Miss destruction in switch statements

[wenpen]

The St destruction of following code won't be called.

#include <cstdio>

struct St{
  St() {
    puts("Ctor");
  }
  ~St(){
    puts("Dtor");
  }
};

int f(int a) {
  switch(a) {
    case 1:{
      St a;
      return 1;
      int x=1;
    }
  }
  return 0;
}

int main() {
  f(1);
}

emit-cir-flag output:

  cir.func @_Z1fi(%arg0: !s32i loc(fused[#loc13, #loc14])) -> !s32i extra(#fn_attr) {
    %0 = cir.alloca !s32i, !cir.ptr<!s32i>, ["a", init] {alignment = 4 : i64} loc(#loc38)
    %1 = cir.alloca !s32i, !cir.ptr<!s32i>, ["__retval"] {alignment = 4 : i64} loc(#loc12)
    cir.store %arg0, %0 : !s32i, !cir.ptr<!s32i> loc(#loc15)
    cir.br ^bb1 loc(#loc39)
  ^bb1:  // pred: ^bb0
    %2 = cir.load %0 : !cir.ptr<!s32i>, !s32i loc(#loc18)
    cir.switch.flat %2 : !s32i, ^bb6 [
      1: ^bb2
    ] loc(#loc16)
  ^bb2:  // pred: ^bb1
    cir.br ^bb3 loc(#loc40)
  ^bb3:  // pred: ^bb2
    %3 = cir.alloca !ty_22St22, !cir.ptr<!ty_22St22>, ["a", init] {alignment = 1 : i64} loc(#loc41)
    %4 = cir.alloca !s32i, !cir.ptr<!s32i>, ["x", init] {alignment = 4 : i64} loc(#loc42)
    cir.call @_ZN2StC2Ev(%3) : (!cir.ptr<!ty_22St22>) -> () loc(#loc22)
    %5 = cir.const #cir.int<1> : !s32i loc(#loc25)
    cir.store %5, %1 : !s32i, !cir.ptr<!s32i> loc(#loc43)
    %6 = cir.load %1 : !cir.ptr<!s32i>, !s32i loc(#loc43)
    cir.return %6 : !s32i loc(#loc43)
  ^bb4:  // no predecessors
    %7 = cir.const #cir.int<1> : !s32i loc(#loc24)
    cir.store %7, %4 : !s32i, !cir.ptr<!s32i> loc(#loc42)
    cir.call @_ZN2StD2Ev(%3) : (!cir.ptr<!ty_22St22>) -> () loc(#loc36)
    cir.br ^bb5 loc(#loc20)
  ^bb5:  // pred: ^bb4
    cir.br ^bb6 loc(#loc16)
  ^bb6:  // 2 preds: ^bb1, ^bb5
    cir.br ^bb7 loc(#loc17)
  ^bb7:  // pred: ^bb6
    %8 = cir.const #cir.int<0> : !s32i loc(#loc27)
    cir.store %8, %1 : !s32i, !cir.ptr<!s32i> loc(#loc44)
    %9 = cir.load %1 : !cir.ptr<!s32i>, !s32i loc(#loc44)
    cir.return %9 : !s32i loc(#loc44)
  } loc(#loc37)

[Lancern]

This is a variant of #348 .

[bcardosolopes]

@Lancern is right. I have some on going work there I need to be resumed, should take care of this too.

[bcardosolopes]

@wenpen if you are looking for issues to work on, I advice against picking this one right now, it involves many parts of CIRGen is quite complex, I'd go for more simple stuff until you get more familiarity with the different parts.

[wenpen]

Just found this issue when construct test case, thank you for advice, I will keep focusing on my unfinished work :)