First, I have found that the triggers are not added to the codegen executors (e.g. Insert). I think This happened because the codegen work was done based on the old executors without triggers.
Adding the feature was thought to be simple, as only a function call should be added at each trigger point. However, I found that the trigger interface is not very kind to the executor. The executor had to 1) check trigger list's nullability, 2) check whether the trigger with a certain type exists, and then 3) call to execute the trigger. I think the trigger facility could abstract all these and have one function call as an interface to outside.
schedutron
commented
6 years ago
First, I have found that the triggers are not added to the codegen executors (e.g. Insert). I think This happened because the codegen work was done based on the old executors without triggers.
Adding the feature was thought to be simple, as only a function call should be added at each trigger point. However, I found that the trigger interface is not very kind to the executor. The executor had to 1) check trigger list's nullability, 2) check whether the trigger with a certain type exists, and then 3) call to execute the trigger. I think the trigger facility could abstract all these and have one function call as an interface to outside.