Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 333327

Summary:

Repo import doesn't work with cross-packageUnit dependencies

Product:

[Modeling] EMF

Reporter:

Caspar D. <caspar_d>

Component:

cdo.core

Assignee:

Caspar D. <caspar_d>

Status:

CLOSED FIXED

QA Contact:

Severity:

normal

Priority:

CC:

saulius.tvarijonas

Version:

4.0

Flags:

stepper: review+

Target Milestone:

---

Hardware:

All

OS:

All

Whiteboard:

Attachments:

Description	Flags
Patch v1	none
Patch v2 - ready to be committed	none

Description Caspar D.

2010-12-30 02:55:46 EST

When importing a set of packageUnits that contain packages
that have dependencies across the packageUnit boundaries,
the import encounters problems during the package import.

The (first) problem seems to be that the packageLoader constructed
in FlushHandler.handleModels calls EMFUtil.createEPackage with
argument lookForResource == true. This causes a lookup of the
resource in the resourceSet, which in turn delegates to the
cdoPackageRegistry, which in turn calls *another* packageLoader
(the repo), which in turn calls loadPackageUnit on the 
MEMStoreAccessor, which throws an UnsupportedOperationException.

Apparent solution: use lookForResource = false. But this might
cause the creation of Resource duds (resources that appear loaded,
but are actually empty), because of the subsequent call to 
packageUnit.load(..) invokes ECoreUtil.resolveAll(*) on every
ePackage obtained from the packageUnit. This also doesn't seem
quite right to me. I think no resolution should be requested until
all packageUnits have been processed...

Comment 1 Caspar D.

2010-12-30 03:05:03 EST

Actually I've changed my mind: calling #createEPackage with 
lookForResource == true, makes sense. (In fact I can't think
of a scenario when calling it with 'false' would make sense.)

The problem is just that we have to keep the resourceSet from
delegating to some other object, which may attempt a demandLoad
or whatever. So, in the patch that I'll upload shortly, I'm
introducing a derivative of ResourceSetImpl that allows the
lookup delegation to be disabled/enabled as needed.

Also, I'll add a boolean arg 'resolve' to the packageUnit.load(*)
call, because I do still believe the resolution should be
postponed until the resourceSet is known to contain the contents
of ALL the packageUnits being imported.

Comment 2 Caspar D.

2010-12-30 04:12:33 EST

Created attachment 185901 [details]
Patch v1

Comment 3 Eike Stepper

2011-01-02 05:29:16 EST

Created attachment 185933 [details]
Patch v2 - ready to be committed

Reformatted slightly

Comment 4 Caspar D.

2011-01-03 22:03:32 EST

Committed to HEAD

Comment 5 Eike Stepper

2011-06-23 03:40:14 EDT

Available in R20110608-1407