AMD loader support for Dojo 1.6 built layer files

[Chris Mitchell]

The Dojo 1.7 AMD loader supports loading Dojo 1.6 formatted source modules, but does not support optimized/built layers for modules. We have received a number of requests from users that are doing early testing on 1.7beta6 for this support. Please consider for 1.7.1, since 1.7 is now feature frozen.

[cjolif]

This issue was partially fixed by #13845. However additional problems arise with for example resource bundles loading.

[Rawld Gill]

In [27797]:

fixed i18n to properly load v1.6- layer files; refs #14092; !strict

[Rawld Gill]

In [27799]:

adding a few lines missed in a merge that formed [27797]; refs #14092; !strict

[Rawld Gill]

In [27800]:

backport [27797] [27799], refs #14092; !strict

[Rawld Gill]

In [27801]:

fixed builder to build flattened layer i18n bundles as available in 1.6-; refs #14092; refs #14169; !strict

[Rawld Gill]

In [27802]:

backport [27801]; refs #14169, refs #14092; !strict

[Rawld Gill]

In [27806]:

fixed layers to write call to preloadLocalizations as per 1.6-; refs #14092, refs #14169; !strict

[Rawld Gill]

In [27807]:

fixed layers to write call to preloadLocalizations as per 1.6-; fixes #14092, fixes #14169; !strict

[Rawld Gill]

preloadLocalizations works as per 1.6- by default. It is possible to force the builder to omit the preloadLocalization calls by

• setting the build control flag preloadLocalizations to falsy
• not setting preloadLocalizations and setting the build control property defaultConfig.async to truthy

If preloadLocalizations is truthy, then preloadLocalization calls will be inserted into layer modules unconditionally.

[Rawld Gill]

v1.6 flattened NLS bundles often contain multiple dojo.provide's and never look like unbuilt v1.6- bundles. This causes the evalBundle in the "dojo-v1x-i18n-Api" has branch as given by the previous repairs to this ticket to fail because it must use checkForLegacyModules to find legacy-style global variables defined consequent to evaluating the bundle.

Also, v1.6 converted localized bundle module names to legal JavaScript? ids by replacing the locale dashes with underscores. This causes checkForLegacyModules in the previous repairs to this ticket to fail.

[Rawld Gill]

[28049] should have ref'd this ticket

[Rawld Gill]

In [28053]:

backport [28049]; refs #14092; !strict

[Colin Snover]

Once built (using the standard build profile w/ noref), dojo/cldr/nls/en/* are getting defined as nonmodule. This is not so good! I’m still trying to trace out why this is happening but thought I’d put feedback here in case someone else might know more quickly.

[Colin Snover]

I’m still not sure why or where this is happening, but the cache entry for the bundle is becoming closured by evalBundle which means that the wrong define is called (the one from inside evalBundle and the module value is never defined. Replacing evalBundle with a no-op function “fixes” the problem.

[Colin Snover]

This is happening because dijit/nls/dijit-all_en-us is treated as a legacy i18n bundle, so *that* module gets wrapped in the evalBundle closure, and then define is shadowed for *all* the functions. Since we aren’t looking for any deferred calls to define, patching evalBundle to look like this seems like it would work:

new Function("bundle", "__evalError", "__checkForLegacyModules", "__mid",
	"var __amdResult, define = function(x){__amdResult= x;}," +
	"ret = [(function(){" +
				"try{eval(arguments[0]);}catch(e){}" +
				"if(__amdResult)return 0;" +
				"if((__checkForLegacyModules = __checkForLegacyModules(__mid)))return __checkForLegacyModules;" +
				"try{return eval('('+arguments[0]+')');}" +
				"catch(e){__evalError.e = e; return __evalError;}" +
			"})(bundle),__amdResult]; define = this.define; return ret;"
)

The only change here is basically that after evaluating the return value, define is replaced with the global define again.

Rawld, thoughts please? (This same fix applies for both before and after [28053].)

Also, even with this change and a noref build, when loading cross-domain, extra requests are being triggered for the /en/ bundles.

[Rawld Gill]

The problem is that the current design doesn't account for flattened (built) NLS resources which were added in 1.7.2. The evalBundle function must detect and properly process the following

1. single NLS bundles expressed as AMD a single AMD module
2. multiple NLS bundles expressed as an AMD a layer
3. single legacy AMD bundles
4. multiple NLS bundles expressed as a flattened NLS layer as built by the legacy builder

The current design fails to handle Case 2.

[Rawld Gill]

In [28124]:

improved evalBundle to handle flattened, built, AMD NLS layers; refs #14092; !strict

[Rawld Gill]

In [28125]:

improved evalBundle to handle flattened, built, AMD NLS layers; refs #14092; !strict

[Rawld Gill]

In [28126]:

improved evalBundle to handle flattened, built, AMD NLS layers (note: [28124] checked in the wrong file and not sure where [28125] came from); refs #14092; !strict

[Rawld Gill]

@csnover: I've tested both not xdomain and xdomain and it seems to work afer 28126; let me know what you find at your convenience.

Keeping ticket open until I get confirmation from others that this repair works universally.

[Colin Snover]

To confirm, [28128] is a more appropriate patch for what was dropped in at comment #22?

[Colin Snover]

Rawld, could you please confirm my previous question? Also, these patches for correctly evaluating built NLS layers need to be backported to 1.7.3.

[Rawld Gill]

Replying to csnover:

To confirm, [28128] is a more appropriate patch for what was dropped in at comment #22?

28128 seems to have nothing to do with this issue. I don't understand your question. Also, there was no commit at #22, so I don't understand that either.

[Colin Snover]

[28124] through [28126] and comment:22.

[Rawld Gill]

Replying to csnover:

[28124] through [28126] and comment:22.

my response to comment:22 is at comment:24. I confirm that [28126] is the best fix I know of at this time. afaik, this is working correctly in trunk. I've rx'd no additional reports/complaints since comment:28, now 6 weeks ago.

[Rawld Gill]

In [28365]:

removed requirment for sync loader when including v1.x i18n backcompat; refs #14092; refs #15057; !strict

[Rawld Gill]

In [28398]:

improved sniff for dojo xhr factory; fixes #14831; refs #14092; !strict

[Rawld Gill]

In [28411]:

tuned i18n code; added exhaustive test for i18n; refs #14092; !strict

[Rawld Gill]

In [28412]:

more of exhaustive test for i18n; refs #14092; !strict

[Rawld Gill]

In [28438]:

fixed error in preloadLocalizations which caused waiting queue to stall; gave fast path for preloadLocalizations for known AMD bundles; fixes #14169; refs #14092; !strict

[Rawld Gill]

If backport to 1.7, see also changes in #14069

[Rawld Gill]

if backport to 1.7, see all changes in #15061

[Rawld Gill]

In [28785]:

backport i18n backcompat fixes; fixes #14092; !strict