Ghostbusting (graphical schematics differences)

Werner Almesberger werner at openmoko.org
Thu Oct 7 14:19:38 EDT 2010


[ Copied to the list. Context: there are occasionally differences that
  don't correspond to any change in the project. It's particularly
  visible in the AUDIO sheet column of
  http://projects.qi-hardware.com/schhist/board-qi-avt2/
  For the past days, I've been trying to track them down. ]

Hi Wolfgang,

I finally found a clue as to what is causing the ghosts: it seems to
be related to memory use in eeschema, maybe an uninitialized variable
or memory corruption.

What changes are some line widths that more or less randomly can be
60 or 120 units (whatever the unit is in this case, probably
deci-mil).

This can be seen in, say,
http://projects.qi-hardware.com/schhist/board-qi-avt2/c3a9386ef1676a4bb5869110f97ba440dd6e4c4e/pdf/AUDIO.pdf

When rapidly switching the pages, the thickness of the pin names of
the FETs changes slightly.

This happens in the Postscript output generated by eeschema, before
all the processing schhist2web does on the data. I only see it happen
on fidelio (the build server) but not on my PC at home. Consecutive
eeschema --plot runs with identical inputs have a roughly equal
probability of producing either result. I don't know yet where the
bug is, but I can make it go away by disabling heap randomization on
fidelio.

As a band-aid, I would suggest to
sysctl -w kernel.randomize_va_space=0
and to rebuild all the schematics histories.

To make the change survive reboots, it would have to be put into
/etc/sysctl.d/local.conf as well. (Not sure if this is sufficient to
make the setting stick. One would have to do a reboot to verify that
nothing else overrides it later.)

- Werner




More information about the discussion mailing list


interactive