669693 - Dehydra: Unhandled type_pack_expansion/decltype_type etc.

Status: RESOLVED FIXED

(Developer Infrastructure :: Source Code Analysis, defect)

Description

[Andrew McCreight [:mccr8]]

When running Dehydra on some files on moz-central, I get a ton of console spew:

/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:67:1: warning: attribute ignored in declaration of ‘class<unnamed>::TransactionPoolEv\
entTarget’
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:67:1: warning: attribute for ‘class<unnamed>::TransactionPoolEventTarget’ must follow\
 the ‘class’ keyword
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp: In constructor ‘std::exception::exception()’:
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_pack_expansion: _ArgTypes ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_pack_expansion: _ArgTypes ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_pack_expansion: _ArgTypes ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_pack_expansion: _ArgTypes ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_pack_expansion: _ArgTypes ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_pack_expansion: _ArgTypes ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_pack_expansion: _ArgTypes ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_pack_expansion: _ArgTypes ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled decltype_type: decltype (((_Tp1)((declval<_Args1>)\
()...), std::__sfinae_types::__one()))
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_argument_pack: _Args ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled decltype_type: decltype ((static_cast<_Tp1>(declva\
l<_Arg1>()), std::__sfinae_types::__one()))
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_argument_pack: _Arg
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_argument_pack: _Args ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_argument_pack: _From
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_argument_pack: _Tp
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_argument_pack: _Tp, _Up
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_argument_pack: _Tp, _Up, _Vp ...
/home/amccreight/tm/dom/indexedDB/AsyncConnectionHelper.cpp:579:1: warning: Dehydra: Unhandled type_argument_pack: _Tp ...

Comment 1

[Andrew McCreight [:mccr8]]

In IRC, jcranmer said this probably means it isn't reifying some structure.

The warning is produced in dehydra_types.c, and is on a default case in a switch  on TREE_CODE (type).

Comment 2

[Andrew McCreight [:mccr8]]

TYPE_PACK_EXPANSION is defined in cp/cp-tree.def and "Represents a type expression that will be expanded into a list of types when instantiated with one or more argument packs".  TYPE_ARGUMENT_PACK is defined in the same file and "Represents an argument pack of types (or templates)".

I'm not really sure what that means, except that the former sounds like a type-level function, and the latter sounds like an argument for a type level function.

One instance of this I found looks fairly innocuous.  There's a method:
  unsigned size() const { return descriptors_.size(); }
Where descriptors_ is defined as:
  std::vector<base::FileDescriptor> descriptors_;
The class itself doesn't look all that weird, except that it is the subtype of a template.

I'm not really sure how this would be represented in Dehydra, unless it has to cancel out the type abstraction and application somehow and create a template.

Comment 3

[Joshua Cranmer [:jcranmer]]

Simple test case that doesn't require STL:
  template <typename... A> void func(A... args) {}
  template <typename... B> class C {};

B... is the type_argument_pack/
A... is the type_pack_expansion.

Comment 4

[Joshua Cranmer [:jcranmer]]

This also needs to be added to dehydra_types.js and lazy_types.js.

A complete list of the types in gcc-4.6.1 headers and which files handle them:

+---- dehydra_types.c
|+--- libs/unstable/lazy_types.js
||+-- libs/unstable/dehydra_types.js
|||
-- ALL
xxx offset_type
xxx enumeral_type
xxx boolean_type
xxx integer_type
xxx real_type
xxx pointer_type
xxx reference_type
    nullptr_type
xxx fixed_point_type
xxx complex_type
xxx vector_type
xxx array_type
xxx record_type
xxx union_type
    qual_union_type
xxx void_type
xxx function_type
xxx method_type
    lang_type
-- C++
x   template_template_parm
xxx template_type_parm
xxx typename_type
xx  typeof_type
x   bound_template_template_parm
    unbound_class_template
    type_argument_pack
    type_pack_expansion
    decltype_type
-- Objective-C
    class_interface_type
    class_implementation_type
    category_interface_type
    category_implementation_type
    protocol_interface_type
-- Ada
    unconstrained_array_type
[Java and C family define no types]

Some notes:
qual_union_type [Similar to UNION_TYPE, except that the expressions in DECL_QUALIFIER in each FIELD_DECL determine what the union contains.  The first field whose DECL_QUALIFIER expression is true is deemed to occupy the union.]
lang_type [Unknown type that the front-end has to worry about]
unbound_class_template [For template template argument of the form `T::template C'.]
type_argument_pack [A collection of type arguments]
type_pack_expansion [A type expression that expands into a list of types]
decltype_type [Appears to be C++0x-specified clone of typeof()]

[ I don't care about Objective C or Ada, just listed them for completeness ]

Comment 5

[(dormant account)]

Attachment: decltype: this works fine for me

Comment 6

[Joshua Cranmer [:jcranmer]]

Apologies, the decltype only appears to be emitted if the type is parameter-dependent, e.g.:
template <typename A, typename B> void f(A x, B y, decltype(x / y) z) {}

Comment 7

[Joshua Cranmer [:jcranmer]]

Attachment: Adds support for decltype, variadic templates

This adds support for decltype and variadic templates in dehydra_types.c, lazy_types.js, and dehydra_types.js, at least as near as I can tell.

Comment 8

[(dormant account)]

Comment on attachment 554897 [details] [diff] [review]
Adds support for decltype, variadic templates

># HG changeset patch
># Parent 468f253e4ae71c74fae2366619693bac10c38827
>
>diff --git a/dehydra_types.c b/dehydra_types.c
>--- a/dehydra_types.c
>+++ b/dehydra_types.c
>@@ -453,16 +453,19 @@ static jsval dehydra_convert_type_cached
> 
>       break;
>     }
>   case COMPLEX_TYPE:
>   case VECTOR_TYPE:
>     /* maybe should add an isTemplateParam? */
>   case TEMPLATE_TYPE_PARM:
>   case TYPENAME_TYPE:
>+  case TYPE_ARGUMENT_PACK:
>+  case TYPE_PACK_EXPANSION:
>+    // FIXME: variadic templates need better support
>     dehydra_defineStringProperty (this, obj, NAME, dehydra_typeString(type));
>     dehydra_defineProperty (this, obj, ISTYPENAME, JSVAL_TRUE);

This is kinda misleading. Feels like isTypename should not be set for these new cases.

>     break;
>   case FUNCTION_TYPE:
>   case METHOD_TYPE:
>     dehydra_convertAttachFunctionType (this, obj, type);
>     break;
>   case ARRAY_TYPE:
>@@ -485,16 +488,17 @@ static jsval dehydra_convert_type_cached
>       if (type_name) {
>         // what the hell else would a template be?
>         xassert (DECL_P (type_name));
>         dehydra_defineStringProperty (this, obj, NAME, decl_as_string (type_name, 0));
>         break;
>       }
>     }
>   case TYPEOF_TYPE:
>+  case DECLTYPE_TYPE:
>     // avoid dealing with typeof mess
>     dehydra_defineStringProperty (this, obj, "typeof_not_implemented", type_as_string (type, 0));

Same here. Should not confuse decltype and typeoftype

>+        elif self.lang == 'c++0x':
>+            command += " -std=c++0x"
>         command += " -c -fplugin=../gcc_" + self.plugin + ".so -o /dev/null"
>         if "PLUGINS_MOZ" in config_opts:
>             command += ' -fplugin-arg="' + self.jsfile
>             argprefix = "--"
>         else :
>             command += " -fplugin-arg-gcc_" + self.plugin + "-script=" + self.jsfile
>             argprefix = "-fplugin-arg-gcc_" + self.plugin + "-"
>         for key, value in self.arguments.iteritems():
>@@ -182,29 +184,26 @@ def extractTests(filename):
>     arguments = spec.get('args')
>     if arguments is None:
>         arguments = {}
> 
>     langs = spec.get('lang')
>     if langs is None:
>         langs = 'c,c++'
>     # ignore C tests if C is not enabled
>-    if not 'C_SUPPORT' in config_opts:
>-        # ignore C only tests
>-        if langs == 'c':
>-            return []
>-        # use only C++ for multiplatform tests
>-        langs = 'c++'
>+    supported = set(['c++', 'c++0x'])
>+    if 'C_SUPPORT' in config_opts:
>+        supported.update('c')
>     langs = langs.split(',')
>     for lang in langs:
>-        if lang not in ('c', 'c++'):
>+        if lang not in ('c', 'c++', 'c++0x'):

not you mean if lang not in supported?

Rest is good work

Comment 9

[Joshua Cranmer [:jcranmer]]

(In reply to Taras Glek (:taras) from comment #8)
> Comment on attachment 554897 [details] [diff] [review]
> Adds support for decltype, variadic templates
> 
> ># HG changeset patch
> ># Parent 468f253e4ae71c74fae2366619693bac10c38827
> >
> >diff --git a/dehydra_types.c b/dehydra_types.c
> >--- a/dehydra_types.c
> >+++ b/dehydra_types.c
> >@@ -453,16 +453,19 @@ static jsval dehydra_convert_type_cached
> > 
> >       break;
> >     }
> >   case COMPLEX_TYPE:
> >   case VECTOR_TYPE:
> >     /* maybe should add an isTemplateParam? */
> >   case TEMPLATE_TYPE_PARM:
> >   case TYPENAME_TYPE:
> >+  case TYPE_ARGUMENT_PACK:
> >+  case TYPE_PACK_EXPANSION:
> >+    // FIXME: variadic templates need better support
> >     dehydra_defineStringProperty (this, obj, NAME, dehydra_typeString(type));
> >     dehydra_defineProperty (this, obj, ISTYPENAME, JSVAL_TRUE);
> 
> This is kinda misleading. Feels like isTypename should not be set for these
> new cases.

As far as I can tell, the isTypename is more or less just an attribute which says "this is a template typename parameter", which the variadic pack types certainly are.

> >   case TYPEOF_TYPE:
> >+  case DECLTYPE_TYPE:
> >     // avoid dealing with typeof mess
> >     dehydra_defineStringProperty (this, obj, "typeof_not_implemented", type_as_string (type, 0));
> 
> Same here. Should not confuse decltype and typeoftype

As far as I can figure out, typeof and decltype are the same thing; gcc tells me to use decltype instead of typeof if you use typeof with -std=c++0x.

> >-        if lang not in ('c', 'c++'):
> >+        if lang not in ('c', 'c++', 'c++0x'):
> 
> not you mean if lang not in supported?

If ENABLE_C_SUPPORT is not enabled, this would throw an error.

Comment 10

[(dormant account)]

(In reply to Joshua Cranmer [:jcranmer] from comment #9)
> (In reply to Taras Glek (:taras) from comment #8)
> > Comment on attachment 554897 [details] [diff] [review]
> > Adds support for decltype, variadic templates
> > 
> > ># HG changeset patch
> > ># Parent 468f253e4ae71c74fae2366619693bac10c38827
> > >
> > >diff --git a/dehydra_types.c b/dehydra_types.c
> > >--- a/dehydra_types.c
> > >+++ b/dehydra_types.c
> > >@@ -453,16 +453,19 @@ static jsval dehydra_convert_type_cached
> > > 
> > >       break;
> > >     }
> > >   case COMPLEX_TYPE:
> > >   case VECTOR_TYPE:
> > >     /* maybe should add an isTemplateParam? */
> > >   case TEMPLATE_TYPE_PARM:
> > >   case TYPENAME_TYPE:
> > >+  case TYPE_ARGUMENT_PACK:
> > >+  case TYPE_PACK_EXPANSION:
> > >+    // FIXME: variadic templates need better support
> > >     dehydra_defineStringProperty (this, obj, NAME, dehydra_typeString(type));
> > >     dehydra_defineProperty (this, obj, ISTYPENAME, JSVAL_TRUE);
> > 
> > This is kinda misleading. Feels like isTypename should not be set for these
> > new cases.
> 
> As far as I can tell, the isTypename is more or less just an attribute which
> says "this is a template typename parameter", which the variadic pack types
> certainly are.
> 
> > >   case TYPEOF_TYPE:
> > >+  case DECLTYPE_TYPE:
> > >     // avoid dealing with typeof mess
> > >     dehydra_defineStringProperty (this, obj, "typeof_not_implemented", type_as_string (type, 0));
> > 
> > Same here. Should not confuse decltype and typeoftype
> 
> As far as I can figure out, typeof and decltype are the same thing; gcc
> tells me to use decltype instead of typeof if you use typeof with -std=c++0x.

Change it to "typeof/decltype_not_implemented" and r+ on the rest

Comment 11

[Joshua Cranmer [:jcranmer]]

Changed the decltype thing and pushed the rest.