RFC: crucible-llvm: Don't represent architecture at the type level #1046
Conversation
It's necessary to represent the pointer width at the type level in order to statically prevent the formation of ill-typed Crucible/What4 expressions. However, there is no similar motivation for representing architectures at the type level. The previous state of affairs was incoherent in the sense that we always provided an `ArchRepr` signaling the use of x86, without actually checking that this architecture was in use: https://github.com/GaloisInc/crucible/blob/041975a5ad4200633613ed30df144c3a9d0e9403/crucible-llvm/src/Lang/Crucible/LLVM/Translation/Monad.hs#L113 We could have fixed this by checking the target triple (functionality recently added to llvm-pretty{,-bc-parser}) and adding constructors for more architectures to `ArchRepr`, but as discussed above there's no motivation to actually do this.
|
I would like to defer a decision on this until I've had a chance to refresh on the unmerged polyglot work (which needs to be merged, but had type-level impact) to see if this is used at the type level for polyglot. |
That sounds fine to me. Could you open an issue with a description of this upcoming feature and a pointer to the implementation as it stands today, and assign it to yourself? That context would help me understand the impact of potential changes like this while that work is pending, and serve as a place to track progress towards merging it. |
|
@kquick Any updates on the polyglot work? |
|
@kquick Any updates on this? |
|
@kquick Are there still plans to merge the polyglot work? |
It's necessary to represent the pointer width at the type level in order to statically prevent the formation of ill-typed Crucible/What4 expressions. However, there is no similar motivation for representing architectures at the type level.
The previous state of affairs was incoherent in the sense that we always provided an
ArchReprsignaling the use of x86, without actually checking that this architecture was in use:crucible/crucible-llvm/src/Lang/Crucible/LLVM/Translation/Monad.hs
Line 113 in 041975a
We could have fixed this by checking the target triple (functionality recently added to llvm-pretty{,-bc-parser}) and adding constructors for more architectures to
ArchRepr, but as discussed above there's no motivation to actually do this.If we want to move forward with this, we should make a follow-up issue for removing all of the
LLVMArchstuff.