[Lldb-commits] [lldb] [lldb][SymbolFileDWARF] CompleteType: Lookup type in the declaration DIE's SymbolFile (PR #120569)

[Michael Buch via lldb-commits]

Michael137 wrote:

> > > > In which case, GetDIEToType().lookup(decl_die) will return a nullptr. This is already a bit iffy because some of the surrounding code assumes we don't call CompleteTypeFromDWARF with a nullptr Type*. E.g., CompleteEnumType blindly dereferences it (though enums will never encounter this because their definition is fetched in ParseEnum, unlike for structures).
> > > 
> > > 
> > > Should we bail out early if the Type* is null and return false to tell `SymbolFileDWARFDebugMap::CompleteType` that it can not complete this type and let it iterate to the symbol file that has the entry in its map.
> > 
> > 
> > Yea that might be even simpler. Let me try that.
> > Also, the other crash i alluded to seems to come from the fact that `UniqueDWARFASTTypeList::Find` will compare `Declaration`s (i.e., line and col info) regardless of whether we're dealing with C++ or not. But `GetUniqueTypeNameAndDeclaration` will return a `0` `Declaration` for C++. So we can end up inserting the `Declaration` for a definition DIE into the map (via `UpdateToDefDIE`), but then try to retrieve it using a `0` `Declaration`. Should have a patch up for that soon (I think we can just skip the declaration check if we're dealing with C++?)
> 
> I agree that checking `Declaration` for C++ die isn't necessary, the unique name should be sufficient. I get that Declaration is cleared for all C++ die in `GetUniqueTypeNameAndDeclaration`, so they all have 0 declaration and checking on declaration will just always be true for C++ right? How does it cause a problem?

It's causing issues when we call `ParseTypeFromDWARF` while we're inside `CompleteRecordType`. E.g., in the test-case in this patch.

`CompleteRecordType` will call `UpdateToDefDIE`, which will update the `Declaration` stored in the map to something non-zero (we get it from the `ParsedDWARFTypeAttributes(def_die)`). Then if we call `ParseTypeFromDWARF`, we will search the map for a `0` `Declaration` (because of `GetUniqueTypeNameAndDeclaration`). So we don't find the type and create a new forward declaration. Which causes problems later down the line

https://github.com/llvm/llvm-project/pull/120569