hid_png3 test fails on i386 arch

[أحمد المحمودي (Ahmed El-Mahmoudy)]

When building pcb 20110918 on i386 architecture, make check fails at hid_png3 test:

Test: hid_png3
../../../src/pcbtest.sh -x png --dpi 300 ../../tests/inputs/gerber_oneline.pcb
FAILED: See outputs/hid_png3/mismatch
FAIL

Attached is a tarball of outputs/hid_png3/mismatch contents

[Juergen Harms (juergen-harms-s)]

Same problem here: the faulty png has one (void) pixel column less in the lead, and an additional column in the right-hand line of the layout)

[Andrew Poelstra (asp11)]

The conversion to nanometers changed the rounding done in the PNG conversion. This is not a bug. The test needs to be changed.

[Andrew Poelstra (asp11)]

Fixed. The commit robot missed it because I typed "Fixes-bug" instead of "Closes-bug" again.

commit af27c2d529a00eb26ec9dcd1e102d7f6c91d7683
Author: Andrew Poelstra <email address hidden>
Date: Sun Dec 18 12:31:39 2011 -0800

    Increase DPI of test hid_png3

    Since the nanometers conversion, hid_png3 has failed due to rounding
    changes. I have changed the test's output DPI from 300 to 600, to
    decrease the chances of this happening in the future.

    Fixes-bug: lp-860037

[Vladimir Zhbanov (vzhbanov)]

Reopen this bug as it still hits me on my both 32-bit i686 machines.

[أحمد المحمودي (Ahmed El-Mahmoudy) (aelmahmoudy)]

Yes, version 20140316 still fails for i386, kfreebsd-i386 & hurd-i386 archs on Debian.

Also the hid_gcode3 test fails for hurd-i386 arch on Debian.

Build logs:
i386: https://buildd.debian.org/status/fetch.php?pkg=pcb&arch=i386&ver=20140316-1&stamp=1395075417
hurd-i386: https://buildd.debian.org/status/fetch.php?pkg=pcb&arch=hurd-i386&ver=20140316-1&stamp=1395076424
kfreebsd-i386: https://buildd.debian.org/status/fetch.php?pkg=pcb&arch=kfreebsd-i386&ver=20140316-1&stamp=1395077313

[gpleda.org commit robot (gpleda-launchpad-robot)]

Bug was fixed by a commit
git master commit 137e17afdcf21bcff53767ac9412ffb7535ca706
http://git.geda-project.org/pcb/commit/?id=137e17afdcf21bcff53767ac9412ffb7535ca706

commit 137e17afdcf21bcff53767ac9412ffb7535ca706
Author: Andrew Poelstra <email address hidden>
Commit: Andrew Poelstra <email address hidden>

    png: add explicit round() calls to the scaling

    After this commit the tests pass on both 32- and 64-bit builds.
    Hopefully this time it's really dead; the previous "fix" af27c2d
    by myself merely increased the DPI of the output, but this one
    attempts to get deterministic rounding behaviour.

    Tests from folks with actual 32-bit machines are welcome.

    Closes-bug: lp-860037

[DJ Delorie (djdelorie)]

Could you please confirm that the actual sizes of traces have not changed in the common case? I took great care in the png math to ensure that, for example, a 10 mil trace at 600 dpi was always 6 pixels across.

[Andrew Poelstra (asp11)]

DJ, Yes, I just checked and the pixel sizes are precisely correct. (Given that I had to change the golden file for this latest commit, this probably means that I had broken things, at least on 64-bit systems, in the nanometers conversion and when I changed the test to 600dpi. Sorry for not being careful then.)

Ahmed, I can reproduce the gcode bug on Linux with a 32-bit build. I've been chasing it for hours and hopefully will have tracked it down by tonight (so we will have passing tests for Debian on all systems :)). In the case that I don't fix it soon, it would be much appreciated if you open another bug for it. Maybe I'll do this... or maybe there already is one.

[أحمد المحمودي (Ahmed El-Mahmoudy) (aelmahmoudy)]

gcode bug is on: https://bugs.launchpad.net/pcb/+bug/1074268

[Juergen Harms (juergen-harms-s)]

I also got stuck on this bug, building pcb for the Mageia distro. I can confirm that the 20140316 stable release does not suffer from this bug any more, and that the patch against this bug (stable Mageia using the 2011 release) is not required any more.

[Bert Timmerman (bert-timmerman)]

Commit is released in pcb-20140316.