Commit_operation bug

adabyron

Hello,

I am developing a plugin in which, at certain moment, I need to switch off the view refreshing, due to performance reasons. So I surround the affected code with Model.start_operation (with the flag disable_ui set to true) and Model.commit_operation. The problem is that I'm getting bug splats when doing the last call (the commit_operation one). Everything is ok until then.

Any idea? The callings are not inside observers. I have some of then in my application, but they don't trigger during the execution of the affected code. I add and delete entities massively (the code is a backtracking algorithm), but as far as I know this is not supposed to be a problem...

Thanks in advance,

Ada

thomthom

I've never had issues with .commit_operation. I suspect that the issue is related to the deletion of entities. You might be trying to interact with invalid entities. Is your code verifying that the entities you deal with is .valid?

(I moved this topic to the Developers section for higher visibility amongst the other ruby developers - until we've gathered more data on what it could be.)

Question: does it splat if you do not use start_operation?
Does it splat if you set the disable_ui flag to false?

TIG

You haven't got the Outliner window rolled-down have you ?
This will splat if you are changing a lot of items that it might be showing...

adabyron

Thank you for your answers

thomthom: No, my code is not verifying the entities. I will fix this, but by the moment I answer your questions:

Q: does it splat if you do not use start_operation? No.
Q: Does it splat if you set the disable_ui flag to false? No.

It only splats when I set the flag to true and I have to delete entities inside the affected code (I have checked this). But now I have noticed a stranger behavior: the view does not refresh until the end, even if I do not put the start_operation and commit_operation callings surrounding the code. I mean, after the backtracking algorithm I can see the result, but I need to put an active_view.refresh calling each time I modify the entities inside the code if I want to see the partial solutions. Does SU refresh the view periodically or just when the entities are changed? I'm a little confused. Anyway, with this behaviour I don't need setting the disable_ui flag to true, so my problem disappears...

TIG: It splats with or without the Outliner window rolled-down.

thomthom

Scripts that does heavy processing can often prevent the SU UI to update - even though SU tries. Using the disable_ui flag simply ensures it does not try to refresh the UI - preserving performance.

Dan Rathbun

Also do not use entities C++ collections directly, make a Ruby Array copy of them, when your loops need to delete items.
You cannot safely iterate and delete items from the C++ collections.
entsA = model.entities.to_a
or
entsA = group.entities.to_a
.

thomthom

Good point.

Dan Rathbun

In addition, some kind of begin .. rescue .. end structure might be used to trap errors, and call .abort_operation in the rescue clause(s), and maybe prevent the BugSplat!.

.. maybe something like this?


begin
  model.start_operation("Task",true)
  entsA = model.entities.to_a
  entsA.each do |e|
    # do stuff 
  end
  model.commit_operation
rescue
  model.abort_operation
end

Anyone have a better example of a basic "operational" block, with error trapping ??

thomthom

That's as simple, and classic, example as it gets.

Jim

I think that in order to catch all classes of exceptions, you need to explicitly give an argument to the rescue clause:


...
rescue Exception => e
puts e.message
puts e.backtrace
end

Or use the appropriate type of Exception, such as LoadError, etc. I don't know how you are supposed to know what type of exception to expect, though.

thomthom

Doesn't it do that be default?

Jim

Back on topic - i would guess, as Dan said, you are probably changing (deleting items from) the collection as you iterate it; which is a no-no.

Jim

I thought, with no argument, that rescue would catch Exceptions of type StandardError, which is probably what is wanted most of the time. But a LoadError and a SyntaxError would not be handled by default.

And I will add that I seem to have got in the habit of using the form I gave as an example, even though it may not be appropriate in all situations.

Dan Rathbun

I believe the rescue clauses' argument list works like the .kind_of? method, in that it is a conditional test that returns true for the exception class named, OR if the named class is a superclass of the actual exception class. (The testing begins at the actual class level, and "walks" the ancestry backward "up the family tree" of superclasses until it the named class is reached. Using Exception as the argument would always execute the rescue clause, as it is the ancestor of ALL exception subclasses.)

Multiple rescue clauses can be defined one after the other, similar to if ... elsif .. elsif ... else ... end statements.

Also more than one Exception class can be listed for each rescue clause.

Finally we can use the global $! to access the exception instead of creating a local via => var, but that's just a matter of personal style.

See: http://www.ruby-doc.org/docs/ProgrammingRuby/html/tut_exceptions.html

adabyron

Dan and Jim: No, I'm not deleting entities as I iterate over the C++ collection. I delete the entities using object attributes, that is, I have something like this:


class Shape

  attr_accessor ;group

  ...

  def erase()
    Sketchup.active_model.entities.erase @group
  end

end

At certain moment, I need to erase all the shapes, so I iterate over MY array of shapes:

shapes.each { |s|
  s.erase
}

thomthom: OK, I didn't know that heavy scripts often prevent the UI to refresh.

By the moment I'm happy with this behavior. Moreover, I have been told to refresh the UI, so the user can see the steps the algorithm is taking (how the entities are added and erased): I don't need to preserv performance any more. Everything is ok (no bug splats at all) when I don't use start_operation and commit_operation.

In case I need to preserv performance again, I will make use of the begin.. rescue.. endidea

Thank you very much

thomthom

But what about the undo stack? If you do not wrap your code in *_operation you get lots of small operations on your undo stack instead of one undoable step. Are you sure your users would be ok with this?

adabyron

thomthom: I have a command that resets the plugin, which users may use. By the moment I'm not worrying about the undo stack, but this is a future task...

But I have just realized one thing:

Sometimes, when I close SketchUp after using my plugin, I get a bug splat. Maybe it is probably due to the same reasons, I mean, when closing SU, the current operation commits, isn't it?

I didn't give it importance at the beginning, because I couldn't reproduce the conditions in which the bug arised... The fact is that, now I realize, everything is OK if I don't delete entities. I have always thought that the bug was some consequence of using EntitiesObservers.

I have some groups.entities in my plugin with this kind of observers attached, but these groups aren't the ones I delete inside the previous affected code. When I need to erase them (outside this code), I check if they have an attached observer and in that case I first remove the observer and then I erase the group.

I have also an EntitiesObserver attached to Sketchup.active_model.entities, which only defines the onElementModified method. This method only does something when the modified entity ID is inside an internal list (there are few entities inside this list).

These observers don't make changes inside the model. Anyway, might them be the source of all my problems?

thomthom

hmm... if you don't modify the model from within the events you usually should be fine. ..but there might be deviating causes. observers has always been a can of worms.

I've tried to make an overview: http://forums.sketchucation.com/viewtopic.php?f=180&t=30793

So - they could be the cause.

Maybe we could get the attention of a Googler to look at this thread...

adabyron

Ok, they ARE the problem.

I have just checked what happens if I don't attach an observer to Sketchup.active_model.entities, and everything is fine with *_operation.

But I still have bug splats when closing SU. I bet this problem has to do with the other EntitiesObservers...

By the moment I need this observers, so I will have to resign myself until the observers run smoothly...

By the way, your overview was so useful to me some weeks ago (I was struggling with another bug). It is a great job

Thank you

thomthom

Which observers and which events did you use?
Can you also briefly describe what you do with the events?
(It'd be nice ot have this info in order to attempts to correlate the issues.)

Btw - I just tried abort_operation to write out texture without adding to the undo stack or messing with the outliner:


tw = Sketchup.create_texture_writer
model = Sketchup.active_model

model.start_operation('Write Textures', true)

tmp = model.definitions.add('Temp_TextureWriter')
g = tmp.entities.add_group
model.materials.to_a.each_with_index { |m,i|
  next if m.texture.nil?
  g.material = m
  tw.load( g )
  p tw.write( g, "c;/temp/mat_#{i}.png" )
}

model.abort_operation

Notice that I make a temp component definition - and therefore do not add anything to the model's entities collection. In worst case you just end up with a temp definition with an empty group. which I think SU will purge automatically at some point.
The code can be improved with error catching - but the concept works.

Think I'll make a wrapper for doing things like this - calling a method with a block where everything inside the block is aborted and therefore not included in the undo stack.