Give call/callbr function types instead of pointer types
#207
Conversation
There are a handful of places in `crucible-llvm` where we assume that the `Call` and `CallBr` instructions have pointer types. As noted in GaloisInc/llvm-pretty-bc-parser#189, however, this is subtly incorrect. This patch: * Bumps the `llvm-pretty-bc-parser` submodule to bring in the changes from GaloisInc/llvm-pretty-bc-parser#207, which corrects this mistake, and * Changes the parts of `crucible-llvm` that match on pointer type to no longer do so.
There are a handful of places in `crucible-llvm` where we assume that the `Call` and `CallBr` instructions have pointer types. As noted in GaloisInc/llvm-pretty-bc-parser#189, however, this is subtly incorrect. This patch: * Bumps the `llvm-pretty-bc-parser` submodule to bring in the changes from GaloisInc/llvm-pretty-bc-parser#207, which corrects this mistake. * Bumps the `llvm-pretty` submodule to be in tandem with `llvm-pretty-bc-parser`. * Changes the parts of `crucible-llvm` that match on pointer type to no longer do so.
|
I'm glad you prompted me to look, because it turns out that ppCallSym :: LLVM => Type -> Value -> Doc
ppCallSym (PtrTo (FunTy res args va)) val = ppType res <+> ppArgList va (map ppType args) <+> ppValue val
ppCallSym ty val = ppType ty <+> ppValue valWe shouldn't be using |
|
I have submitted GaloisInc/llvm-pretty#104 with the necessary changes to support pretty-printing Because bumping the |
|
I've added a (Note that I haven't yet added a test case that uses opaque pointers, as I will do that in a separate patch.) |
There are a handful of places in `crucible-llvm` where we assume that the `Call` and `CallBr` instructions have pointer types. As noted in GaloisInc/llvm-pretty-bc-parser#189, however, this is subtly incorrect. This patch: * Bumps the `llvm-pretty-bc-parser` submodule to bring in the changes from GaloisInc/llvm-pretty-bc-parser#207, which corrects this mistake. * Bumps the `llvm-pretty` submodule to be in tandem with `llvm-pretty-bc-parser`. Note that because this brings in the changes from GaloisInc/llvm-pretty#104, I opted to explicitly disallow LLVM's opaque pointers in `crucible-llvm` for now, but this is something that will likely change in subsequent patches. * Changes the parts of `crucible-llvm` that match on pointer types to no longer do so.
See
Note [Typing function applications]for the rationale.This fixes #189. This is also an important step towards being able to support opaque pointers, as described in #177.