(C backend) minting 16 or more short's by prefixing `i` creates potentially illegal C code

[robstewart57]

For foreach loops, Orcc pulls apart the condition expression and mints scalar variable names like i0, i1, j0, j1, and so on. E.g. the following CAL

foreach int(size=16) i in 0 .. 2 do x := x+i; end
foreach int(size=16) i in 0 .. 2 do x := x+i; end

Becomes

u16 x;
i16 i;
i16 i0;
x = 0;
i = 0;
while (i <= 2) {
  x = x + i;
  i = i + 1;
}
i0 = 0;
while (i0 <= 2) {
  x = x + i0;
  i0 = i0 + 1;
}

Where i16 is defined in libs/orcc-runtime/inclues/types.h.

typedef short i16;

There is a corner case when there exists more than 16 foreach statements in an action. Here, the variable name i16 is created, and the declaration for i16 inside the function hides the typedef declaration for i16 as a typedef name. E.g.

actor Foo() ==> :
  my_action: action  ==> 
  var uint(size=16) x := 0
  do
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  foreach int(size=16) i in 0 .. 2 do x := x+i; end
  println(x);
  end
end

Gets compiled to:

static void my_action() {
    u16 x;
    i16 i;
    i16 i0;
    i16 i1;
    i16 i2;
    i16 i3;
    i16 i4;
    i16 i5;
    i16 i6;
    i16 i7;
    i16 i8;
    i16 i9;
    i16 i10;
    i16 i11;
    i16 i12;
    i16 i13;
    i16 i14;
    i16 i15;
    i16 i16;
    i16 i17;
    x = 0;
    i = 0;
    while (i <= 2) {
        x = x + i;
        i = i + 1;
    }
    i0 = 0;
    while (i0 <= 2) {
        x = x + i0;
        i0 = i0 + 1;
    }
    i1 = 0;
    while (i1 <= 2) {
        x = x + i1;
        i1 = i1 + 1;
    }
    i2 = 0;
    while (i2 <= 2) {
        x = x + i2;
        i2 = i2 + 1;
    }
    i3 = 0;
    while (i3 <= 2) {
        x = x + i3;
        i3 = i3 + 1;
    }
    i4 = 0;
    while (i4 <= 2) {
        x = x + i4;
        i4 = i4 + 1;
    }
    i5 = 0;
    while (i5 <= 2) {
        x = x + i5;
        i5 = i5 + 1;
    }
    i6 = 0;
    while (i6 <= 2) {
        x = x + i6;
        i6 = i6 + 1;
    }
    i7 = 0;
    while (i7 <= 2) {
        x = x + i7;
        i7 = i7 + 1;
    }
    i8 = 0;
    while (i8 <= 2) {
        x = x + i8;
        i8 = i8 + 1;
    }
    i9 = 0;
    while (i9 <= 2) {
        x = x + i9;
        i9 = i9 + 1;
    }
    i10 = 0;
    while (i10 <= 2) {
        x = x + i10;
        i10 = i10 + 1;
    }
    i11 = 0;
    while (i11 <= 2) {
        x = x + i11;
        i11 = i11 + 1;
    }
    i12 = 0;
    while (i12 <= 2) {
        x = x + i12;
        i12 = i12 + 1;
    }
    i13 = 0;
    while (i13 <= 2) {
        x = x + i13;
        i13 = i13 + 1;
    }
    i14 = 0;
    while (i14 <= 2) {
        x = x + i14;
        i14 = i14 + 1;
    }
    i15 = 0;
    while (i15 <= 2) {
        x = x + i15;
        i15 = i15 + 1;
    }
    i16 = 0;
    while (i16 <= 2) {
        x = x + i16;
        i16 = i16 + 1;
    }
    i17 = 0;
    while (i17 <= 2) {
        x = x + i17;
        i17 = i17 + 1;
    }
    printf("%u\n", x);
    // Update ports indexes
}

This is illegal C after the i16 i16 line:

Foo.c:60:6: error: expected ‘;’ before ‘i17’
  i16 i17;

[robstewart57]

As pointed about by @endrix , the straight forward CAL solution is to not use i in a CAL foreach statement, i.e. to rewrite

foreach int(size=16) i ..

To

foreach int(size=16) j ..

Nevertheless, strictly speaking this is a corner case bug that the Orcc frontend does not currently guard against.

[alorence]

Hi Rob,

Try to update the line https://github.com/orcc/orcc/blob/master/eclipse/plugins/net.sf.orcc.cal/src/net/sf/orcc/frontend/StmtTransformer.java#L200
from

Var loopVar = procedure.newTempLocalVariable(Typer.getType(variable), variable.getName());

to

Var loopVar = procedure.newTempLocalVariable(Typer.getType(variable), "lp_" + variable.getName());

It should do the trick. Sorry I can't do it by myself neither test if it completely fixes your issue. But the conflict between the loop variable (i + index in the current procedure) and the types defined in the c code is probably generated in the Orcc frontend when the CAL code is parsed to generate the IR. Renaming the variable with a prefix should avoid conflicts when the variable is symply called 'i' (that is often the case for loop vars).