Update deps

[brianhuffman]

No description provided.

[brianhuffman]

I did add one FIXME to this PR, in SAWScript/Crucible/LLVM/MethodSpecIR.hs, saying that we should eventually do something with any translation warnings that may now be returned by the LLVM-to-Crucible translation. I figured we could open a separate ticket to track that tech debt.

[robdockins]

It looks like the test failures are all due to the new def/use analysis claiming that some LLVM virtual register use is not properly dominated by its definition. That suggests to me that there is some bug causing the analysis to be too conservative.

[brianhuffman]

Well, it looks like I get to spend another afternoon doing builds and tests. I'll bisect and see where this failure was introduced.

[robdockins]

I'm pretty confident it's due to GaloisInc/crucible#886 (in fact, this commit is probably the proximate cause GaloisInc/crucible@c914655)

I'm taking a look at the involved bitcode files now to see if there is an obvious cause.

[robdockins]

The two cases I've found that are causing trouble are both 1) generated from rust code and 2) involve the return value of an invoke instruction. I think the problem is that invoke is an unusual instruction that is both a block terminator and can define a virtual register; probably it is not being handled correctly.

EDIT: indeed, I'll have to fix this case. The use sets of arguments to invoke and the use sets of the blocks it jumps to are treated the same; however, they need to be handled differently, as the normal return successor block should have the result of the invoke in scope.

[robdockins]

I think that GaloisInc/crucible#895 should resolve this problem.

[brianhuffman]

I just reordered some commits and force-pushed a branch without the offending crucible update. I'll test GaloisInc/crucible#895 on top of that separately.

[robdockins]

I don't have any real context for the Macaw changes, but this seems OK to me.

[brianhuffman]

I talked separately with @travitch specifically about adapting to GaloisInc/macaw#226, and he agreed that using unsupportedSyscalls was the right thing to do there.

[brianhuffman]

It looks like a couple of tests have failed due to failure to download yices. I'll restart the tests and hopefully they'll work without changes.