Closed kungfooman closed 8 months ago
On the other hand, if another kind of type error would happen in the same function, it wouldn't generate a type error either.
Currently I skip error validation code generation for error locations in the ignoreLocations
array - however, I can also imagine to add a runtime option to show them anyway, but with a special "marking" to them... lets say with an orange text or background, because they are "expected derivations from the norm".
As mentioned here:
https://github.com/kungfooman/RuntimeTypeInspector.js/blob/cf0a5b8e1ad534662be6f6ef316a18b6ac64c88e/plugin-rollup/index.mjs#L39-L44
Why?
In order to find the most possible bugs, I basically treat
+-Infinity
andNaN
whenever I find them as a potential error. Yet some functions/methods use them consciously, like in the case of Transformers.js (where the logits processor on the first few runs is processing Float32Array's where all elements are-Infinity
except one, which forces certain special tokens to be generated).In short, if a location is white listed in the
ignorelocations
array, it will not generate a type error.On the other hand, if another kind of type error would happen in the same function, it wouldn't generate a type error either.
So I'm still trying to find the easiest and most precise way to create a precise white list here.