Bug 4142 – Date object does not always adjust daylight savings correctly

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Carsten Guenther]

This will fix

    ecma/Date/15.9.4.3.js
    ecma/Date/15.9.5.28-1.js

There seems to be a problem with daylight savings when a time change crosses the daylight savings start/
end dates.

[Geoffrey Garen]

(In reply to comment #0)

FYI, Mac IE has the same problem. Win IE and Mac FF don't.

[George Staikos]

Adding this before the return in KJS::makeTime() fixes it for me:  
  
    // Determine if we passed over a DST change boundary  
    if (!utc) { 
      time_t tval = mktime(t) + utcOffset + int((ms + yearOffset)/1000);  
      struct tm t3;  
      localtime_r(&tval, &t3);  
      t->tm_isdst = t3.tm_isdst;  
    } 
  
Does not seem to introduce any regressions.  I am a bit worried about overflow  
and rounding though.  It might break down very close to the changeover period.  
  

[Geoffrey Garen]

Created attachment 4271 [details]
Patch

Implemented George's suggestion.

[Darin Adler]

Comment on attachment 4271 [details]
Patch

What's the rationale for calling the local variable "t3"?

The FIXME says "on some systems", but this will happen on all systems; there's
no system where time_t has enough range to include all the years we want our
function to handle. So I suggest re-wording it. Otherwise, looks fine.

[George Staikos]

There is already a t and a t2, and although t3 wasn't perhaps the best sequel, 
I didn't see fit to leave it out. :-) 
 
And yes, "all systems" is correct.  I don't know if it's a realistic problem 
to put more effort into at this time though. 

[John Sullivan]

I checked this in. It did fix ecma/Date/15.9.5.28-1.js, but ecma/Date/15.9.4.3.js was apparently fixed by 
some earlier checkin.