[DebugInfo] Removed assert on missing CountVarDIE

[sdesmalen]

Summary

The assert for a DISubrange's CountVarDIE to be available fails
when the dbg.value() has been optimized away for any reason.
Having the assert for that is a little heavy, so instead removing
it now in favor of not generating the 'count' expression.

Addresses http://llvm.org/PR36263 .

[aprantl]

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

1391

Why not:

if (auto *CV = SR->getCount().dyn_cast<DIVariable*>()) {
  if (DIE *CountVarDIE = getDIE(CV))
    addDIEEntry(DW_Subrange, dwarf::DW_AT_count, *CountVarDIE);
  else
     // should this. case be handled?
} else if (Count != -1)
  addUInt(DW_Subrange, dwarf::DW_AT_count, None, Count);

[aprantl]

Can you add a testcase by, e.g, manually removing the dbg.value for the count from an existing testcase?

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

1392

That else statement was more meant as a question for you :-)

[sdesmalen]

• Renamed test/DebugInfo/Generic/disubrange.ll -> test/DebugInfo/Generic/disubrange_vla.ll
• Added test/DebugInfo/Generic/disubrange_vla_no_dbgvalue.ll

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

1392

Ahh, I wasn't sure :-) So if there is not enough information about the 'count' at the point of arriving here, there isn't much else you can do than not generating the information. Of course it is best to make sure this doesn't happen without a good reason (hence https://reviews.llvm.org/D43386).

[sdesmalen]

I expect this code to undergo some more changes in the near future because it doesn't work properly with inlining (http://llvm.org/PR36323) and so needs more effort to better retrieve the DIEs for inlined count expressions and to generate correct DWARF for the inlined variable-length types. I started looking into that, but haven't had the time to come up with a patch yet.

[davide]

@sdesmalen is this the same issue as https://bugs.llvm.org/show_bug.cgi?id=36376 ?

[sdesmalen]

In D43387#1011468, @davide wrote:

@sdesmalen is this the same issue as https://bugs.llvm.org/show_bug.cgi?id=36376 ?

Yes, that seems like the same issue, thanks for pointing out!

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

void DwarfUnit::constructSubrangeDIE(DIE &Buffer, const DISubrange *SR,
   int64_t DefaultLowerBound = getDefaultLowerBound();
   int64_t Count = -1;
   if (auto *CI = SR->getCount().dyn_cast<ConstantInt*>())
     Count = CI->getSExtValue();
 
   if (DefaultLowerBound == -1 || LowerBound != DefaultLowerBound)
     addUInt(DW_Subrange, dwarf::DW_AT_lower_bound, None, LowerBound);
 
-  if (auto *CV = SR->getCount().dyn_cast<DIVariable*>()) {
-    // 'finishVariableDefinition' that creates the types for a variable is
-    // always called _after_ the DIEs for variables are created.
-    auto *CountVarDIE = getDIE(CV);
-    assert(CountVarDIE && "DIE for count is not yet instantiated");
+  DIE *CountVarDIE = nullptr;
+  if (auto *CV = SR->getCount().dyn_cast<DIVariable*>())
+    CountVarDIE = getDIE(CV);
+
+  if (CountVarDIE)
     addDIEEntry(DW_Subrange, dwarf::DW_AT_count, *CountVarDIE);
-  } else if (Count != -1)
-    // FIXME: An unbounded array should reference the expression that defines
-    // the array.
+  else if (Count != -1)
     addUInt(DW_Subrange, dwarf::DW_AT_count, None, Count);
 }
 
 DIE *DwarfUnit::getIndexTyDie() {
   if (IndexTyDie)
     return IndexTyDie;
   // Construct an integer type to use for indexes.
   IndexTyDie = &createAndAddDIE(dwarf::DW_TAG_base_type, getUnitDie());
   addString(*IndexTyDie, dwarf::DW_AT_name, "sizetype");
   addUInt(*IndexTyDie, dwarf::DW_AT_byte_size, None, sizeof(int64_t));