David has pretty much summed it up, but to restate: When we implemented this, we had to be able to distinguish the two cases below: 1. there are no IDX_parent in an accelerator table because the producer did not generate them. 2. there are no IDX_parent in an accelerator table because the parent is not indexed.
Case 1 is represented by the absence of IDX_parent, whereas Case 2 is captured by the "flag_present" case. We also needed case 2 to be space-efficient, which is accomplished through the use of a Form that has no data associated with it (other than its abbreviation entry). There is more context on this in the thread below. https://discourse.llvm.org/t/rfc-improve-dwarf-5-debug-names-type-lookup-parsing-speed/74151/42 Maybe this is something we could submit for standardization? > On Sep 3, 2024, at 15:05, Adrian Prantl via Dwarf-discuss > <dwarf-discuss@lists.dwarfstd.org> wrote: > > CC'ing Felipe Piovezan, who implement this feature, in case he wants to fill > in more details. > > -- adrian > >> On Sep 3, 2024, at 10:06 AM, David Blaikie via Dwarf-discuss >> <dwarf-discuss@lists.dwarfstd.org> wrote: >> >> Some context: >> https://discourse.llvm.org/t/rfc-improve-dwarf-5-debug-names-type-lookup-parsing-speed/74151/44 >> https://github.com/llvm/llvm-project/pull/77457 >> >> Essentially clang/llvm produces DW_IDX_parent with DW_FORM_flag_present when >> the entry has a parent, but that parent isn't indexed (I think this mostly >> comes up for the entities that are direct children of the unit DIE). >> >> This is designed to make it more efficient for consumers (lldb, in >> particular) to do quicker comparisons from the index for qualified names. >> >> My vague recollection is the idea that the consumer needs that info to know >> the difference between "this thing may or may not have a parent" and "this >> thing has an unnamed/unindexed parent". >> >> On Sat, Aug 31, 2024 at 1:03 PM David Anderson via Dwarf-discuss >> <dwarf-discuss@lists.dwarfstd.org <mailto:dwarf-discuss@lists.dwarfstd.org>> >> wrote: >>> Dwarf5. The Name Table .debug_names >>> >>> Background: >>> 6.1.1.2, page 141 >>> Parent debugging information entry, a reference to the >>> index entry for the parent. This is represented as the >>> offset of the entry relative to the start of the entry >>> pool. >>> >>> Table 7.232 Name index attribute encodings >>> DW_IDX_parent constant >>> >>> >>> However clang version 18.1.8 generates >>> >>> DW_IDX_parent DW_FORM_flag_present >>> when there is no parent entry to refer to. >>> >>> And that does not seem to match what DWARF5 >>> says to do. >>> >>> So what is 'true' supposed to mean for DW_IDX_parent? >>> >>> A bit puzzled here. >>> Thanks for any insight. >>> David Anderson >>> >>> -- >>> Lisp Users: Due to the holiday next >>> Monday, there will be no garbage >>> collection. --seen on Slashdot >>> -- >>> Dwarf-discuss mailing list >>> Dwarf-discuss@lists.dwarfstd.org <mailto:Dwarf-discuss@lists.dwarfstd.org> >>> https://lists.dwarfstd.org/mailman/listinfo/dwarf-discuss >> -- >> Dwarf-discuss mailing list >> Dwarf-discuss@lists.dwarfstd.org >> https://lists.dwarfstd.org/mailman/listinfo/dwarf-discuss > > -- > Dwarf-discuss mailing list > Dwarf-discuss@lists.dwarfstd.org > https://lists.dwarfstd.org/mailman/listinfo/dwarf-discuss
-- Dwarf-discuss mailing list Dwarf-discuss@lists.dwarfstd.org https://lists.dwarfstd.org/mailman/listinfo/dwarf-discuss
