Regression using last version(s) of ImageMagick

[Sean Enck]

On 2018-10-13, Steve Allewell wrote:
> On 12/10/2018 20:42, Sean Enck wrote:
> > On 2018-10-08, Sean Enck wrote:
> >> It appears importing an image has regressed (no longer works) against 
> >> kxstitch 2.1.1 when using the latest (well archlinux latest) versions of 
> >> image/libmagick. The current (not working version) in arch:
> >>> imagemagick-7.0.8.12-1-x86_64.pkg.tar.xz
> >>> libmagick-7.0.8.12-1-x86_64.pkg.tar.xz
> >>
> >> I downgraded my versions to verify and import worked (no crash):
> >>> imagemagick-7.0.7.13-1-x86_64.pkg.tar.xz
> >>> libmagick-7.0.7.13-1-x86_64.pkg.tar.xz
> >>
> >> The steps are:
> >> 1. open kxstitch
> >> 2. go to import image
> >> 3. select an image (any image, I haven't seen it work)
> >> 4. doesn't matter if you do anything before hitting "OK" when the image 
> >> would go to the normal grid
> >> 5. full crash, can provide a trace if there is anything special in kxstitch 
> >> or any specific tracing method is preferred.
> >>
> >> Thanks for any help/insight,
> >> --Sean
> > 
> > I was able to do some digging and found the issue started around ImageMagick 
> > 7.0.7.24 (works) -> 7.0.7.25 (broken). I traced it to the FlossScheme.cpp 
> > color matching where nullptr was being returned and I patched this in 
> > Arch using the referenced patch below.
> > 
> > https://aur.archlinux.org/cgit/aur.git/plain/0001-Change-nearest-color-lookup-logic-for-finding-floss-.patch?h=kxstitch
> > 
> Hi Sean
> 
> Apologies for not getting back to you earlier.  I noticed you sent a
> message to the list on the 8th, but it doesn't look like I got a copy of it.
> 
> I have version 7.0.8 of ImageMagick too and confirm the bug.
> 
> I had a look at the patch and it does the job, although it does not
> require float for near and closest; abs and QColor red, green, blue
> functions all return int.
> 
> I did a little rewrite which works for me, please check it for yourself:
> -------------------------------------------------------------------------------
> Floss *FlossScheme::find(const QColor &color) const
> {
>     QListIterator<Floss *> flossIterator(m_flosses);
> 
>     Floss *matched = nullptr;
>     int closest = 100;
> 
>     while (flossIterator.hasNext()) {
>         Floss *floss = flossIterator.next();
>         QColor c = floss->color();
> 
> 	// the color mapping may not be perfect so search for a near match.
>         int distance = abs(color.red()-c.red()) +
> abs(color.green()-c.green()) + abs(color.blue()-c.blue());
> 
>         if (distance < closest) {
>             matched = floss;
>             closest = distance;
>         }
>     }
> 
>     return matched;
> }
> -----------------------------------------------------------------------------
> 
> Do you want to commit the change?
> 
> I will have a look to see what changed in ImageMagic between those two
> versions and see if there are any improvements to make.
> 
> Regards
> 
> Steve
>

Thanks for another set of eyes on it!

I initially emailed (without troubleshooting more than what I put) 
because I didn't have the time until (days) later to look into the code.
I'm not (generally) a C++ guy so I need a little extra time to try
things (before realizing I could just use the call you had as a the
criteria, I tried another which did require a float and didn't go back
once I realized what you already had was an int approximation... I just
needed to track it). I started trying to see what changed in ImageMagick
but I have no idea what kxstitch uses from it so it was more of "where, 
in the code does this seem to be happening?"

I applied your update locally and it appears to work just the same as my
initial patch (aka bug still gone).

Happy to commit the (joint) change, I wasn't sure what the process 
is/was? I initially used format-patch to try and bundle it fully as a
consumable idea/be able to patch the arch packaging.

--Sean