705344 - Use IDL for Components.(utils.)lookupMethod

Status: RESOLVED FIXED

(Core :: XPConnect, defect)

Description

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

Attachment: Patch v1

Comment 1

[Andreas Gal :gal]

Comment on attachment 576974 [details] [diff] [review]
Patch v1

If you enter a request also enter a compartment. However is the auto request actually needed?

Comment 2

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

No idea, this is pre-existing code.

Comment 3

[Bobby Holley (:bholley)]

Comment on attachment 576974 [details] [diff] [review]
Patch v1

I'm hoping this method will go away entirely (see bug693733 comment 13). But in the mean time, this is a righteous change. :-)

Also, I'm not too worried about Andreas' comment, since the code stays the same, and requests are going away anyway in 8 weeks.


>-    argv[0] = OBJECT_TO_JSVAL(obj);
>-    rv = nsXPConnect::GetXPConnect()->GetJSObjectOfWrapper(cx, obj, &obj);
>+    nsresult rv = nsXPConnect::GetXPConnect()->GetJSObjectOfWrapper(aCx, obj, &obj);
>     if (NS_FAILED(rv))
>         return rv;

Looks like we've been assuming nsXPConnect::GetXPConnect() is never null here, which is a likely indicator that the thing in bug 705333 is ok.

r=bholley.

Comment 4

[Bobby Holley (:bholley)]

As mentioned in bug 705333, the aFoo argument naming convention needs to be removed here as well.

Comment 5

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

Looks like you were right after all: https://tbpl.mozilla.org/php/getParsedLog.php?id=7722856&tree=Try

Comment 6

[Bobby Holley (:bholley)]

I'm totally confused - where in the removed code was the compartment being entered?

Comment 7

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

No idea; I'm removing this code, not claiming to understand what it does :)

Comment 8

[Bobby Holley (:bholley)]

Comment on attachment 576974 [details] [diff] [review]
Patch v1

Ah, right.

>+    nsresult rv = nsXPConnect::GetXPConnect()->GetJSObjectOfWrapper(aCx, obj, &obj);

So this call, among other things, strips security wrappers (including cross-compartment wrappers) off of |obj|...

>-    if (!member->NewFunctionObject(inner_cc, iface,
>-                                   JSVAL_TO_OBJECT(argv[0]),
>-                                   &funval))
>+    if (!member->NewFunctionObject(inner_cc, iface, obj, &funval))

which is precisely why the old code here passes the original argument, rather than obj (which has been clandestinely mutated).

So that needs to be fixed. This was hard to catch because the change in obj happened without any explicit assignment. So let's fix it by introducing a new JSObject called unwrappedObj (or something), and doing GetJSObjectOfWrapper(aCx, obj, &unwrappedObj);

Comment 9

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

Eek! Thanks for catching that.

Comment 10

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

https://hg.mozilla.org/mozilla-central/rev/50b70f639c2c