[IMPALA-9762] Fix GCC7 compilation issue: shift-count-overflow in tuple-row-compare.cc

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 4.0.0
• Fix Version/s: Impala 4.0.0
• Component/s: Backend
• Epic Link: GCC7_newabi_upgrade
• Target Version: Impala 4.0.0
• Epic Color: ghx-label-4
• Assignee: Joe McDonnell
• Reporter: Joe McDonnell
• Created: 19/May/20 05:34
• Updated: 30/May/20 02:39
• Resolved: 30/May/20 02:39

Description

GCC 7 compilation fails on tuple-row-compare.cc with the following:

 

/home/joe/view2/Impala/be/src/util/tuple-row-compare.cc: In instantiation of \u2018U impala::TupleRowZOrderComparator::GetSharedFloatRepresentation(void*, U) const [with U = unsigned int; T = double]\u2019:
/home/joe/view2/Impala/be/src/util/tuple-row-compare.cc:405:53:   required from \u2018U impala::TupleRowZOrderComparator::GetSharedRepresentation(void*, impala::ColumnType) const [with U = unsigned int]\u2019
/home/joe/view2/Impala/be/src/util/tuple-row-compare.cc:357:41:   required from \u2018int impala::TupleRowZOrderComparator::CompareBasedOnSize(const impala::TupleRow*, const impala::TupleRow*) const [with U = unsigned int]\u2019
/home/joe/view2/Impala/be/src/util/tuple-row-compare.cc:343:49:   required from here
/home/joe/view2/Impala/be/src/util/tuple-row-compare.cc:454:33: error: left shift count >= width of type [-Werror=shift-count-overflow]
     return static_cast<U>(~tmp) << std::max((sizeof(U) - sizeof(T)) * 8, (uint64_t)0);
            ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/joe/view2/Impala/be/src/util/tuple-row-compare.cc:457:33: error: left shift count >= width of type [-Werror=shift-count-overflow]
     return (static_cast<U>(tmp) << std::max((sizeof(U) - sizeof(T)) * 8, (uint64_t)0)) ^
            ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
cc1plus: all warnings being treated as errors

 

What is happening is that GCC is creating some template instantiations of TupleRowZOrderComparator::GetSharedRepresentation() and TupleRowZOrderComparator::CompareBasedOnSize(). The code at issue is doing this calculation:

 

  return (static_cast<U>(val) <<
      std::max((sizeof(U) - sizeof(T)) * 8, (uint64_t)0)) ^ mask;

 

However, for these template instantiations, sizeof(T) > sizeof(U). That condition is impossible and never executed in Impala, but the instantiation exists. The compiler is intepreting the sizeof(U) - sizeof(T) not as a negative integer, but as a very very large positive integer (maybe due to the uint64_t 0). It wins the max calculation and triggers the shift-count-overflow.

The fix is to do the max calculation with signed integers and then cast it to unsigned for the shift.