301413 - Valid bindings generate MDG loop warning

Status: RESOLVED FIXED

(Core Graveyard :: XForms, defect)

Description

[aaronr]

I simplified down a testcase that was showing a group that should have been
hidden since it was disabled.  I notice a warning in the .exe's output:
"WARNING: XForms: There are loops in the MDG".  I was running with the patch for
301168, too.

Allan, since it is a warning in the MDG, could you take a look at it?

Comment 1

[aaronr]

Attachment: testcase

by altering any number of things in the testcase source, the testcase will
begin to work.	Like the relevancy expression in the first bind will always
equate to "true()", and if I use "true()", then I no longer see the problem. 
And if I change an unused bind (use3dont) to point to another unused node like
/root/useArray/use[2]/do, then the problem will stop occurring.  Really weird.

Comment 2

[Allan Beaufour]

(In reply to comment #0)
> Allan, since it is a warning in the MDG, could you take a look at it?

Will do.

Comment 3

[Allan Beaufour]

Attachment: Reduced testcase

Comment 4

[Allan Beaufour]

Attachment: One-line fix (including some extra debug)

The problem was that nsXFormsMDGEngine::AddMIP() was adding the dependencies of
a MIP as dependencies of the same type as itself. Fx.: a relevant was dependent
on another node's relevant property. That's incorrect, it should always depend
on the other node's calculate property (as it is the only one influencing
it...).

Comment 5

[Allan Beaufour]

(In reply to comment #4)
> Created an attachment (id=190402) [edit]
> One-line fix (including some extra debug)

Pretend there's a comment explaining that in AddMIP() :) I'll add it before
check in.
 
> The problem was that nsXFormsMDGEngine::AddMIP() was adding the dependencies of
> a MIP as dependencies of the same type as itself. Fx.: a relevant was dependent
> on another node's relevant property. That's incorrect, it should always depend
> on the other node's calculate property (as it is the only one influencing
> it...).

Comment 6

[Allan Beaufour]

Attachment: w/comments in AddMIP

Comment 7

[Doron Rosenberg (IBM)]

checked in

Comment 8

[Allan Beaufour]

I think there's still a bug though. If the MIPs were calculates the MDG would
say there's a loop, even though there's not. This is a problem with the way the
MDG finds dependencies.