[PATCH] D7895: Anonymous namespaces are missing import DW_TAG_imported_module.

Tue Aug 18 11:34:42 PDT 2015

| I don't understand why "without requiring* additional smarts on the debugger side" is a goal.
It is explicitly a goal of DWARF to be language-neutral, and it is contrary to that goal to require a debugger to know a C++ rule about anonymous namespaces in order to conjure up a correct list of named entities.  I'd go so far as to say it's a bug in the producer.

| We require lots of additional smarts on the debugger side to deal with ABI knowledge
AFAICT all of those smarts are to deal with debugger features intended to imitate things a *compiler* would do, like support dynamically calling a function in the process-under-debug.  Nobody has shown why a debugger needs ABI knowledge in order to interpret the DWARF correctly, in its intended mode of providing a language-neutral mapping of source to object.
Certainly if your fancy-pants debugger wants to pretend to be a compiler, JITting ABI-compatible code and such, it needs to know a bunch more stuff (like the ABI), but that's not a justification for emitting DWARF that *can't* be interpreted correctly without specific C++ rule knowledge.

| (SFINAE being an obvious example of happy fun times),
SFINAE (AFAIK) applies to template selection, and since you don't have a template declaration in the DWARF, that's irrelevant here.
--paulr

From: Eric Christopher [mailto:echristo at gmail.com]
Sent: Tuesday, August 18, 2015 11:06 AM
To: David Blaikie
Cc: Robinson, Paul; reviews+D7895+public+7827be49c0b04087 at reviews.llvm.org; Romanova, Katya; Frédéric Riss; Duncan P. N. Exon Smith; llvm-commits
Subject: Re: [PATCH] D7895: Anonymous namespaces are missing import DW_TAG_imported_module.

On Tue, Aug 18, 2015 at 11:03 AM David Blaikie <dblaikie at gmail.com<mailto:dblaikie at gmail.com>> wrote:
On Tue, Aug 18, 2015 at 10:59 AM, Eric Christopher <echristo at gmail.com<mailto:echristo at gmail.com>> wrote:

On Tue, Aug 18, 2015 at 10:55 AM David Blaikie <dblaikie at gmail.com<mailto:dblaikie at gmail.com>> wrote:
On Tue, Aug 18, 2015 at 10:43 AM, Robinson, Paul <Paul_Robinson at playstation.sony.com<mailto:Paul_Robinson at playstation.sony.com>> wrote:
The as-if suggests that an explicit artificial import is an entirely appropriate, natural and good way to handle it, because it models the language behavior exactly without *requiring* additional smarts on the debugger side.

I agree it is a way to handle it - but I don't understand why "without requiring* additional smarts on the debugger side" is a goal.

That said, it seems to be a bit of a double standard here. We require lots of additional smarts on the debugger side to deal with ABI knowledge, and in fact have numerous odd workarounds in DWARF to deal with parameter passing that would be less of an issue if we had generalized ABI support (as an example). How is this case any different?

The conclusion with the ABI stuff was that it would be impossible for a client to model it correctly - there were cases Richard demonstrated that there was no reasonable extra DWARF we could produce in the type that would allow the debugger to make the right choice about the calling convention to use when that type was passed (I believe there were particular template cases?). Also, the argument was made that these rules change over time - and the debugger wouldn't know which set of rules to use without being told which ones the compiler had used when it compiled this file (OK, so if we had enough type information, we could've just put a "ABI rev number" on the CU).

For both of those reasons, it is necessary that we actually just put a tag on the type specifying how it should be passed.

Or specify ABI per function of course.

These cases don't apply here - C++ has fun/weird/interesting name lookup rules (and, admittedly, I'd wager we still don't provide enough DWARF to make the name lookup work every time (SFINAE being an obvious example of happy fun times), but generally we rely on consumers to do best effort rather than by trying to describe it all explicitly in DWARF - the latter seems pretty much impractical)

Oh I agree. :)

-eric

- Dave

-eric

natural/good - I just don't agree with. The stated goal in many cases has been that the DWARF model the source as written. This is a case where that's easy and obvious to do.

* it's not requiring - it's just a degraded experience (as you say, quality of implementation - as with many places where, if the debugger doesn't understand C++, it gets a less-than-ideal user experience)

Not all as-if statements have to be modeled directly in the DWARF, because they aren't relevant to DWARF or DWARF has other ways of achieving the same effect.  For example, [namespace.udir]/2 says a using-directive means "the names appear as if they were declared in the nearest enclosing namespace…" but there's no reason to generate piles of declarations in the outer namespace because, we have DW_TAG_imported_module which means exactly that.
I don't see any good reason for artificial namespaces to be treated specially just because the namespace is anonymous and the 'using' directive isn't explicit in the source.

From my perspective it seems like you're asking the producer to special case anonymous namespaces so the debugger doesn't have to. I'd simply argue the opposite - I'd like to not special case them in the producer, instead producing source-accurate debug info as much as possible, and leave it to the debugger to implement appropriate behavior.

The Cobol example seems perfect: We rely on consumers modeling languages correctly to import outer namespaces into inner ones in languages that support that. In the languages that don't, we rely on the consumer to know that and model it differently. If a consumer does not do this, the user gets a degraded experience (their cobol name lookup is wrong (or perhaps their non-cobol name lookup is wrong, if the debugger decided to only model the cobol side and assume all other languages work that way too)).

This seems directly analogous to anonymous namespaces - we describe the source and let the consumer model it appropriately.

--paulr

From: David Blaikie [mailto:dblaikie at gmail.com<mailto:dblaikie at gmail.com>]
Sent: Tuesday, August 18, 2015 10:27 AM
To: Robinson, Paul
Cc: reviews+D7895+public+7827be49c0b04087 at reviews.llvm.org<mailto:reviews%2BD7895%2Bpublic%2B7827be49c0b04087 at reviews.llvm.org>; Romanova, Katya; Eric Christopher; Frédéric Riss; Duncan P. N. Exon Smith; llvm-commits
Subject: Re: [PATCH] D7895: Anonymous namespaces are missing import DW_TAG_imported_module.

On Tue, Aug 18, 2015 at 10:23 AM, Robinson, Paul <Paul_Robinson at playstation.sony.com<mailto:Paul_Robinson at playstation.sony.com>> wrote:
| suggests that DW_TAG_imported_module could be used for C++ using directives (of which there are none in the source code in question).

[namespace.unnamed]/1 says there is an implicit (as-if) using directive.

Lots of things are as-if, doesn't mean we model them in the DWARF that way, it's just an easy way of defining the behavior in terms of other stuff that's already defined in the standard. Generally we try to model things like the actual source where possible, it seems/usually sounds like/etc.

--paulr

From: David Blaikie [mailto:dblaikie at gmail.com<mailto:dblaikie at gmail.com>]
Sent: Tuesday, August 18, 2015 10:15 AM

To: Robinson, Paul
Cc: reviews+D7895+public+7827be49c0b04087 at reviews.llvm.org<mailto:reviews%2BD7895%2Bpublic%2B7827be49c0b04087 at reviews.llvm.org>; Romanova, Katya; Eric Christopher; Frédéric Riss; Duncan P. N. Exon Smith; llvm-commits
Subject: Re: [PATCH] D7895: Anonymous namespaces are missing import DW_TAG_imported_module.

All I'm saying is that I think this is a fine thing to have as a debugger-specific tuning for your debugger.

DWARF-the-standard is language neutral, yes, so it provides features that can be used to describe many different languages. It describes (non-normatively) possible uses of DWARF features for language features, and suggests that DW_TAG_imported_module could be used for C++ using directives (of which there are none in the source code in question). It doesn't suggest anywhere that it should, let alone must, be used to explicitly describe name lookup rules (or we'd need those using directives in nested scopes for all outer scopes, for example).

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150818/b9b87fa1/attachment.html>