Partly revert the following commit:
commit 3090983c31
Author: Ron Sluiter <rsluiter@users.noreply.github.com>
Date: Wed Jul 29 15:50:23 2015 +0000
Bug fix for target position (VAL/DVAL/RVAL) initialization error
when the motor record is configured to do retries.
And from the release notes:
6) Kevin Peterson discovered an initialization bug when the motor record is
configured to do retries. When configured to do retries, the motor
record issues incremental rather than absolute moves. If the motor
behaves badly (e.g., a piezo stiction motor) the controller's absolute
step count can be far from its' absolute readback position. The motor
record initialization logic that determines how the target positions
are initialized at boot-up was not taking into consideration whether
or not the motor record is configured to do retries, and therefore,
whether or not absolute or incremental moves were issued by the motor
record. With this release, if the motor record is configured to do
retries, then the controller's absolute position is ignored (because
the controller's absolute position was "corrupted" by retries) and the
autosave/restore position is used to set the controllers position.
Switching between retries on and off (relative vs. absolute moves)
between reboots will cause unpredictable results and is not covered
by this bug fix.
Files modified: motor_init_record_com() in MotorSrc/motordevCom.cc
init_controller() in MotorSrc/devMotorAsyn.c
Commit 3090983c improves the situation for setups where autosave is
used, and makes things worse when autosave is not used.
When autosave is not used and the motor is configured to do retries,
the the DVAL field is loaded into the controller regardless.
And because DVAL is 0.0 at startup, the controller position is lost.
The first issue report is found here:
"Problem with R6-10 issue 6 fix - controller position can be lost on reboot"
https://github.com/epics-modules/motor/issues/85
And the we have another issue report here:
"IOC zeroes encoder on startup"
https://github.com/motorapp/Galil-3-0/issues/13
A possible way forward is discussed here:
"Add a field to the motor record to allow always restoring the
autosaved position"
https://github.com/epics-modules/motor/issues/151
This commit adds the "RSTM" field:
- 0 Disables restoring the autosaved position
- 1 Always restores the autosaved position
- 2 Uses the same logic as motorRecord 6.9 (or older)
- 3 Uses the same logic as motorRecord 6.10
This numbering maps 0 and 1 somewhat to false/true, uses 2/3 for
special handlings and leaves room for more choices.
E.g. "use the encoder value, if valid, fall back to autosave if not.
Or "use the URIP/RDBL" if possible.
I am getting off-topic,
those improvements should be done in later commits anyway.
The 2nd and 3rd parameter in send_mess() can and should
be a 'const char *' instead of just 'char *'.
Modern compilers complain here, so that the signature now
gets the const.
Update drivers from the following list to use the new send_mess():
modules/motorAcs
modules/motorAcsTech80
modules/motorAerotech
modules/motorFaulhaber
modules/motorIms
modules/motorKohzu
modules/motorMclennan
modules/motorMicos
modules/motorMicroMo
modules/motorNewFocus
modules/motorNewport
modules/motorOms
modules/motorOriel
modules/motorPI
modules/motorParker
modules/motorPiJena
modules/motorSmartMotor
modules/motorThorLabs
And while there, fix one more "const char *" in motordrvCom.cc
This is from Matthew Pearson, pearsonmr@ornl.gov
A number of users and staff have requested that the motor record have a
setpoint deadband field (eg. SPDB), which prevents any motion from
happening if the readback position is within SPDB.
Without a SPDB we have issues with motors moving unnecessarily.
For example, if the RBV is 0.001 and we set VAL to 0.0, if we have a
non-zero backlash correction then the motor will move unnecessarily
to take out backlash.
Even without backlash correction we often enable and disable the
drive amplifier for no good reason.
Currently the effective setpoint deadband is equal to MRES,
but this is often too small to be an effective setpoint deadband.
The new SPDB field would default to 0.0
The new code hooks into do_work():
When the new setpoint is within DPDB, set too_small and don't move.
The new code does not use snipptes like "abs(npos - rpos)", these
already produce warnings on 64 bit sytems.
abs() is declared to work on int, but we feed long values.
Simply compare the coordinates in engineering units.
Mark Rivers