Allow intrinsics to have multiple input arguments #9527
Conversation
[ci skip-jvm-tests]
…port for intrinsics taking more than one argument. # Delete this line to force CI to run the JVM tests. [ci skip-jvm-tests]
# Delete this line to force CI to run the JVM tests. [ci skip-jvm-tests]
stuhood
force-pushed
the
stuhood/multiple-inputs-to-intrinsics
branch
from
e919335
to
3ff8f2f
Compare
|
The commits are useful to review independently. |
| intrinsics.insert( | ||
| Intrinsic { | ||
| product: types.directory_digest, | ||
| inputs: vec![types.input_files_content], |
There was a problem hiding this comment.
It would be nice to be able to tie the inputs length to the length on the values vec expected by the intrinsic function. The &args[0], &args[1] look scary in isolation (what if there are 3 input types and the function evolves to only using 2 or needing 4. There is and will be a small enough set of intrinsics that just making sure these align by hand is probably good enough.
There was a problem hiding this comment.
There is and will be a small enough set of intrinsics that just making sure these align by hand is probably good enough.
Yea. For better or worse, these have evolved away from a large switch statement and toward dynamic dispatch like this, because it makes the callsite clearer. It's possible that moving back toward the switch statement is feasible, but for now I think that the fact these are now isolated to one file makes this lack of type safety manageable.
Problem
Intrinsics have been functions from oneValueto anotherValuefor a while now, but this does not allow an intrinsic function to depend on multipleParams simultaneously.Solution
Move the definitions of
Intrinsics next to their implementations, and change their signatures toVec<Value> -> Valueto allow for further expansion of inputs.Finally, to prepare for multi-platform speculation make
MultiPlatformExecuteProcessthe first intrinsic with multiple arguments by adding aPlatformrequirement.