Bug 11184 – fixing failed two failed mozilla Date reg tests on windows

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Ian Eng]

Two failed test cases:
ecma/Date/15.9.5.2-1.js,
ecma/Date/15.9.5.2.js

That's because Date.toString() doesn't adjust DST when computing UTC offset. 
For example, on WinXP, DST is set, following problem produces wrong results.

var now = new Date();
print(now.toString());
var then = new Date(now.toString());
print(then.toString());

When DST is set, it produces:
--> Thu Oct 05 2006 23:56:32 GMT-0800 (Pacific Daylight Time)
--> Fri Oct 06 2006 00:56:32 GMT-0800 (Pacific Daylight Time)

The problem comes from "now.toString()", it should be 
--> Thu Oct 05 2006 23:56:32 GMT-0700 (Pacific Daylight Time)

and when DST is not set, it is 'GMT-0800'.

I think the problem is in date_object.cpp, gmtoffset() function doesn't consider DST. I made a simple patch to fix it, but I am not expert on either KJS or Date. Someone please review it carefully.

[Ian Eng]

Created attachment 10943 [details]
get correct GMT offset when DST is on

[Darin Adler]

Comment on attachment 10943 [details]
get correct GMT offset when DST is on

We should not be doing a const_cast like this. If we need to make a change like this, then dateToMseconds should be changed to take a const tm*.

(And, in passing, "dateToMseconds" is a strangely-capitalized name; we should fix that too. And it's very strange that isDST takes a const double& instead of a double. But neither of those has anything to do with this patch.)

[Kevin McCullough]

This is my fault, and a fix will be in place today.  For performance and accuracy reasons the gmtoffset function will be reverted to it's previous implementation.

[Ian Eng]

Created attachment 10949 [details]
revised patch

Darin, thanks for your review. I took your suggestion and changed the first parameter of dateToMseconds to "const tm&". I did grep the function name in *.h and *.cpp under WebKit, all references found are in date_object.cpp. And these calls are fixed. I hope no other references are missed.

As you mentioned that renaming of several other functions has nothing to do with this patch. I am not going to touch that part.

[Kevin McCullough]

Comment on attachment 10949 [details]
revised patch

I think using const reference is a great idea.  But gmtoffset in date_object.cpp has changed.  Please see revision 16855.

[Ian Eng]

(In reply to comment #5)
> (From update of attachment 10949 [details] [edit])
> I think using const reference is a great idea.  But gmtoffset in
> date_object.cpp has changed.  Please see revision 16855.
> 

Cool, thanks.