galoisinc / lean-llvm

LLVM support for the lean theorem prover

lean-llvm

This contains a Lean 4 library with bindings to the LLVM library. It is intended to make it easy to work with LLVM in Lean.

See https://github.com/galoisinc/reopt-vcg for most up-to-date info on how this library is currently used/built/etc.

It is made available under the Apache 2 license. See the LICENSE file for more details.

lean-llvm's Issues

Can I try this out

I am very interested in this project. Is it possible to provide me with some instructions to build the code and try the examples out?

`parseBitcodeFile` error reporting is broken

Calling llvm.ffi.parseBitcodeFile with an invalid file (e.g., an LLVM source assembly file instead of a bitcode file) causes an internal lean assertion to fail, e.g.:

../../bin/../src/runtime/lean.h:634: lean_ctor_num_objs: Assertion `lean_is_ctor(o)' failed.

Presumably there's just something off in how we're using set_io_error here.

Also, we may want a slightly more explicit return type---e.g. IO (Except String Module) or similar instead of just IO Module---to match the underlying LLVM parsing interface (it returns an Except essentially as well) and leave raising an IO error for unexpected behaviors. This last note about return types also applies to the llvm.ffi.parseAssembly function recently added (at the moment in a branch), which currently just gets the underlying LLVM error message string and calls set_io_error (here).

Test that change to `src/llvm_exports.cpp` is sound

This is the change I'm not positive is correct =\

Here is a commit removing various usages of cnstr_get_scalar and cnstr_set_scalar right before the commit that removed those two template functions from lean4: leanprover/lean4@16d88b0

Those don't end up using cnstr_set... however, there is other code in the test directories that seems very similar to what we're doing here, e.g. https://github.com/leanprover/lean4/blob/ec9f4b579dc6b4a20407d2735f8ad91de29bfdb5/src/tests/util/object.cpp#L266

So yah, I'm not sure.

Originally posted by @pnwamk in #16

parseAssembly should take a data layout param

Currently parseAssembly uses a hard coded data layout string for expediency. We should probably just take a data_layout structure and use it's toString instance method or similar (which doesn't yet exist I believe).