[Lldb-commits] [lldb] r106045 - /lldb/trunk/include/lldb/Core/dwarf.h

[Jason Molenda]

On Jun 15, 2010, at 2:43 PM, Chris Lattner wrote:

> Can Core/dwarf.h use (and extend) llvm/include/llvm/Support/Dwarf.h instead of duplicating much of it?

Yeah, I can do that.  I need to add the full set of DW_OP_{lit,reg,breg}{0..31} constants to include/llvm/Support/Dwarf.h and lib/Support/Dwarf.cpp at the very least.  The printers over in Dwarf.cpp are a little annoying - the DW_TAG ones include the "DW_" prefix, the rest do not.  It always grates one me when I see the prefix missing in program output - I have yet to see a context where the extra three characters would have killed anyone.  I'm adding functions like

const char *
DW_ACCESS_value_to_name (uint32_t val)
{
  static char invalid[100];
  const char *llvmstr = AccessibilityString (val);
  if (llvmstr == NULL)
  {
      snprintf (invalid, sizeof (invalid), "Unknown DW_ACCESS constant: 0x%x", val);
      return invalid;
  }

  std::string str;
  if (strlen (llvmstr) < 3 || llvmstr[0] != 'D' || llvmstr[1] != 'W' || llvmstr[2] != '_')
    str = "DW_";
  str += llvmstr;
  ConstString const_str (str.c_str());

  return const_str.GetCString();
}

to lldb's DWARFDefines.cpp to compensate.  We also have a "value to englishy name" utility function that removes all prefix chars (e.g. "DW_FORM_") and changes '_'s to ' ' for printing; they're used in a few places.  To reimplement those on top of the llvm printers I needed to add the modified string to lldb's const string pool and return a pointer to that.  But for the simple value_to_name printers, it's a lot of extra fussing about to ensure that the strings have "DW_" at the front.


In general I prefer the generated approach used for the header/printer functions in lldb -- there is a text file like

/* START: [7.5.4] Figure 16 "Tag Encodings" (pp. 125-127) in DWARFv3 draft 8 */
/* PREFIX: DW_TAG */
/* VALIDATTRIBS: DWARFv3 */

array_type 0x01
class_type 0x02
entry_point 0x03
enumeration_type 0x04
formal_parameter 0x05
imported_declaration 0x08
label 0x0a
lexical_block 0x0b
member 0x0d
pointer_type 0x0f

etc. so that only one table is maintained.  A simple script emits the header file and C source file that prints the texts.  In practice the contents change so rarely that it's not a big deal.


OK to check in the additional constant values to llvm?  I'll need to wait until Greg is back next week so he can update our llvm snapshot we're building against before I can commit the lldb changes.


J