[Dwarf-Discuss] debug_names - what should go in ?

[Pavel Labath]

Hello dwarf-discuss,

over the last couple of months, I have been implementing DWARF v5
debug_names generation support in llvm. Implementation is not yet fully
finished, but I have reached a point where I can start testing the
generated index for completeness.

I wrote a tool (based on LLVM's DWARF verifier) which goes through the
generated debug info and compares it to the contents of the name index.
Right now it takes any debug info entry with a DW_AT_name, applies the
exclusions listed in section 6.1.1.1 of the specification and then reports
if the entry is not present in the index.

I chose to start with default-to-in policy for the verifier, so I can
evaluate whether any reported discrepancies for possible inclusion or
blacklisting *and* because the wording at the start of the said section was
a bit indeterminate to me -- it uses somewhat vague terms like "subprogram"
and "variable" (***), whereas the rest of the section speaks about specific
tags (DW_TAG_variable, etc.).

So far, this approach has found a couple of bugs in the generator, some
things which I obviously needed to blacklist (DW_TAG_compilation_unit), and
a couple of questionable reports. Two of the last kind are
DW_TAG_imported_declaration and DW_TAG_enumerator (not
DW_TAG_enumeration_type).

Both of these define globally visible entities and so if they were not
present, a debugger relying only on the name index would fail to find the
imported enitity or the enumerator even though they should be visible by
the standard scope rules. And yet, both of these are clearly *not* a "named
subprogram, label, variable, type, or namespace". The extra complication
for DW_TAG_imported_declaration is that it can define a named entity even
when it does not have a DW_AT_name attribute (3.2.3: If no name is present,
then the name by which the entity is to be known is the same as the name of
the entity being imported.).

I believe it makes sense to include both of these things in the index
(including the implicitly-named imported declarations), however I would
also like to stick to the specification as much as possible. I know the
spec gives implementors freedom to include extra information in the index,
but I don't want to do that too much, as that would complicate matters for
consumers which need to support multiple producers, some of which may have
a stricter implementation of the spec.

So, I wanted to hear your thoughts on this first. How strictly should we
interpret the first paragraph of section 6.1.1.1 (***)? Do you think it
makes sense to add these entries to the name index?

regards,
pavel labath


(***) ""The name index must contain an entry for each debugging information
entry that defines a named subprogram, label, variable, type, or namespace,
subject to the following rules:""