Review pac

review of the first pipeline as code i wrote (now named pac2 oddly enough)
pipeline as code

Summary¶

• had a working prototype in days
• has been finished after a few more days i think.
• then slowly started to evolve out of hand.
• eventually died since never finished “perfectly”

Breakdown¶

The original goal was: create Pyblish but better, since PRs to fix it didn’t work out. I was bothered by all the Pyblish issues so decided to restart from a clean slate.

First I researched all the things that bothered me with Pyblish, and tried small prototypes to address them. After a few days I had a nice prototype for pac. It went fast, mainly because I was so immersed (obsessed) in Pyblish and validation at the time.
It hit all the goals Pyblish aimed to address, but cleaner.

I since have had to make 2 or 3 validation systems at other jobs, and fell back to using Pyblish since it works out of the box. But it sucked and reflected in a subpar validation system being rolled out.
Meanwhile I had this working prototype lying around, but since it wasn’t finished it didn’t feel ready use at work. I don’t want to bring unfinished projects in work and end up being biased and coding on my little fun project during work.

Things I learned.¶

• i was trying to make a graph, so researched a bunch of existing solutions and wrote notes.
• nodes don’t need self.id since python objects all have an id id(my_object)
• nodes don’t need a name, since all python objects have a name. self.__class__.__name__
• nodes don’t need actions, since all python objects have methods, which can be private or public.
• i was a bit fuzzy on the exact way, but nodes have connections. since they are always called by another node, which can be (recursively) accessed through the caller_object, using inspect.stack(). (Code felt little hacky but did the job)
• i could do complex class manipulation, e.g. overriding def __setattr__ so we always stored a link to the node assigned to an attribute. This was really cool but mainly ended up abstracting things. Nice when stuff worked, but a lil harder to debug.
• serialization, config and raw config are surprisingly confusing. Might have made more sense to split that up. see composition vs inheritance.
• nodes don’t need a run method, since python objects have __call__

my progamming review