Out of bounds exception in x86-64-linux compiler when using -dwarf=true -symbols=true

[k-sareen]

$ V3C_OPTS="-dwarf=true -symbols=true" aeneas bootstrap

gives me

Compiling (/home/$USER/git/virgil/bin/bootstrap/x86-linux/Aeneas -> /home/$USER/git/virgil/bin/current/x86-64-linux/Aeneas)...
!BoundsCheckException
        in SsaX86_64Gen.assemble() [aeneas/src/x86-64/SsaX86_64Gen.v3 @ 879:49]
        in SsaMachGen.assembleInstrs() [aeneas/src/mach/MachBackend.v3 @ 1069:33]
        in X86_64Backend.genCodeFromSsa() [aeneas/src/x86-64/X86_64Backend.v3 @ 145:39]
        in MachBackend.genAllCode() [aeneas/src/mach/MachBackend.v3 @ 1343:39]
        in LinuxTarget.emit() [aeneas/src/os/Linux.v3 @ 128:35]
        in Compilation.emit() [aeneas/src/main/Compiler.v3 @ 268:45]
        in Compilation.compile() [aeneas/src/main/Compiler.v3 @ 184:32]
        in Aeneas.compile() [aeneas/src/main/Aeneas.v3 @ 81:56]
        in Aeneas.main() [aeneas/src/main/Aeneas.v3 @ 25:39]

for the x86-64-linux target.

Originally posted by @k-sareen in #308 (comment)

[titzer]

I've fixed one of these bugs, but there are couple more lurking. The handling of debug information in the middle of the compiler isn't well-tested and it looks like SsaDeleteVar needs added type information for normalization.

[k-sareen]

Ah thanks. Is debug information not stable yet? It'd be good to just have them always on. And it'd shake the compiler a bit more so that bugs like these are caught.