COMMON block elimination misbinds variables sometimes

[raehik]

When updating to use fortran-src 0.6.0 with Declarators in CommonGroups, I confirmed that array types are now handled correctly in COMMON block elimination. However, the example I was using had a weird error where the generated binds are sometimes the wrong way round.

Following is a copy-paste from my comment in #136 .

---

This program (from riptutorial.com) now works (after annotating ICOUNT and ISTACK)...

       PROGRAM STACKING
       INTEGER ICOUNT
       INTEGER ISTACK
       COMMON /HEAP/ ICOUNT, ISTACK(1023)
       ICOUNT = 0
       READ *, IVAL
       CALL PUSH(IVAL)
       CALL POP(IVAL)
       END

       SUBROUTINE PUSH(IVAL)
       COMMON /HEAP/ ICOUNT, ISTACK(1023)
       ICOUNT = ICOUNT + 1
       ISTACK(ICOUNT) = IVAL
       RETURN
       END

       SUBROUTINE POP(IVAL)
       COMMON /HEAP/ ICOUNT, ISTACK(1023)
       IVAL = ISTACK(ICOUNT)
       ICOUNT = ICOUNT - 1
       RETURN
       END

...almost. Somehow, the generated code binds the module variables incorrectly 2/3 times:

       PROGRAM STACKING
       use Heap, icount => Heap_istack, istack => Heap_icount
       ICOUNT = 0
       READ *, IVAL
       CALL PUSH(IVAL)
       CALL POP(IVAL)
       END

       SUBROUTINE PUSH(IVAL)
       use Heap, icount => Heap_istack, istack => Heap_icount
       ICOUNT = ICOUNT + 1
       ISTACK(ICOUNT) = IVAL
       RETURN
       END

       SUBROUTINE POP(IVAL)
       use Heap, icount => Heap_icount, istack => Heap_istack
       IVAL = ISTACK(ICOUNT)
       ICOUNT = ICOUNT - 1
       RETURN
       END

Fixing the binds grants a compiling program (with apparently the same behaviour).

*Previously, it would fail with a strange error due to ISTACK(1023) being parsed as a subscript, which then gets converted to a function call.

[raehik]

I've tried adding type annotations to every subroutine instead of just the top one (the original program was implicitly typed, which fortran-src doesn't support). Weirdly, swapping top-level defs doesn't swap their bindings: the bottom def is always correctly bound, and the other two are wrong.

I didn't get a good idea for how the bindings are generated when I looked before. Surely it must be related to that?