Closed PeterBowman closed 4 years ago
Looks good! I'm going to be touching RobotServer
/RobotClient
pair at https://github.com/asrob-uc3m/yarp-devices, would look like a nice opportunity to experiment!!
How to make this work on RF modules (e.g. transCoords app):
int main()
{
MyMonitor monitor; // implements yarp::os::SearchMonitor
yarp::os::ResourceFinder rf;
// tweak 'rf', e.g. set default context and so on
rf.setMonitor(&monitor, "transCoords");
roboticslab::TransCoords mod;
mod.configure(rf); // check return value!
// at this point, retrieve options stored in 'monitor' and
// print them on stdout if --verbose was passed on
return mod.runModule(); // no need to configure again
}
We'll need a class that implements yarp::os::SearchMonitor
(no Doxygen, see definition) and provides a public method for obtaining option bottles. Compare YarpDevMonitor
.
Thanks a lot for this info! Always loved the behavior, but found the intrinsics quite cryptic, especially for modules.
Bad thing is, YARP should have already provided an implementation, otherwise we are forced to write a tiny library only for this and replicate the same code in several places... This is what yarpdev
runs behind the scene: ref. The sample main
function from my previous comment would check verbose
and reuse that code to print the report (replace PolyDriver
with MyMonitor
in the function's signature).
Maybe we could develop some kind of helper class that could be PR'ed directly into YARP?
Sure, the sooner the better. I'm afraid of those find_package(YARP 3.0.0 REQUIRED)
in each single app (unless we stick to preprocessor conditionals).
need to review all check calls and add missing comments/defaults
Speaking of the yarp::os::Searchable class (see inheritance diagram, it has several implementations), prefer
bool check(const ConstString &key, const ConstString &comment) const;
Bottle findGroup(const ConstString &key, const ConstString &comment) const;
instead of
check(const ConstString &key) const;
Bottle findGroup(const ConstString &key) const;
The former generate a report that may be used by the monitor in verbose mode. Also, the following are available:
bool check(const ConstString &key, Value *&result, const ConstString &comment="") const;
Value check(const ConstString &key, const Value &fallback, const ConstString &comment="") const;
I'm going to list here all repos that have been improved by providing or expanding on option documentation (now passing appropriate defaults and description in config.check()
and the like) and propagating it through device wrappers:
Thanks a lot for this!
Sometimes, device options hide each other, like this:
if (config.check("optionA"))
{
yarp::os::Value v = config.check("optionB", yarp::os::Value(MY_DEFAULT));
}
The check for optionB
could be pulled out from its enclosing if
clause so that it's parsed by --verbose
regardless of the presence of optionA
. Then, additional logic would take care of the implicit relationship between both options. Is this the way to go?
Or, since I like simplicity:
if (config.check("optionA", "enables A (and additional options)"))
{
// ...
}
Thus, users would be aware of the existence of hidden options (and how to enable them).
Custom devices wrapped by controlboardwrapper2
cannot display options via --verbose
due to an upstream bug (tracked at https://github.com/robotology/yarp/pull/1608).
Related upstream PR regarding --help
: https://github.com/robotology/yarp/pull/1473.
I'm going to list here all repos that have been improved...
All set but openrave-yarp-plugins. @jgvictores I'd really appreciate if you could take a look at existing device options and review/fill their description fields :) (low priority).
Okay, it'll take a while, but I will find a moment to do it!
yarp-devices still WIP due to https://github.com/roboticslab-uc3m/yarp-devices/issues/175, marking as blocked.
Note to self: use simpler unit notation and replace parens with square brackets. Example: [m/s] or [deg/s]
instead of (meters/second or degrees/second)
.
Added some documentation on ORYP at https://github.com/roboticslab-uc3m/openrave-yarp-plugins/commit/e0adfbc6953659a5d20568eee27c1c6691e279e2
Applied to asrob-uc3m/yarp-devices
here; works, but ran into https://github.com/asrob-uc3m/yarp-devices/issues/12
I'd rather close https://github.com/roboticslab-uc3m/yarp-devices/issues/175 and track that bug here. To sum up, PolyDriver::getValue
does not work as expected when the instance has been configured and opened with a Searchable
object to which a monitor has been attached to (usually via options.setMonitor(config.getMonitor())
). See https://github.com/roboticslab-uc3m/yarp-devices/issues/175#issuecomment-375406583 for full explanation. Might need to circumvent the way we use getValue
in that repo (example) or enhance the value lookup mechanism upstream. For that reason, I'd keep the blocked
label for now.
There is no simple way to sort this out on the upstream side without proposing a breaking change of the (undocumented) SearchMonitor
interface, which lacks option getters (check sources).
On our end, the CanBusControlboard device might hold a vector of Property
instances besides the existing one for PolyDriver
objects (ref) in order to store and access device configurations. This is a hack.
IMHO, a big issue is too much being done on the CanbusControlboard. Things such as home should be done within Technosoft open or new method which is involved for all devices and each one implements their behavior via polymorphism.
The only special case where I introduced the original ugly hackish code was connecting the CuiAbsolute to the TechnosoftIpos, but that (apart from different) could be done better.
Custom devices wrapped by controlboardwrapper2 cannot display options via --verbose due to an upstream bug (tracked at robotology/yarp#1608).
Merged into YARP's master
branch, scheduled for the 2.3.72.1 release.
This issue has been stalled for too long, therefore...
I'd rather close roboticslab-uc3m/yarp-devices#175 and track that bug here.
I changed my mind (again) and opened https://github.com/roboticslab-uc3m/yarp-devices/issues/207, in the only repo affected by said bug.
The list https://github.com/roboticslab-uc3m/questions-and-answers/issues/49#issuecomment-372620773 is done except for a few undocumented config options in openrave-yarp-plugins. @jgvictores could you please take a look? Apart from that, I need to check whether any new devices that have been added in the past year conform to these new standards.
One last question: is the behavior of --verbose
actually better than the old --help
? Also, note to self: look for remaining occurrences of the latter.
Regarding RFModule: either open an issue upstream considering https://github.com/roboticslab-uc3m/questions-and-answers/issues/49#issuecomment-368339553, or forget.
The list #49 (comment) is done except for a few undocumented config options in openrave-yarp-plugins. @jgvictores could you please take a look?
I have that repo pretty abandoned, but I'll take a look. Specifically, I had done advances on https://github.com/roboticslab-uc3m/openrave-yarp-plugins/issues/82, raised its priority to get done with it.
One last question: is the behavior of
--verbose
actually better than the old--help
? Also, note to self: look for remaining occurrences of the latter.
My intuition behind --help
was to be implement a --verbose
that additionally quits before running anything, mostly to be able to read the output before all the verbosity of actually running. Would implementing this explicitly make any sense, or do you think we can just force users to stop a running program and scroll up?
There is not much to do at or-yp, anyway, just a few undocumented occurrences of .check(
at OpenraveYarpPluginLoader.
Would implementing this explicitly make any sense, or do you think we can just force users to stop a running program and scroll up?
It's a pity we don't know the outcome of https://github.com/robotology/yarp/pull/1473. Some thoughts:
--help
would need to invoke --verbose
under the hood.--verbose
itself is position-dependent, that is, if we fail to pass some check in DeviceDriver::open
and end up in a return false;
clause, all further instructions won't be reached, especially accounting for those documentation comments passed to Searchable::check
. In other words, the output of --verbose
may be incomplete.::open
method: either parse all YARP options and store them in local variables before using them in conditional checks or whatever (i.e. register doc comments before doint anything else), or defer the early exit of a failed configuration (let return false;
be the very last instruction in said method), respectively. Such a big change would enforce a new paradigm of writing YARP devices which I'm not sure to be fond of, but it's definitely doable.--help
to behave? Should it output additional, textual clues which are not specific to YARP options shown by --verbose
? Must it actually propagate to subdevices, just like --verbose
does, now?PolyDriver
and RFModule
. Our implementation of a --help
handler could differ between those, also we need to know whether YARP catches this option anywhere and enforces its own solution (see linked PR above).To sum up:
check
calls: same as abovegetValue
bug (https://github.com/roboticslab-uc3m/yarp-devices/issues/207): solved--help
versus --verbose
: not sure how to proceed, we use neither of both anyway...Closing due to inactivity/too broad a scope/lack of interest on tedious subtasks.
The
yarpdev
utility suggests using a--verbose
CLI parameter in order to list all available options, their description, default value, etc. This comes in handy when working with our own devices, which undergo frequent changes in their usage (or, simply put, are barely or not documented at all). Traditionally, we introduced the--help
parameter, but it's hard to maintain (needs hardcoding plus some care to keep it up to date).YARP offers its own mechanisms for such tasks, tightly bound to the
yarp::dev::PolyDriver
class: theYarpDevMonitor
utility (not shown in Doxygen generated docs nor even exported, check PolyDriver.cpp). Such instance is created by allPolyDriver
objects and can be propagated to subdevices (either wrapped devices, or devices opened by other devices in theirDeviceDriver::open
method). Example:Try this via
yarpdev --device ...
with and without--verbose
. Example:Some thoughts:
--help
actually made the device stop and exit (but remember:std::exit
is evil)check
calls and add missing comments/defaultsyarpdev
, but we'll surely want to achieve the same results with RF modules