pcbdiff fails due to image size mismatch

[ldoolitt]

Even when two boards are the same size, the pcbdiff script sometimes generates images of differing size.
That in turn causes ImageMagick's "composite -stereo" command to fail, and no visual diff is shown.
I can avoid the effect in my use case by removing the "--only-visible" flag to pcb.

The attached pair.tar.gz includes two pcb files, check11.pcb and check12.pcb. To reproduce, "pcbdiff check11.pcb check12.pcb". Sorry the files are so big.

I can believe this behavior is triggered by graphical elements hanging off the edge of the nominal board area, e.g., the line forming the board outline is 10 mils wide and therefore hangs out by 5 mils.
At least for incomplete board designs -- of the kind that I want to track in version control -- sloppiness around the board edge is commonplace for me.

[ldoolitt (ldoolitt)]

Patch attached.
While I was in there, I made the 200 dpi default over-ride-able with the environment variable PCBDIFF_DPI.

[Traumflug (mah-jump-ing)]

Comments on geda-users:

Markus Hitter:

At the first glance I'm not sure wether this is the right fix, it might be a better idea to fix the reason for the differently sized PNG exports.

Kai-Martin Knaak:

> It's not obvious to me what --only-visible is supposed to mean
> to pcb in a batch context.

I use this option in my offline printing script.
   http://www.gedasymbols.org/user/kai_martin_knaak/layoutdruck.sh
It crops the print to the extents of whatever is in teh layout.

However, the option does not work for a print of the bottom side aka
"solderside". Instead, I achieve the same by feeding the postscript file to
epstool.

[Traumflug (mah-jump-ing)]

Another thing is, all options usable in the GUI are provided on the command line, too, without explicitely coding it this way. I'm not even sure wether it's possible to remove a command line option while keeping it in the GUI.

That said, the question here is wether pcbdiff needs this option.

[Traumflug (mah-jump-ing)]

Surprise! Trying the two testcase files I get this:

mah@piccard:~/RepRap/pcb$ tools/pcbdiff '/home/mah/RepRap/pcb/PCB.00002935.backup' '/home/mah/RepRap/pcb/PCB.00020492.backup'
Segmentation fault (core dumped)
convert: no decode delegate for this image format `/tmp/pcbdiff.IlNs3iHPZ3' @ error/constitute.c/ReadImage/532.
convert: missing an image filename `/tmp/pcbdiff.IxCs5f7Wby' @ error/convert.c/ConvertImageCommand/3011.
composite: no decode delegate for this image format `/tmp/pcbdiff.IxCs5f7Wby' @ error/constitute.c/ReadImage/532.
composite: missing an image filename `/tmp/pcbdiff.Eb21XgpbhO' @ error/composite.c/CompositeImageCommand/1616.
display: no decode delegate for this image format `/tmp/pcbdiff.Eb21XgpbhO' @ error/constitute.c/ReadImage/532.

It's pcb which crashes on the empty layout. Viewing it in the GUI works.

[Traumflug (mah-jump-ing)]

I think it's fixed now. See these two:

http://git.geda-project.org/pcb/commit/?id=46cba84697759663f27b653d62d517dcbfeab5cb
http://git.geda-project.org/pcb/commit/?id=ab6ebc701d8ab6fbc67c5232215d28a1fa3eae0a

Larry, sorry for reviewing the patches so much, I'm a nitpicker. ;-)

[Bert Timmerman (bert-timmerman)]

And now for something completely different:

I copied/modified a blinking pcb graphics diff.

Kind regards,

Bert Timmerman.

[Traumflug (mah-jump-ing)]

> And now for something completely different

Huh? Please don't hijack bugs. You can open a new one at any time.