Another sign/clearsign related bug?



On Fri, Sep 20, 2013 at 7:55 PM, Raymond Toy <toy.raymond at gmail.com> wrote:

>
>
>
> On Fri, Sep 20, 2013 at 10:25 AM, Raymond Toy <toy.raymond at gmail.com>wrote:
>
>> No problem.  I can't remember exactly what link I used before, but
>> http://git-scm.com/book/en/Git-Tools-Debugging-with-Git is a pretty good
>> start.
>>
>> However, in this particular case it might be tricky because the current
>> code is a good revision and the old code is a bad revision. I'll try this
>> out myself to see if I can get it started and let you know.
>>
>>
> It is tricky.  The current version is good and the old version is bad,
> unlike the typical case.  So, when I try that git complains that the good
> version is not an ancestor of the bad. (Don't see why that matters too
> much.)
>
> Anyway, my solution was to flip the meaning of good and bad.  git seems to
> be happy with that and is now bisecting.  Just gotta remember that when I
> get the wrong result, I have to say git bisect good.
>
> 58 revisions left...
>
> I'm going to stop here and let you take it.  Let me know if you have
> problems.  We'll work it out.
>

Well, I decided to do it since building maxima is fairly fast on my machine
and the test is very simple.  The whole process took about 20-30 minutes,
mostly waiting for maxima to compile.

According to git bisect the revision that made it work is

commit b650fb35311b33070c1bcc9c640d66671e89a7d6
Author: Rupert Swarbrick <rswarbrick at gmail.com>
Date:   Wed May 22 12:15:05 2013 +0100

    Fix bug in asksign

    I introduced this bug when refactoring the code in
    February. Basically, I hadn't realised that when someone calls
    asksign(x^2), we actually ask for the sign of x (asking whether it's
    zero or nonzero) and then treat any negatives as positives. This patch
    puts the behaviour back in.


Ray