[Bug debug/90197] [8/9 Regression] Cannot step through simple loop at -O -g

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90197

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
             Status|UNCONFIRMED                 |NEW
            Version|unknown                     |8.3.1
           Keywords|                            |wrong-debug
   Last reconfirmed|                            |2019-04-23
                 CC|                            |aoliva at gcc dot gnu.org,
                   |                            |rguenth at gcc dot gnu.org
     Ever confirmed|0                           |1
   Target Milestone|---                         |8.4

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  The issue is that the line number program only contains line number
6
and the consumer (gdb) doesn't handle a jump in a stmt sequence as "ending"
a line, thus 'next' advances to after the loop.  We expand from

  <bb 2> [local count: 118111600]:
  [t.c:5:3] # DEBUG BEGIN_STMT
  [t.c:5:8] # DEBUG BEGIN_STMT
  [t.c:5:12] # DEBUG i => 0
  # DEBUG i => 0
  # DEBUG base => base_7(D)
  [t.c:5:3] if (count_9(D) > 0)
    goto <bb 3>; [89.00%]
  else
    goto <bb 5>; [11.00%]

  <bb 3> [local count: 105119324]:
  ivtmp.10_4 = (unsigned long) dst_10(D);
  _14 = (unsigned int) count_9(D);
  _15 = _14 * 15;
  _21 = base_7(D) + _15;

  <bb 4> [local count: 955630224]:
  # base_17 = PHI <base_7(D)(3), [t.c:5:40] base_13(4)>
  # ivtmp.10_6 = PHI <ivtmp.10_4(3), ivtmp.10_5(4)>
  # DEBUG i => NULL
  # DEBUG base => base_17
  [t.c:6:5] # DEBUG BEGIN_STMT
  _16 = (void *) ivtmp.10_6;
  [t.c:6:12] MEM[base: _16, offset: 0B] = base_17;
  [t.c:5:30] # DEBUG i => D#1
  [t.c:5:40] base_13 = base_17 + 15;
  [t.c:5:40] # DEBUG base => base_13
  # DEBUG i => D#1
  # DEBUG base => base_13
  ivtmp.10_5 = ivtmp.10_6 + 4;
  [t.c:5:3] if (base_13 != _21)
    goto <bb 4>; [89.00%]
  else
    goto <bb 5>; [11.00%]

  <bb 5> [local count: 118111601]:
  [t.c:7:1] return;

notice how the only stmt marker inside the loop body is for t.c:6.

gimplification shows

test (unsigned int * dst, unsigned int base, int count)
[t.c:4:1] {
  [t.c:5:3] # DEBUG BEGIN_STMT
  [t.c:5:3] {
    int i;

    [t.c:5:8] # DEBUG BEGIN_STMT
    [t.c:5:12] i = 0;
    [t.c:5:3] goto <D.1912>;
    <D.1911>:
    [t.c:6:5] # DEBUG BEGIN_STMT
    [t.c:6:8] _1 = (long unsigned int) i;
    [t.c:6:8] _2 = _1 * 4;
    [t.c:6:8] _3 = dst + _2;
    [t.c:6:12] [t.c:6:8] *_3 = base;
    [t.c:5:30] i = i + 1;
    [t.c:5:40] base = base + 15;
    <D.1912>:
    [t.c:5:3] if (i < count) goto <D.1911>; else goto <D.1913>;
    <D.1913>:

so the debug begin_stmt for the loop is attached before the IV
initialization.  Not sure if that has consequences.

The testcase behaves as expected with -gno-statement-frontiers

Alex?