[mdlug] C Pointer Problem

[A. Zimmer]

On Fri, 10 Apr 2015 20:40:32 -0400
Gregory Czerniak <gregczrk at gmail.com> wrote:

>
> #define peiGetConvKer(image)    (*(imageType**)&(image)->clientData[1])
>
>  My guess is
> that it's finding the address of "image", dereferencing clientData[1]
> (getting the integer), type-casting this integer as a double imageType
> pointer, and then de-referencing.  By doing that, you'd end up with a
> single imageType pointer pointing to whatever that integer value was.
>

Thanks for this response.

Your description does make sense, especially when we consider how the
value is being initially stored, as seen in this macro definition:

#define peiSetConvKer(image,ker)	(*(imageType**)&(image)->clientData[1])=(ker)

Here, the variable "ker" is actually a pointer to the "imageType"
structure and the parentheses around ker, (ker), would mean that
a pointer to a pointer is being stored in the array.  But the
clientData[] array stores 32-bit integers, which is compatible with
pointers on 32-bit systems but differs from the 64-bit pointers
on 64-bit systems.  This probably contributes to cause of the seg faults.

Anyway, I've decided to abandon the conversion project.  I would have
to spend way too much time reverse engineering the code to get a better
understanding of what is actually happening before I could even begin
to refactor.  Since I can always boot into a 32-bit system using a USB disk,
such as Gentoo LiveDVD, I can still use the program whenever I need it.

It would be nice to have a pure 64-bit version, however,  The program,
called peiHalfTone, is used to completely automate the process of
descreening grayscale digital images and its results are the best I've
seen not only in the open source Linux world but anywhere else as well.

In case you may be interested, I have posted an example of its
performance.

An original image containing a very obtrusive halftone pattern
is shown here:

http://s16.postimg.org/5f9435xmt/elevator_orig.jpg

After processing with peiHalfTone, the result is shown
here:

http://s10.postimg.org/ov8k02eeh/elevator_conv.jpg

Notice how the halftone pattern is virtually eliminated with
all frequency measurements being done automatically.

As a comparison, processing with a custom FFT frequency mask
that I hand-crafted myself is shown next:

http://s12.postimg.org/bxtlgq6ml/elevator_fft.jpg

This last result, which required the manual creation of a frequency
mask, is close but not as good.  Spending more time would have
improved the result but using peiHalfTone is nearly instantaneous.

So thanks again for your input.  I think that I have improved my
ability to understand complex pointer constructions but it's
doubtful that I'll ever encounter anything similar.  Most people
seem to prefer more straightforward code.