[Bf-committers] proposal for a small UI improvement for driver setup

[dima glib]

Did anyone consider having an independent math expression evaluation
engine? (e.g. some library from
https://github.com/ArashPartow/math-parser-benchmark-project)
It could be added either as a separate driver type, or auto-detected based
on whether a "pure-math" engine can compile a given expression.
This would likely remove the need to enable scripting for 99% of driver
use-cases. Would it be too complicated to implement?


On Sun, Jun 4, 2017 at 2:58 PM, Joshua Leung <aligorith at gmail.com> wrote:

> Hi,
>
> I agree that the current situation isn't great, and does need to be
> improved.
>
> However, IMO the proposed solution is not good either. Specifically, as
> presented, it does seem that to imply that that checkbox will always get
> shown. There are several issues with this:
> 1) It is highly unlikely that you will want to disable the thing again once
> enabled.
> 2) That setting applies to all scripts/settings, not just the current
> driver. It is therefore misleading to include it as a checkbox there.
> 3) When loading existing files (with the autorun setting disabled), there's
> the duplication that Sergey mentioned (i.e. info header + this panel).
> 4) IIRC, just putting the checkbox there isn't enough to get everything
> running properly for the current session only.
>
> -------
>
> That said, when this autorun setting is disabled (via userpref or
> commandline options), there *is* the annoying problem that newly created
> drivers cannot be run, with no obvious way to get them working. This is a
> usability issue.
>
> As a short-term compromise, I propose that beside/under the error prompt,
> we include the "Reload Trusted" operator button (shown in the info header
> when loading files when autorun is disabled). We'd have to prompt the user
> to save their work (or do it for them), and perhaps the label would need to
> be different (something like "Enable Python autorun").
>
> ------
>
> In the long-term though, IMO we really should look into building our own
> little "mini-Python interpreter" for interpreting driver expressions.
> Campbell's work with sandboxing the Py interpreter (e.g. by whitelisting
> bytecode opcodes) is a promising if fragile approach. However, it doesn't
> mitigate the multithreading problems with Python and the GIL (i.e. there
> can only be a single Py interpreter instance running/evaluating code at
> one). I know that Bassam and a few other riggers have been having problems
> with some of their rigs running slowly (or even crashing) under the new
> depsgraph due to multiple pydrivers getting scheduled to run at once, and
> everything grinding to a halt as they're evaluated one by one. Thus, by
> building our own "driver expression interpreter" that only handles the
> subset of Python-like syntax actually needed to evaluate driver
> expressions, we can solve both security + performance issues at once.
>
> As a fallback, "full" Python evaluation can still occur if our interpreter
> cannot handle a particular expression (i.e. the rigger tried to access a
> custom function, or tried to do something "fancy" with indexing/attribute
> access). That then would still required autorun to be enabled, and would
> still be subject to the GIL restrictions. However, since fewer drivers
> should now be affected by the Py bottleneck/limitation, it's still a net
> positive for users in general over the current situation.
>
> Regards,
> Joshua
>
>
>
>
>
> On Sun, Jun 4, 2017 at 11:06 PM, Gaia Clary <gaia.clary at machinimatrix.org>
> wrote:
>
> > The current solution to this situation is also not a good idea i believe
> :)
> >
> > However, isn't there a difference here between ?
> >
> > - a global definition in user preferences
> > - a session related setting
> >
> > In that sense i believe my proposal is not that bad, as it allows to
> > set the autorun option right on spot (where the drivers are defined)
> > but only for the current session. And it avoids the display of an error
> > and it seemsvery intuitive to me.
> >
> > Also i believe when there is a bad idea at all here, then it is to force
> > the user to enable scripting when all they want is to use drivers.
> > Of course i know Campbell has put some effort into this to see how python
> > could be made more secure so that using it within drivers would no longer
> > be a security problem, and "python in drivers" could be always enabled.
> > But as far as  i know there is no satisfying solution for this,
> > so drivers still need to have auto run enabled.
> >
> > Anyways, it was just a proposal, if its not useful, then its ok for me.
> > I asked at least :)
> >
> > cheers,
> > Gaia
> >
> > On 04.06.2017 10:26, Sergey Sharybin wrote:
> > > Hi,
> > >
> > > This isn't a good idea. You should not expose same user setting all
> over
> > > the interface. All those settings should be kept in a centralized
> place.
> > >
> > > On Sun, Jun 4, 2017 at 1:04 AM, Aaron Carlisle <carlisle.b3d at gmail.com
> >
> > > wrote:
> > >
> > >> I think it is a good idea, I also think that it would be fine to have
> > this
> > >> in 2.79.
> > >>
> > >> Aaron Carlisle
> > >>
> > >> Picture taker | Bit cruncher | Pixel pusher | Document writer | Artist
> > >> Project administrator for the Blender 3D Documentation Project
> > >>
> > >> On Sat, Jun 3, 2017 at 1:04 PM, Gaia Clary <
> > gaia.clary at machinimatrix.org>
> > >> wrote:
> > >>
> > >>> Hi;
> > >>>
> > >>> Assume you have disabled "Autorun Python Scripts" by default.
> > >>> Now add a new driver and step into the graph editor to edit
> > >>> the Driver python expression.
> > >>>
> > >>> Currently you will see an error text in the panel.
> > >>> But what about a change like in the image here:
> > >>>
> > >>>       http://pasteall.org/pic/show.php?id=116080
> > >>>
> > >>> If this is an acceptable improvement, is that something that
> > >>> could possibly already go into Blender 2.79 or would that be only
> > >>> good for Blender 2.8 ?
> > >>>
> > >>> cheers,
> > >>> Gaia
> > >>>
> > >>> _______________________________________________
> > >>> Bf-committers mailing list
> > >>> Bf-committers at blender.org
> > >>> https://lists.blender.org/mailman/listinfo/bf-committers
> > >>>
> > >> _______________________________________________
> > >> Bf-committers mailing list
> > >> Bf-committers at blender.org
> > >> https://lists.blender.org/mailman/listinfo/bf-committers
> > >>
> > >
> > >
> >
> > _______________________________________________
> > Bf-committers mailing list
> > Bf-committers at blender.org
> > https://lists.blender.org/mailman/listinfo/bf-committers
> >
> _______________________________________________
> Bf-committers mailing list
> Bf-committers at blender.org
> https://lists.blender.org/mailman/listinfo/bf-committers
>