Name mangling mistakes with -g

[brson]

This is a backtrace from compiletest built with -g

#11 0x00007ffff7aee3ff in context::call_shim_on_c_stack (this=0x43d700, args=0x445e80, fn_ptr=0x7ffff7af2f48) at
 ../src/rt/arch/x86_64/context.h:58
#12 0x00007ffff7af4066 in call_upcall_on_c_stack (args=0x445e80, fn_ptr=0x7ffff7af2f48) at ../src/rt/rust_upcall
.cpp:37
#13 0x00007ffff7af3144 in upcall_shared_malloc (nbytes=52, td=0x0) at ../src/rt/rust_upcall.cpp:200
#14 0x000000000042be44 in glue_take::S::_7fffa7de31d4d8d4 ()
#15 0x000000000042ebf6 in _ZN9glue_take32R_5B_compile_lib_path_3D_Srun_lib_path17_78527795f021202cE ()
#16 0x0000000000426d8e in run_tests::_37dd24d3b9435856 ()
#17 0x000000000042fe28 in __morestack ()
#18 0x000000000041c0d0 in main::_c5e23313747479c7 ()
#19 0x000000000041c162 in _rust_main ()
#20 0x00007ffff7aec619 in task_start_wrapper (a=0x44272c) at ../src/rt/rust_task.cpp:282
#21 0x0000000000000000 in ?? ()

gdb doesn't understand one of the names.

[graydon]

Guess is this is because we're sanitizing after mangling. Mangling encodes name-component lengths, and sanitizing goes through and converts various objectionable single-characters to multi-character sequences (eg. @ becomes boxed).

Should sanitize before-or-during mangling.

[graydon]

I have a patchlet that fixes half of this -- the half of doing mangling and sanitizing in the wrong order. But there's another worse half that will be more work: any time we generate a symbol with the same name as another, LLVM silently renames it and appends a numeric suffix. This breaks mangling. We have to ... very carefully go through trans and ensure we never make duplicate names.

I think that's worth splitting into a sub-bug and deferring. The major piece I'll land on this bug shortly.