Friday, 9 October 2015

Python hou.runVex()

In a previous post I managed to find and implement the math required to find intersection point between 2 lines in 3D, which was interesting and a challenge.But in that and other situation I often wonder how I can exploit Houdini more when it comes to this kind of math, as Houdini's nodes already implement so much. Quite commonly cases arise where I'd like to run some python code against a node, ie. Invoke a node, pass some geometry and return the output.

That's where I got to thinking that VEX actually has a lot of powerful functions that Houdini's python doesn't, Then I remembered coming across the hou.runVex() method. After spending the best part of a week getting to grips with this workflow, thought I'd write down some of my findings, since there's not much out there on this topic.

Things of note when using hou.runVex(vexfile, inputs)
  1. The inputs argument can't be an empty dictionary, even if there aren't arguments defined on the cvex function.
  2. I've found that cvex must be the context function, at least when working with the Python SOP. Although don't quote me on that needs more testing.
  3. When compiling, it's safest to rename the .vfl when changes are made, at least for a successful compilation to vex. e.g.. Say you have a file named attrib.vfl and you compile it to attrib.vex successfully, upon making code changes to the .vfl it's safer to rename the file attrib_01.vfl and compile this to a corresponding .vex file. This will ensure that the vex file reflects the changes.
  4. There's conditions to passing lists or arrays in the inputs dictionary, they must have the same length and each lists items must be of the same type.
  5. Also related to input arguments having length > 1 , e.g. Lists or arrays. Theses types will affect the exports/bound variables length, causing outputs to also display matching length even when their default value is of a single value.
  6. Another point worth noting regarding inputs when using lists is that they effectively form a kind of loop, running the vex code as many times as there are items in the list. This ability could be utilised to emulate functionality akin to the vex/vop nodes parameter run over : points, prims, vertices, detail.
  7. It also seems, a least for vector type inputs, they must be contained in a list or array type, ie When passing a single vector, it will need to be in an list of 1 item, or the function's input argument default defined in vex will be used. This doesn't seem to be the case with strings or integers, which makes me wonder whether it's due to vector types having native length in it's components.
  8. This point needs more testing, getting some very strange results which are difficulto track down.I think there is a memory issue, although it could be due to way vex works! Not entirely sure. Can't recall exactly, but in the docs it talks about how vex is compiled and that it doesn't allow for recursive code. One odd effect of running vex via Python in this way, is that you can refer to the returned value before you've made the call. eg
      print a
      a = hou.runVex()
      This will work fine!
      Then again it doesn't, as it shouldn't.
    I think there maybe a memory related issue, because although the vex is running as expected, it seems to store the result in memory and that's why the above call to variable "a" before it's definition works, occasionally. Now when I change position values via an upstream transform node, the results of the vex also transform when they definitely shouldn't be. The kicker is when I save the hip with the upstream xform and reopen the hip the results are as expected. Also the variable ref that was working now doesn't, errors as expected. which brings me back to the vex in memory being the problem.
In Short, It works! Mostly.
As long as the above points are considered, hou.runVex() should work as expected. This workflow has the potential to be very helpful and powerful, which I'd like to use more often. Though the last point is a bit of a concern.