[llvm-dev] [PATCH] D14358: DWARF's forward decl of a template should have template parameters.

David Blaikie via llvm-dev llvm-dev at lists.llvm.org
Wed Dec 9 12:31:30 PST 2015


On Wed, Dec 9, 2015 at 11:59 AM, Robinson, Paul <
Paul_Robinson at playstation.sony.com> wrote:

> Maybe we are being too pedantic about the names.  I'll have to go back and
> look in detail at why we decided to do that.
>
>
>
> In any case, arguably 5.5.8 (Class Template Instantiations) 1 only applies
> to definitions of a type, not declarations. ("Each formal parameterized
> type declaration appearing in the template definition is represented by a
> debugging information entry with the tag DW_TAG_template_type_parameter")
>
> Not so fast… It's a template definition of a type declaration.  DWARF 5 is
> less ambiguous about this IMO, although you are actually very good at
> finding the ambiguities!  The relevant text in DWARF 5 current draft is: "A
> debugging information entry that represents a template instantiation will
> contain child entries describing the actual template parameters."  Are you
> willing to argue that this type declaration is not an instantiation?  (If
> not, what is it?)
>

Nope, it is an instantiation, for sure.

Beyond that, I'd still argue that existing implementation experience points
to this being pretty reasonable - and without the motivating use case
(which only exists in your fork) it seems hard to justify adding the extra
data.


>  Why would that clause apply to attributes any less than it applies to
> DIEs?
>
> It does apply equally.  However, nearly all attribute descriptions are
> specified as "may have" and therefore can be omitted freely without being
> non-conforming.
>

Fair point (& the same goes for the example I was going to give of omitting
DW_TAG_friends - they're "may" too)

Types are a bit more vague (as to whether omitting unreferenced types is
supported by the standard) DWARF 4 just says "Structure, union, and class
types are represented by debugging information entries ...".

Any size numbers for this change?

In any case, it seems like it might make sense for you to upstream your
template naming change and put it under the PS4 debugger tuning option, and
put this change there too, once the motivation for it is in-tree. At that
point, while I'd be curious about the size tradeoff, it'd be essentially
academic.

- David


> --paulr
>
>
>
> *From:* David Blaikie [mailto:dblaikie at gmail.com]
> *Sent:* Wednesday, December 09, 2015 11:28 AM
>
> *To:* Robinson, Paul
> *Cc:* Marshall, Peter; llvm-dev; cfe-commits (cfe-commits at lists.llvm.org)
> *Subject:* Re: [PATCH] D14358: DWARF's forward decl of a template should
> have template parameters.
>
>
>
>
>
>
>
> On Wed, Dec 9, 2015 at 11:11 AM, Robinson, Paul <
> Paul_Robinson at playstation.sony.com> wrote:
>
> Actually no, we prefer to have the original typedef names in the
> instantiation name, for source fidelity.
>
>
>
> Then perhaps you should keep this change in your tree too - since that's
> where the need is?
>
>
>
>   "Name as it is in the source" or something reasonably close.  Unwrapping
> typedefs is going too far.
>
>
>
> Yet this isn't the choice upstream in Clang or GCC. I don't know about
> other DWARF generators, but it seems your interpretation isn't the way some
> other people/implementers are reading the DWARF spec.
>
> [This seems like it would present a multitude of challenges to any DWARF
> debugger dealing with this kind of debug info - it'd have to know far more
> about the rules of the C++ language (which you've previously argued in
> favor of avoiding) to perform a variety of operations if the types don't
> match up fairly trivially.]
>
> In any case, arguably 5.5.8 (Class Template Instantiations) 1 only applies
> to definitions of a type, not declarations. ("Each formal parameterized
> type declaration appearing in the template definition is represented by a
> debugging information entry with the tag DW_TAG_template_type_parameter")
> which, I agree, seems like a bug in the spec to not /allow/ them on
> declarations, but I'd equally argue requiring them would seem too narrow to
> me.
>
>
>
> Re. "looseness" of the DWARF spec, it is not so loose as you like to
> think.  Attributes tend to be fairly optional or can be used "in novel
> ways" but the DIEs and their relationships are not like that.  "Where this
> specification provides a means for describing the source language,
> implementors are expected to adhere to that specification."
>
>
>
> Why would that clause apply to attributes any less than it applies to
> DIEs? It seems like a fairly broad statement.
>
> I forget whether we already discussed it - but do you have any size data
> (preferably/possibly from a fission build or otherwise measurement of "just
> the debug info" not the whole binary) on, for example, a clang selfhost?
>
> - Dave
>
>
>
> --paulr
>
>
>
> *From:* David Blaikie [mailto:dblaikie at gmail.com]
> *Sent:* Wednesday, December 09, 2015 10:49 AM
> *To:* Robinson, Paul
> *Cc:* Marshall, Peter; llvm-dev; cfe-commits (cfe-commits at lists.llvm.org)
>
>
> *Subject:* Re: [PATCH] D14358: DWARF's forward decl of a template should
> have template parameters.
>
>
>
>
>
>
>
> On Wed, Dec 9, 2015 at 10:40 AM, Robinson, Paul <
> Paul_Robinson at playstation.sony.com> wrote:
>
> That doesn't seem to be the DWARF I'm seeing from Clang (& it'd be
> surprising if we used the typedef (or otherwise non-canonical) name in the
> class name):
>
>
>
> Finally getting back to this…..  Ha.  We don't unwrap the typedefs ("name
> as it is in the source"), while the upstream compiler does.
>
>
>
> Yeah, I imagine you'd want to fix that as I expect it would cause you
> other problems, no? (or is there some reason you have this change to the
> compiler? I imagine it'd be hard to have that divergence by accident?)
>
>
>
> Providing the template-parameter DIEs is still the correct thing to do per
> the DWARF
>
> spec.
>
>
>
> I still don't agree that the DWARF we produce here is incorrect (the DWARF
> spec is pretty loose on "correctness" of DWARF). If there's some practical
> problem/use case it'd be useful to understand it so we make sure we're
> fixing it the right way.
>
> - Dave
>
>
>
> --paulr
>
>
>
> *From:* David Blaikie [mailto:dblaikie at gmail.com]
> *Sent:* Friday, November 13, 2015 11:21 AM
> *To:* Marshall, Peter; llvm-dev
> *Cc:* Robinson, Paul
>
>
> *Subject:* Re: [PATCH] D14358: DWARF's forward decl of a template should
> have template parameters.
>
>
>
>
>
>
>
> On Fri, Nov 13, 2015 at 6:16 AM, <Peter_Marshall at sn.scee.net> wrote:
>
> Hi Paul,
>
> Sorry for the delay, I've been out of the office.
>
> I think this example shows that name matching does not always work:
>
> template<typename T> class A {
> public:
>         A(T val);
> private:
>         T x;
> };
>
> struct B {
>         typedef float MONKEY;
>
>         A<MONKEY> *p;
> };
>
> B b;
>
> struct C {
>         typedef int MONKEY;
>
>         A<MONKEY> *p;
> };
>
> C c;
>
> This gives this DWARF:
>
> +-0000003f DW_TAG_structure_type "B"
>    -DW_AT_name  DW_FORM_strp  "B"
>   +-00000047 DW_TAG_member "p"
>      -DW_AT_name  DW_FORM_strp  "p"
>     +-DW_AT_type  DW_FORM_ref4  0x00000054
>       +-00000054 DW_TAG_pointer_type
>         +-DW_AT_type  DW_FORM_ref4  0x00000059
>           +-00000059 DW_TAG_class_type "A<MONKEY>"
>              -DW_AT_name  DW_FORM_strp  "A<MONKEY>"
>              -DW_AT_declaration  DW_FORM_flag_present
>
> +-00000073 DW_TAG_structure_type "C"
>    -DW_AT_name  DW_FORM_strp  "C"
>   +-0000007b DW_TAG_member "p"
>      -DW_AT_name  DW_FORM_strp  "p"
>     +-DW_AT_type  DW_FORM_ref4  0x00000088
>       +-00000088 DW_TAG_pointer_type
>         +-DW_AT_type  DW_FORM_ref4  0x0000008d
>           +-0000008d DW_TAG_class_type "A<MONKEY>"
>              -DW_AT_name  DW_FORM_strp  "A<MONKEY>"
>              -DW_AT_declaration  DW_FORM_flag_present
>
>
>
> That doesn't seem to be the DWARF I'm seeing from Clang (& it'd be
> surprising if we used the typedef (or otherwise non-canonical) name in the
> class name):
>
> (I've trimmed a few irrelevant attributes)
>
> 0x0000001e:   DW_TAG_variable [2]
>
>                 DW_AT_name [DW_FORM_strp]       ( .debug_str[0x0000004c] =
> "b")
>
>                 DW_AT_type [DW_FORM_ref4]       (cu + 0x0033 =>
> {0x00000033})
>
>
>
> 0x00000033:   DW_TAG_structure_type [3] *
>
>                 DW_AT_name [DW_FORM_strp]       ( .debug_str[0x00000059] =
> "B")
>
>
>
> 0x0000003b:     DW_TAG_member [4]
>
>                   DW_AT_name [DW_FORM_strp]     ( .debug_str[0x0000004e] =
> "p")
>
>                   DW_AT_type [DW_FORM_ref4]     (cu + 0x0048 =>
> {0x00000048})
>
>
>
> 0x00000047:     NULL
>
>
>
> 0x00000048:   DW_TAG_pointer_type [5]
>
>                 DW_AT_type [DW_FORM_ref4]       (cu + 0x004d =>
> {0x0000004d})
>
>
>
> 0x0000004d:   DW_TAG_class_type [6]
>
>                 DW_AT_name [DW_FORM_strp]       ( .debug_str[0x00000050] =
> "A<float>")
>
>                 DW_AT_declaration [DW_FORM_flag_present]        (true)
>
>
>
> 0x00000052:   DW_TAG_variable [2]
>
>                 DW_AT_name [DW_FORM_strp]       ( .debug_str[0x0000005b] =
> "c")
>
>                 DW_AT_type [DW_FORM_ref4]       (cu + 0x0067 =>
> {0x00000067})
>
>
>
> 0x00000067:   DW_TAG_structure_type [3] *
>
>                 DW_AT_name [DW_FORM_strp]       ( .debug_str[0x00000064] =
> "C")
>
>
>
> 0x0000006f:     DW_TAG_member [4]
>
>                   DW_AT_name [DW_FORM_strp]     ( .debug_str[0x0000004e] =
> "p")
>
>                   DW_AT_type [DW_FORM_ref4]     (cu + 0x007c =>
> {0x0000007c})
>
>
>
> 0x0000007b:     NULL
>
>
>
> 0x0000007c:   DW_TAG_pointer_type [5]
>
>                 DW_AT_type [DW_FORM_ref4]       (cu + 0x0081 =>
> {0x00000081})
>
>
>
> 0x00000081:   DW_TAG_class_type [6]
>
>                 DW_AT_name [DW_FORM_strp]       ( .debug_str[0x0000005d] =
> "A<int>")
>
>                 DW_AT_declaration [DW_FORM_flag_present]        (true)
>
>
>
>
>
> As there are no template parameters for the forward declaration of either
> A<MONKEY>
> they are indistinguishable.
>
> The reason we currently have no need for the parameters in a template name
> is because we
> reconstruct template names from their parameter tags. This allow the
> pretty printing to match
> the templates from the DWARF to match our demangled symbols from the ELF
> symbol table.
>
> -Pete
>
>
>
>
> From:        "Robinson, Paul" <Paul_Robinson at playstation.sony.com>
> To:        David Blaikie <dblaikie at gmail.com>, "Marshall, Peter" <
> Peter_Marshall at sn.scee.net>
> Cc:        "reviews+D14358+public+d3104135076f0a10 at reviews.llvm.org"
>    <reviews+D14358+public+d3104135076f0a10 at reviews.llvm.org>,
> "cfe-commits (cfe-commits at lists.llvm.org)" <cfe-commits at lists.llvm.org>
> Date:        10/11/2015 01:08
> Subject:        RE: [PATCH] D14358: DWARF's forward decl of a template
> should have template parameters.
> ------------------------------
>
>
>
>
> | when/where/why are types acquired from the mangled names of ELF
> symbols, rather than from corresponding DWARF?
>
> Pete, can you help me out here?  David seems to want an ironclad case for
> not being able to do something any other way, before he will let me put the
> template type parameters on the declaration of a template instantiation.
>  (He does not deny that doing so would be valid DWARF, only that it can't
> possibly be *useful* DWARF.)
> Thanks,
> --paulr
>
> *From:* David Blaikie [mailto:dblaikie at gmail.com <dblaikie at gmail.com>]
> * Sent:* Monday, November 09, 2015 4:08 PM
> * To:* Robinson, Paul
> * Cc:* reviews+D14358+public+d3104135076f0a10 at reviews.llvm.org;
> cfe-commits (cfe-commits at lists.llvm.org)
> * Subject:* Re: [PATCH] D14358: DWARF's forward decl of a template should
> have template parameters.
>
>
>
> On Mon, Nov 9, 2015 at 3:55 PM, Robinson, Paul <
> Paul_Robinson at playstation.sony.com> wrote:
> | Why is matching by name insufficient/not correct?
> I'm told we look at the mangled names in the ELF symbol table, demangle
> them, and look in the DWARF for the corresponding types.
>
> Not quite sure I follow that - when/where/why are types acquired from the
> mangled names of ELF symbols, rather than from corresponding DWARF? (eg:
> the DWARF describing the type of a function's parameter?)
>
>   Now, the mangled name (for predefined types in particular) provides a
> type description, not the name-as-emitted-by-Clang, and in fact the same
> type can have more than one name ('const int' and 'int const' for a trivial
> example).  The name Clang provides in the DWARF does not necessarily match
> the name produced by the demangler; this makes name-matching way more
> trouble than you'd think.  We're not interested in teaching the debugger
> how to parse template instantiation names.
> Having the template type parameter means we have an unambiguous
> description of the type, and can match it easily.
>
> | including unreferenced entities fails source fidelity
> I'll assume you meant to say _excluding_ unreferenced entities fails
> source fidelity,
>
> Indeed
>
> which is quite true, but there is a valid engineering tradeoff in that
> what the DWARF actually contains (or not, in the case of, say, unused
> function declarations or unreferenced class contents) represents one
> possible valid source that could have produced the same object.  (I'm
> curious why an unreferenced formal parameter of a function still gets
> described, if this is your argument for omitting template parameters.)
>
> Omitting parameters would make the function description unusable for
> callers, for example - so there's some value in describing them so that a
> debugger can evaluate expressions involving calls to the function, yes?
>
> Omitting template parameters however is not the same as omitting
> unreferenced entities, because the template parameters *are* referenced—by
> the template instantiation itself;
>
> Not quite sure I follow that logic. It's quite possible to have
> unreferenced template parameters:
>
>  template<typename>
>  void f() { }
>
> and, omitting them from the source does not produce a valid program.
>
> Omitting the names still produces a valid program - though I'm not quite
> sure which omission you're referring to. (& even if we omit the names, we
> still describe the parameters - as we do for unused/unnamed function
> parameters)
>
>   Now, one of the 3 debuggers Clang explicitly supports (i.e. gdb) seems
> not to mind that they're missing, but the other two would benefit from
> having these things, and I would really like to have Clang produce these
> things.
>
> It sounds like the LLDB bug you cited is being treated as an LLDB bug, not
> a Clang one, for now. So I'm not sure it's relevant to justifying Clang
> changes just yet, unless they come back & suggest that they don't actually
> have enough information to implement the features they would like to
> implement.
>
> & equally I'd like to understand the features that you'd like to build
> with this info that can't be built without it (as a minimum: features that
> GDB doesn't support, since any features GDB does support seem to be
> implementable with the current info Clang and GCC emit)
>
> - David
>
>
> Thanks,
> --paulr
>
> *From:* David Blaikie [mailto:dblaikie at gmail.com]
> * Sent:* Monday, November 09, 2015 1:46 PM
>
> * To:* Robinson, Paul
> * Cc:* reviews+D14358+public+d3104135076f0a10 at reviews.llvm.org;
> cfe-commits (cfe-commits at lists.llvm.org)
> * Subject:* Re: [PATCH] D14358: DWARF's forward decl of a template should
> have template parameters.
>
>
>
> On Thu, Nov 5, 2015 at 11:05 AM, Robinson, Paul <
> Paul_Robinson at playstation.sony.com> wrote:
> | What was your primary motivation?
> A similar concern to PR20455 from our own debugger.  It much helps
> matching up the forward declaration and definition to have the parameters
> properly specified.
>
> Why is matching by name insufficient/not correct?
>
>
> | maybe it's possible to remangle the template using just the string name
> I have no idea what you're talking about here.
>
> Looking at PR20455 you linked, LLDB isn't finding the right function
> because of mangling:
> call to a function 'basic_string<char, char_traits<char>
> >::operator[](int) const' ('_ZNK12basic_stringIc17char_traits<char>EixEi')
> that is not present in the target
> It hasn't created the correct mangled name of operator[] - what I was
> saying is it might be possible to parse the template parameter from the
> pretty name, and use that to produce the mangled name. It /looks/ like GDB
> can manage this. Maybe only because we also include the mangled name of the
> member function? Not sure.
>
> | | Choosing to emit a forward/incomplete declaration in the first place
> fails source fidelity,
> | How so?
> When the source has a full definition but Clang chooses to emit only the
> declaration, per CGDebugInfo.cpp/shouldOmitDefinition().
>
> Sure, in the same way that including unreferenced entities fails source
> fidelity - all tradeoffs to reduce debug info size.
>
> Though the behavior is visible in a simpler example that doesn't have that
> failing (& if your change goes in, the test case should probably be
> simplified like this):
>
> template<typename T> struct foo;
> foo<int> *f;
>
> --paulr
>
> *From:* David Blaikie [mailto:dblaikie at gmail.com]
> * Sent:* Thursday, November 05, 2015 12:10 AM
> * To:* Robinson, Paul
> * Cc:* reviews+D14358+public+d3104135076f0a10 at reviews.llvm.org;
> cfe-commits (cfe-commits at lists.llvm.org)
>
> * Subject:* Re: [PATCH] D14358: DWARF's forward decl of a template should
> have template parameters.
>
>
>
> On Wed, Nov 4, 2015 at 11:32 PM, Robinson, Paul <
> Paul_Robinson at playstation.sony.com> wrote:
> Would citing PR20455 help?  It wasn't actually my primary motivation but
> it's not too far off.  Having the template parameters there lets you know
> what's going on in the DWARF, without having to fetch and parse the name
> string of every struct you come across.  Actually I'm not sure parsing the
> name string is unambiguous either; each parameter is either a typename, or
> an expression, but without the parameter DIEs you don't know which,
> a-priori.  (What does <foo> mean? Depends on whether you think it should be
> a type name or a value; you can't tell, syntactically, you have to do some
> lookups.  Ah, but if you had the parameter DIEs, you would Just Know.)
>
> For LLDB's needs, I'm not sure it's sufficient either - but I wouldn't
> mind an answer before we use it as the basis for this change (it sounds
> like maybe it's possible to remangle the template using just the string
> name, rather than needing an explicit representation of the parameters)
>
> What was your primary motivation?
>
>  Choosing to emit a forward/incomplete declaration in the first place
> fails source fidelity,
>
> How so? You might have only a template declaration (template<typename T>
> struct foo; foo<int> *f;) or you may've only instantiated the declaration
> (the C++ language requires you to instantiate or avoid instantiating
> certain things in certain places, so in some contexts you /only/ have an
> instantiated declaration, not a definition)
>
> but it is a practical engineering tradeoff of compile/link performance
> against utility; and, after all, the source *could* have been written that
> way, with no semantic difference.  But, if we're going to emit a white-lie
> incomplete declaration, we should do so correctly.
>
> Again, "correct" in DWARF is a fairly nebulous concept.
>
> --paulr
>
> P.S. We should talk about this forward-declaration tactic wrt LTO
> sometime.  I have a case where a nested class got forward-declared; it's
> entirely conceivable that the outer class with the inner forward-declared
> class would end up being picked by LTO, leaving the user with no debug info
> for the inner class contents.
>
> I believe this Just Works(tm). The things that can vary per-insntance of a
> type (implicit special members, member template implicit specializations,
> and nested types*) are not added to the type's child list, but they
> reference the child as their parent. So they continue to apply no matter
> which instance of the type is picked for uniquing (because of the
> name-based referencing, so the nested type definition just says "my parent
> is _Zfoo" and whatever _Zfoo we end up picking in the LTO linking/metadata
> deduplication will serve that role just fine)
>
> * we could just do a better job of modelling nested types (& other
> non-globally scoped types) in a way that more closely models the source by
> emitting a declaration where they were declared, and a definition where
> they are defined (with the usual DW_AT_specification to wire them up)
>
>
> *From:* David Blaikie [mailto:dblaikie at gmail.com]
> * Sent:* Wednesday, November 04, 2015 8:30 PM
> * To:* reviews+D14358+public+d3104135076f0a10 at reviews.llvm.org; Robinson,
> Paul
> * Subject:* Re: [PATCH] D14358: DWARF's forward decl of a template should
> have template parameters.
>
>
>
> On Wed, Nov 4, 2015 at 5:53 PM, Paul Robinson via cfe-commits <
> cfe-commits at lists.llvm.org> wrote:
> probinson added a comment.
>
> GCC 4.8.4 on Linux doesn't produce these, but DWARF 4 section 5.5.8 says a
> class template instantiation is just like the equivalent non-template class
> entry, with the exception of the template parameter entries.  I read that
> as meaning an incomplete description (i.e. with DW_AT_declaration) lets you
> omit all the other children, but not the template parameters.
>
> As usual, I think it's pretty hard to argue that DWARF /requires/ anything
> (permissive & all that). And I'm not sure that having these is particularly
> valuable/useful - what use do you have in mind for them?
>
> Wouldn't hurt to have some size info about the cost here, though I don't
> imagine it's massive, it does open us up to emitting a whole slew of new
> types (the types the template is instantiated with, and anything that
> depends on - breaking/avoiding type edges can, in my experience, be quite
> beneficial (I described an example of this in my lightning talk last
> week)).
>
>
> I don't think omitting the template DIEs was an intentional optimization,
> in the sense of being a decision separate from deciding to emit the
> incomplete/forward declaration in the first place.  They were just omitted
> because we were omitting everything, but everything turns out to be
> non-compliant.
>
>
> http://reviews.llvm.org/D14358
>
>
>
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>
>
>
>
>
>
> **********************************************************************
> This email and any files transmitted with it are confidential and intended
> solely for the use of the individual or entity to whom they are addressed.
> If you have received this email in error please notify postmaster at scee.net
> This footnote also confirms that this email message has been checked for
> all known viruses.
> Sony Computer Entertainment Europe Limited
> Registered Office: 10 Great Marlborough Street, London W1F 7LP, United
> Kingdom
> Registered in England: 3277793
> **********************************************************************
>
> P* Please consider the environment before printing this e-mail*
>
>
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20151209/6cc5ef50/attachment.html>


More information about the llvm-dev mailing list