================
@@ -134,7 +134,9 @@ bool TypeQuery::ContextMatches(
if (ctx == ctx_end)
return false; // Pattern too long.
- if (ctx->kind == CompilerContextKind::Namespace && ctx->name.IsEmpty()) {
+ if ((ctx->kind & CompilerContextKind::Namespace) ==
----------------
labath wrote:
Yeah, so the current assumption is that when matching types, the pattern can
contain wildcards (you can say "i'm looking for a type or a namespace at a
given position"), but the thing you're matching is always precise (it will say
"i'm a type" or "I'm a namespace", but it can't say "I don't know"). For DWARF
this isn't an issue because we always have that information. I don't know if
that's the case for PDB, or if it's just a side-effect of how it performs the
query (*). If it's not possible to do it the other way, then we may have to
change the way we do matching, but I think it'd be better to not do that.
(*) Currently, the PDB code kind of cheats by taking the type name and parsing
it as a type query (SymbolFileNativePDB.cpp:1736 in this patch), but then using
the result of that parsing as the *thing to match*. This is bad for two reasons:
- if you're parsing from a name, it's clear that you won't be able to see
whether enclosing scopes are a type or a namespace. The information just isn't
there --`Foo::Bar` looks the same regardless of whether `Foo` is a namespace or
not.
- It's going to be slow because this approach requires you to fully materialize
the type (and everything it depends on) before knowing if this is even the type
you're looking for. This may not be as big of an issue for PDB as it is for
DWARF because (so I hear) in doesn't have template information (which is the
big source of fanout in this process) and it may have less of these searches
internally (because it's already linked, in DWARF we need to do this just to
search for a definition DIE), but it still means that a search for
`foo::iterator` will materialize every iterator out there.
For this reason it would be better type context was created differently,
without relying on a `Type` object. For DWARF this happens in
`DWARFDIE::GetTypeLookupContext`, but I don't know if something like that works
in PDB. The patch description seems to indicate that this information does not
exist, but I don't understand how *anything* could work in that case. E.g. I
don't know what kind of Clang AST would we created for these types. Are they
properly nested in their expected namespace? If so how do you figure out that
namespace?
https://github.com/llvm/llvm-project/pull/149876
_______________________________________________
lldb-commits mailing list
lldb-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
