails):

strongName =
answers[computePropValue($intern_47)][computePropValue($intern_48)][computePropValue($intern_59)];

It fails on the first call, namely: computePropValue($intern_47)
With $intern_47 containing "browser.engine".
Below the method being called. This line results a value null, such that an
exception is thrown:
var value = providers[propName](), allowedValuesMap = values[propName];
Because "browser.egine" is a derived property, it contains no provider, but
as such an exception is thrown.
What is going wrong here ? please some help?

----
function computePropValue(propName){
* var value = providers[propName](), allowedValuesMap =
values[propName];*
if (value in allowedValuesMap) {
return value;
}
var allowedValuesList = [];
for (var k in allowedValuesMap) {
allowedValuesList[allowedValuesMap[k]] = k;
}
if (__propertyErrorFunc) {
__propertyErrorFunc(propName, allowedValuesList, value);
}
throw null;
}

-----

The Property in it's own BrowserEngine.gwt.xml file:
----
<module>
<define-property name="browser.engine" values="gecko,webkit,presto,trident"
/>

<set-property name="browser.engine" value="gecko">
<when-property-is name="user.agent" value="gecko1_8" />
</set-property>
<set-property name="browser.engine" value="webkit">
<when-property-is name="user.agent" value="safari" />
</set-property>
<set-property name="browser.engine" value="presto">
<when-property-is name="user.agent" value="opera" />
</set-property>
<set-property name="browser.engine" value="trident">
<any>
<when-property-is name="user.agent" value="ie6" />
<when-property-is name="user.agent" value="ie8" />
<!-- ie9 and ie10 will fallback to ie8 -->
</any>
</set-property>
</module>

-----

[Ed]

I didn't think it would help, but I just added a default browser.engine
property to the BrowserEngine.gwt.xml file, just below the
<define-property...:
<set-property name="browser.engine" value="webkit" /> <!-- provide a
default as last resort -->

Strange enough this does help, as it does starts now.
The code to determine the strong name has changed to:
strongName =
answers[computePropValue($intern_47)][computePropValue($intern_58)];

As you can see the "browser.engine" property has been removed as argument.
Before it was:
strongName = answers[computePropValue($intern_47)][computePropValue($
intern_48)][computePropValue($intern_59)];

Why this difference?
btw: it seems to work in both chrome/FF.

[Colin Alworth]

If you compile in PRETTY instead of DETAILED, it won't intern those
strings, but still will leave the output mostly readable (just no packages).

Without seeing the rest of the structure of the module files, it is hard to
speculate, but we're using more or less the same idea successfully, though
we're doing it backward. Inside of GXT we defined a broader set of user
agents, and then distill down to just the 6 (as of gwt 2.4, 2.5)
permutations that GWT itself supports. This takes over GWT's own
property-provider and uses ours instead. The chief difference is that
user.agent already has a property-provider, but we're trying to instruct it
which property to use instead.

In lieu of your full module/setup, here's a quick sample I threw together
that seems to do more or less what you are requiring, and seems to fail in
the same way:

<module rename-to="test">
<inherits name="com.google.gwt.core.Core" />

<define-property name="foo" value="a, b, c" />

<define-property name="bar" value="x, y" />

<set-property name="bar" value="x">
<any>
<when-property-is name="foo" value="a" />
<when-property-is name="foo" value="b" />
</any>
</set-property>
<set-property name="bar" value="y">
<when-property-is name="foo" value="c" />
</set-property>

<property-provider name="foo"><![CDATA[return
window.location.query;]]></property-provider>

<entry-point class="path.to.client.Test" />
</module>

public class Test implements EntryPoint {
@Override
public native void onModuleLoad() /*-{
console && console.log && console.log('loaded successfully');
}-*/;
}

The foo provider is compiled out as expected:
providers['foo'] = function(){
return window.location.search[1];
}

but bar gets a default handler, one that clearly doesn't make sense here:
providers['bar'] = function(){
return __gwt_getMetaProperty('bar');
}

By adding <set-property name="bar" value="x" />, the code works, but it
doesn't correctly select a y value when it should (the permutation selector
code is just gone).

Through any case where bar is present in the selector, I continue to see
this block:
unflattenKeylistIntoAnswers(['x', 'a', 'gecko1_8'],
'20D7C847BEB9A7D69CDF024B3FA7AE54');
unflattenKeylistIntoAnswers(['x', 'b', 'gecko1_8'],
'20D7C847BEB9A7D69CDF024B3FA7AE54');
unflattenKeylistIntoAnswers(['y', 'c', 'gecko1_8'],
'20D7C847BEB9A7D69CDF024B3FA7AE54');
unflattenKeylistIntoAnswers(['x', 'a', 'ie9'],
'2F5083A323F0E99E2BC83F08F6D97D35');
unflattenKeylistIntoAnswers(['x', 'b', 'ie9'],
'2F5083A323F0E99E2BC83F08F6D97D35');
unflattenKeylistIntoAnswers(['y', 'c', 'ie9'],
'2F5083A323F0E99E2BC83F08F6D97D35');
//etc
demonstrating that the 'collapsing' work is being done - there is never a
x-y, nor a y-a/y-b line. From this, I'm suspecting that the design of this
feature is to minimize permutations, not remove selector scripts altogether
- from
http://code.google.com/p/google-web-toolkit/wiki/ConditionalProperties,

"This alternateFeatures property is called *derived* because its value can
be determined solely from other deferred-binding properties. There is never
a need to execute a property provider for a derived property. Moreover,
derived properties do not expand the permutation matrix and have no
deployment cost."

Misleading, not not actually inaccurate - there *isn't* a need to execute
the property provider, but the current implementation seems to still do it
anyway, and fails when it cannot be done. The only advantage we are getting
out of conditional properties is the minimized set of permutations.

Post by Ed
I didn't think it would help, but I just added a default browser.engine
property to the BrowserEngine.gwt.xml file, just below the
<set-property name="browser.engine" value="webkit" /> <!-- provide a
default as last resort -->
Strange enough this does help, as it does starts now.
strongName =
answers[computePropValue($intern_47)][computePropValue($intern_58)];
As you can see the "browser.engine" property has been removed as argument.
strongName = answers[computePropValue($intern_47)][computePropValue($
intern_48)][computePropValue($intern_59)];
Why this difference?
btw: it seems to work in both chrome/FF.

[Ed Bras]

@Colin: thanks for isolated example that shows the problem.

[Colin Alworth]

Looks like my sample may have been too simplistic - upon re-reading
http://code.google.com/p/google-web-toolkit/wiki/ConditionalProperties it
looks like they do indeed suggest using a default as the 'correct' way to
do things:

<!-- Provide a default -->
<set-property name="alternateFeatures" value="false" />


I'll try to build a more complex test case that actually uses the
properties defined, just in case the compiler is getting to clever for me.

Post by Colin Alworth
If you compile in PRETTY instead of DETAILED, it won't intern those
strings, but still will leave the output mostly readable (just no packages).
Without seeing the rest of the structure of the module files, it is hard
to speculate, but we're using more or less the same idea successfully,
though we're doing it backward. Inside of GXT we defined a broader set of
user agents, and then distill down to just the 6 (as of gwt 2.4, 2.5)
permutations that GWT itself supports. This takes over GWT's own
property-provider and uses ours instead. The chief difference is that
user.agent already has a property-provider, but we're trying to instruct it
which property to use instead.
In lieu of your full module/setup, here's a quick sample I threw together
that seems to do more or less what you are requiring, and seems to fail in
<module rename-to="test">
<inherits name="com.google.gwt.core.Core" />
<define-property name="foo" value="a, b, c" />
<define-property name="bar" value="x, y" />
<set-property name="bar" value="x">
<any>
<when-property-is name="foo" value="a" />
<when-property-is name="foo" value="b" />
</any>
</set-property>
<set-property name="bar" value="y">
<when-property-is name="foo" value="c" />
</set-property>
<property-provider name="foo"><![CDATA[return
window.location.query;]]></property-provider>
<entry-point class="path.to.client.Test" />
</module>
public class Test implements EntryPoint {
@Override
public native void onModuleLoad() /*-{
console && console.log && console.log('loaded successfully');
}-*/;
}
providers['foo'] = function(){
return window.location.search[1];
}
providers['bar'] = function(){
return __gwt_getMetaProperty('bar');
}
By adding <set-property name="bar" value="x" />, the code works, but it
doesn't correctly select a y value when it should (the permutation selector
code is just gone).
Through any case where bar is present in the selector, I continue to see
unflattenKeylistIntoAnswers(['x', 'a', 'gecko1_8'],
'20D7C847BEB9A7D69CDF024B3FA7AE54');
unflattenKeylistIntoAnswers(['x', 'b', 'gecko1_8'],
'20D7C847BEB9A7D69CDF024B3FA7AE54');
unflattenKeylistIntoAnswers(['y', 'c', 'gecko1_8'],
'20D7C847BEB9A7D69CDF024B3FA7AE54');
unflattenKeylistIntoAnswers(['x', 'a', 'ie9'],
'2F5083A323F0E99E2BC83F08F6D97D35');
unflattenKeylistIntoAnswers(['x', 'b', 'ie9'],
'2F5083A323F0E99E2BC83F08F6D97D35');
unflattenKeylistIntoAnswers(['y', 'c', 'ie9'],
'2F5083A323F0E99E2BC83F08F6D97D35');
//etc
demonstrating that the 'collapsing' work is being done - there is never a
x-y, nor a y-a/y-b line. From this, I'm suspecting that the design of this
feature is to minimize permutations, not remove selector scripts altogether
- from
http://code.google.com/p/google-web-toolkit/wiki/ConditionalProperties,
"This alternateFeatures property is called *derived* because its value
can be determined solely from other deferred-binding properties. There is
never a need to execute a property provider for a derived property.
Moreover, derived properties do not expand the permutation matrix and have
no deployment cost."
Misleading, not not actually inaccurate - there *isn't* a need to execute
the property provider, but the current implementation seems to still do it
anyway, and fails when it cannot be done. The only advantage we are getting
out of conditional properties is the minimized set of permutations.

Post by Ed
I didn't think it would help, but I just added a default browser.engine
property to the BrowserEngine.gwt.xml file, just below the
<set-property name="browser.engine" value="webkit" /> <!-- provide a
default as last resort -->
Strange enough this does help, as it does starts now.
strongName =
answers[computePropValue($intern_47)][computePropValue($intern_58)];
As you can see the "browser.engine" property has been removed as
strongName = answers[computePropValue($intern_47)][computePropValue($
intern_48)][computePropValue($intern_59)];
Why this difference?
btw: it seems to work in both chrome/FF.