Probably very much depends on what you do with Blender. I've been trying to port my exporter plugin from Maya to Blender, and it has been nothing but frustrating. It's been quite a while since I wrote it for Maya, but I remember the Maya API (C++ and MEL) to be well documented, well structured, and quite pleasant to work with. I was always amazed at Maya's basic architecture: Every UI action maps to a command that is run by the interpreter. This makes scripting very easy.
Blender's Python API on the other hand is very poorly documented, and - to me - seems poorly thought out. It's not really designed as an API, it's just exposing the existing C++ code, and usability is not a concern. It feels tacked on, leaving lots of gaps. Somewhere I've seen a quote by Ton Rosendaal that Blender is for artists, not for coders. If you want Blender to be a competitor for Maya that seems very short sighted.
The one thing that impressed me though, was the speed by which the Blender devs react to tickets. Hats off to them!
Another problem is that the code changes from version to version, so a big chance that example you found no longer works. While annoying it’s not insurmountable, but a good doc would help a lot indeed! Perhaps with the LTS release that’s a thing to solve.
Unless the example is from 2.79 and earlier it shouldn't take any or only minimal adjustments. There have been some API changes between versions since 2.81, but they usually wouldn't affect a lot of add-ons or are very easy to change.
All the API changes are also documented in the release notes [0] and the API docs allow to view the Python API for each version [1].
What change between versions caused you problems and what parts of the API docs were lacking?
What is a problem, though, is that when you google an issue, and find a solution, you then have to upgrade that solution to your version of Blender. That is very annoying because unless you're familiar with Blender, you can't tell if the solution is wrong, or correct but outdated.
From my experience script examples usually indicate what version they were made for and add-ons have to declare the minimum supported Blender version in their `bl_info`, which tells what API version it was implemented for. I know that at least on Blender's Stack Exchange the community tries to provide updated answers if anything relevant has changed in a new version.
I can see that older scripts for 2.79 and earlier may cause the issue you're describing, but this doesn't seem like a Blender specific issue. How did Maya handle API changes between versions? I would assume that there were breaking changes at some point that caused similar issues.
I have no experience with blender but have been developing for maya for 7+ years or so. Python in Maya is very much a afterthought and extremely slow (even for Python standards). There is noting pythonic about the api which by all means is just a port of the mel interface.
The documentation is great, agreed, but I am not sure you can attribute that to the python part of Maya. The documentation structure goes back to pre python days.
If any of the major studios spent a fraction of their development budget on Blender rather than Maya (and contributed back to the community) I think we could see great strides in polishing Blenders usability for scripting.
> That is pretty much how it works in Blender. All the buttons and properties in the UI are accessible through the Python API.
Yes, it seems that way, and I was hopeful at first. But then I tried to use it.
In Maya, you can just copy paste a command from the command window, and it will work.
When you do that in Blender, it will probably give you an unspecific error message like "not possible in the current context". Because you need to manually switch Blender e. g. from Object mode to Edit mode (Faces), and your command only works in that context. So you develop a script that works initially, but later you find that it only works if Blender is in a certain state. That state is nowhere documented, you have to guess what it wants.
While Maya uses arguments to commands, Blender uses the equivalent of global variables.
That is specific to operators (Blender's implementation of buttons) since they are closely tied to their use in the UI. They may expect a specific context which is checked by the operator's `poll()` function. The namespace the operator is in should hint at what that context is, e.g. `bpy.ops.mesh` works with mesh data which is accessed in edit mode. Other operator namespaces directly reference area types, those need to be executed within that area. For instance operators `bpy.ops.view3d` will most likely require to be executed in an area of type `VIEW3D`. That is a requirement because the data that the operator needs is otherwise not accessible. I understand that the approach you are used to in Maya is more convenient.
When you have issues with calling an operator due to a wrong context, there are different ways to approach the problem:
1. Figure out what the necessary context is and switch modes, areas, etc. to match the requirements.
2. Figure out what the necessary context is and override the context.
3. Avoid using operators.
Approach 1.) could be accomplished by checking where the operator is originally used in Blender. For instance the `bpy.ops.mesh.remove_doubles()` is available in edit mode in the Mesh > Clean up menu. Thus edit mode is required to run the operator. This will work for many operators because they don't have very complicated requirements.
Approach 2.) is for when you must use an operator, it requires a complicated context and you know about Blender's internals. You can look at the operator's (poll) implementation and construct a specific context that is passed into the operator [0]. This should only be used if there is no better alternative.
Approach 3.) is preferable when you are working with mesh data. Instead of using regular operators you can use bmesh instead [1]. It doesn't require switching to edit mode or passing a context. It's usually the more efficient approach as well.
The comment above is more of an explanation of how this currently works in Blender and how one can approach the aforementioned challenge.
I do agree that the Python API is often just a thin wrapper around the C/C++ code [0] and that the documentation of the required context for operators is a bit lacking for people who are not familiar with Blender's implementation. This is also acknowledged in the API docs [1]. However, I do think the API is quite usable and not particularly complicated to understand.
I have the same observation as a fairly novice maya programmer. it's absolutely priceless to be able to do something using the "regular" UI and just see every single command that is run and then just copy paste the appropriate bits and stitch them together with various conditionals and tests.
maya is even MORE layered than just ui -> script commands, there's the intermediate layer of the node editor which shows the scene data as a set of interconnected nodes and what information they're passing between each other (with a few odd omissions) basically any function in maya can be accessed at least 3 ways at 3 different levels of technical skill. where in blender there are still fairly large missing chunks to translate between the UI level, node level and script level.
all that being said it's a very impressive 3d suite and I highly welcome any and all competition!
that would be pretty telling, I expected python to be big advantage for blender but if the internals/model are fluffy, and fluffier than good old maya i'd be worried
Blender's Python API on the other hand is very poorly documented, and - to me - seems poorly thought out. It's not really designed as an API, it's just exposing the existing C++ code, and usability is not a concern. It feels tacked on, leaving lots of gaps. Somewhere I've seen a quote by Ton Rosendaal that Blender is for artists, not for coders. If you want Blender to be a competitor for Maya that seems very short sighted.
The one thing that impressed me though, was the speed by which the Blender devs react to tickets. Hats off to them!