104653 - NaN's appear in download dialog.

Status: VERIFIED WONTFIX

(Core Graveyard :: File Handling, defect)

Description

[Andrew Simmonds]

NaN's currently show all over the download dialog instead of numbers, say 20%
remaining it will be NaN% remaining this happens at random throughout the whole
download to other parts such as time remaining and elapsed time.

I'm pretty sure that compiling with gcc3 is the cause of this problem.

A picture of one such dialog will be attached.

Comment 1

[Andrew Simmonds]

Attachment: dialog with a few NaN's in it.

Comment 2

[Boris Zbarsky [:bzbarsky]]

Please try a new build.  This was fixed on the evening of Oct 11.

If you are still seeing it with an Oct 12 or later build, please reopen.

*** This bug has been marked as a duplicate of 70859 ***

Comment 3

[Andrew Simmonds]

I clobber built last night.
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:0.9.5+) Gecko/20011014

This a little different to that bug in which its happening to every single
number shown on the dialog not just one. Aswell as ftp and binary downloads.

Comment 4

[R.K.Aa.]

Just clobbered and built on linux, RH7.1. Downloading behaves just fine.
Not a NaN in sight. WFM.

Reporter: Are you building optimized? Please add details.

Comment 5

[Andrew Simmonds]

As I said earlier I built with gcc 3.0

Comment 6

[Andrew Simmonds]

and opt -O2

Comment 7

[R.K.Aa.]

Building with -02 here too. Adding comment about this bug in bug 53486

Comment 8

[R.K.Aa.]

Reducing severity:
This isn't happening with the default compiler, and  bug 70859 (about same
error) had severity trivial.

Comment 9

[John Morrison]

this is very likely the consequence of bug 102725 "gcc -O2 problems converting 
numbers to strings". In that case, when js is compiled -O2, the expression 
'(new Date()).getTime().toString()' will return something like ":03121132921",
so anywhere that you 'subtract' a stringified number from a number, the result
will be 'NaN'.

You can work around it, probably, with the setting of no-strict-aliasing 
described in the other bug. A fix is being developed.

Comment 10

[Andrew Simmonds]

Thanks alot John, i'm now right on the track of the problem.
It appers to be a problem in js/src/jsdtoa.c
If I manually compile that file with -O1 the bug disappears.
Looks like I might need to further hunt through that file and volatile a few
bits and pieces and see what happens.

Comment 11

[Bill Law]

I'm resolving this.  If there's anything that does need to be done, reopen and
assign to the person who will be doing it.

Comment 12

[John Morrison]

I expect that with bug bug 102725 fixed, then this is no longer an issue.

Comment 13

[Andrew Simmonds]

verified now to be fixed due to bug 102725 checkin

Comment 14

[Matt Foster]

Erm...I'm still seeing this on cvs build from 20020110.  RH7.2, gcc 2.96 with
-O2 opt enabled.  Have I done something wrong, or is this a regression?  The js
bug 102725 has supposedly been fixed, so I'm confused.