RE: question on wrong result for small numbers from strtod(), strtof()

["Howland Craig D (Craig)" <howland at LGSInnovations dot com>]

Daniel:
     I did get a chance to look some more, briefly.
     Here's a better repair than the first attempt--although my guess is
that it is incomplete.  The first patch was trying to fix a mistake in
parsing after it had been done.  This second attempt makes it read
properly--fixing the parsing instead of trying to clean up afterwards.
However, it does not address things that get done in some of the support
files.  So while it will fix one problem, I don't know if it solves the
entire issue.  Give it a try and let me know the results.
     If it does not fix the problem, please give the details of the
s2b call at line 692 for a failure case.  (That's the first external
call that I suspect might still be a problem, but which I have not
yet examined closely.)
				Craig

-----Original Message-----
From: Howland Craig D (Craig) 
Sent: Monday, October 25, 2010 12:27 PM
To: 'Daniel Calcoen'; newlib@sourceware.org
Subject: RE: question on wrong result for small numbers from strtod(),
strtof()

So I fixed one case and broke some others.  I'll think about it some
more, but I probably won't get to it for a few days.
 
It is possible for you to use your double-is-64-bits option on a
permanent basis?  (The strtod code was originally written assuming
64-bit double.  Tracking this down might be very difficult.)

-----Original Message-----
From: Daniel Calcoen [mailto:Daniel.Calcoen@cern.ch] 
Sent: Monday, October 25, 2010 11:58 AM
To: Howland Craig D (Craig); newlib@sourceware.org
Subject: RE: question on wrong result for small numbers from strtod(),
strtof()

I checked twice that the patch was correctly applied, I rebuild newlib
and
run the test

This is not the correct patch, now I have more wrong values [tests
4,5,12,13,14,18,19 and 20]

...

Attachment: strtod.patch
Description: strtod.patch