DWARF Standard


HOME
SPECIFICATIONS
FAQ
ISSUES



040618.1 James Cownie Explicit Type for Enumerations Extension Accepted with modifications James Cownie


The problem
-----------

Consider a C++ code like this on a machine with a 32 bit int :-

  #include 
  #include 

  int fn (int )
  {
    printf ("fn (int)\n");
    fflush (stdout);
    return 1;
  }

  int fn (unsigned int )
  {
    printf ("fn (unsigned int)\n");
    fflush (stdout);
    return 2;
  }

  enum limits {
    mostpos = 0x7fffffff,
    mostneg = (1<<31)
  };

  enum limits2 {
    big    = 0x7fffffff,
    bigger = 0x80000000u
  };

  int main (int argc, char ** argv)
  {
    limits l;

    printf ("mostpos = %d, mostneg = %d, sizeof (limits) %d\n", mostpos,
	     mostneg, sizeof(l));

    fn (mostpos);
    fn (bigger);

    return 0;
  }


Two different compilers generate virtually identical DWARF like this
for the enumerations limits and limits2.

 <1><52b>: Abbrev Number: 23 (DW_TAG_enumeration_type)
     DW_AT_name        : limits
     DW_AT_byte_size   : 4
 <2><53a>: Abbrev Number: 24 (DW_TAG_enumerator)
     DW_AT_name        : mostpos
     DW_AT_const_value : 2147483647
 <2><548>: Abbrev Number: 24 (DW_TAG_enumerator)
     DW_AT_name        : mostneg
     DW_AT_const_value : -2147483648

 <1><557>: Abbrev Number: 23 (DW_TAG_enumeration_type)
     DW_AT_name        : limits2
     DW_AT_byte_size   : 4
 <2><567>: Abbrev Number: 24 (DW_TAG_enumerator)
     DW_AT_name        : big
     DW_AT_const_value : 2147483647
 <2><571>: Abbrev Number: 24 (DW_TAG_enumerator)
     DW_AT_name        : bigger
     DW_AT_const_value : -2147483648

Observe that these enumeration types and enumerators are described
identically. There is no information here which allows the debugger to
determine that the underlying types of the enumerations differ.

This means that it is impossible for the debugger correctly
to evaluate the expressions "fn(mostpos)" or "fn(bigger)", since it
can't work out the underlying type of the enumeration types, and
therefore can't resolve the overloaded functions.
[proposal here, not using DW_AT_type, not shown.]


Chris Quenelle  responded.
Jim Blandy suggested DW_AT_type.
Michael Eager responded positively to  this notion.

PROPOSAL:

In section 5.7, Enumeration Type Entries, the following
paragraph be added:

"The DW_TAG_enumeration_type may have a DW_AT_type
attribute referring to the data type which implements
the type. "

Italics:
"If the DW_AT_type attribute refers to a
type with a DW_AT_name a debugger might choose to display
the type name in some way, depending on the source language."

=======================================

Accepted without italic text.


All logos and trademarks in this site are property of their respective owner.
The comments are property of their posters, all the rest © 2007-2017 by DWARF Standards Committee.