[Bug c++/16295] New: -pedantic option resolves compiler error "sorry, unimplemented: ..."

[Bug c++/16295] New: -pedantic option resolves compiler error "sorry, unimplemented: ..."

[kai dot ludwig at uni-tuebingen dot de]

g++-3.3.3 aborts with message

"sorry, unimplemented: `tree_vec' not supported by dump_expr"

when compiling WITHOUT -pedantic or -pedantic-errors flags.
WITH these flags it compiles without complaint.

-- 
           Summary: -pedantic option resolves compiler error "sorry,
                    unimplemented: ..."
           Product: gcc
           Version: 3.3.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: kai dot ludwig at uni-tuebingen dot de
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[kai dot ludwig at uni-tuebingen dot de]

------- Additional Comments From kai dot ludwig at uni-tuebingen dot de  2004-06-30 11:43 -------
Created an attachment (id=6660)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6660&action=view)
preprocessed file compiled without -pedantic flag : fails

generated by:

g++ -I. -I/usr/local/include/loki -I/home/kl/software/src/quantity/Quantities
-I/home/kl/software/src/quantity/BSUtilities -v -save-temps -Wall -Werror
main.cpp -L/home/kl/software/lib -lPhysicalQuantities -lQuantity -lBSUtilities

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[kai dot ludwig at uni-tuebingen dot de]

------- Additional Comments From kai dot ludwig at uni-tuebingen dot de  2004-06-30 11:44 -------
Created an attachment (id=6661)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6661&action=view)
preprocessed file compiled with -pedantic flag: succeeded

generated by

g++ -I. -I/usr/local/include/loki -I/home/kl/software/src/quantity/Quantities
-I/home/kl/software/src/quantity/BSUtilities -v -save-temps -Wall -Werror
-pedantic main.cpp -L/home/kl/software/lib -lPhysicalQuantities -lQuantity
-lBSUtilities

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[kai dot ludwig at uni-tuebingen dot de]

------- Additional Comments From kai dot ludwig at uni-tuebingen dot de  2004-06-30 11:46 -------
Created an attachment (id=6662)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6662&action=view)
can't compile this main routine without -pedantic flag


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[kai dot ludwig at uni-tuebingen dot de]

------- Additional Comments From kai dot ludwig at uni-tuebingen dot de  2004-06-30 11:56 -------
(In reply to comment #0)
> g++-3.3.3 aborts with message
> 
> "sorry, unimplemented: `tree_vec' not supported by dump_expr"
> 
> when compiling WITHOUT -pedantic or -pedantic-errors flags.
> WITH these flags it compiles without complaint.

My environment:
SuSE Linux 9.1 with included GNU compiler collection

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-06-30 14:30 -------
Confirmed. The two preprocessed source files are actually identitical. 
I get this here: 
 
g/x> /home/bangerth/bin/gcc-3.3.4-pre/bin/c++ -c w.ii  
In file included 
from /home/kl/software/src/quantity/Quantities/PhysicalQuantiti 
es/Acceleration.h:27, 
                 from /home/kl/software/src/quantity/Quantities/PhysicalQuantiti 
es/PhysicalQuantities.h:28, 
                 from main.cpp:1: 
/home/kl/software/src/quantity/Quantities/Quantity/Variable.h:865: internal 
comp 
iler error: tree 
   check: expected tree_list, have tree_vec in dump_decl, at cp/error.c:978 
Please submit a full bug report, 
with preprocessed source if appropriate. 
See <URL:http://gcc.gnu.org/bugs.html> for instructions. 
 
 
g/x> /home/bangerth/bin/gcc-3.3.4-pre/bin/c++ -c w.ii -pedantic 
 
It is indeed strange that the code should compile with -pedantic, but 
not otherwise. Note that the different message I get is likely due to 
the fact that I configured with --enable-checking. 
 
I'll try to come up with a smaller example. 
 
W. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-06-30 14:30:56
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-06-30 14:41 -------
One note: the code is actually invalid. We crash on the line 
       typedef NonPrefixable<GenericClass<AddT>, GenericUnit, DT> AddU; 
but NonPrefixable really only takes two template arguments, not three. 
 
W. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-06-30 15:00 -------
OK, here's something small: 
---------------- 
template <typename> struct O { 
    template <typename> struct I { 
        struct II {}; 
    }; 
}; 
 
template <typename T> struct X { 
    typedef typename O<T>::I<int>::II type1; 
    typedef O<type1>                  type2; 
}; 
---------------------- 
It still crashes with -pedantic, so something is different. Note that 
the crash also goes away if I fix the code: one has to write 
  typedef typename O<T>::template I<int>::II type1; 
 
However, this is fixed in 3.4 and mainline, where we get an error message 
about this, rather than an ICE. 
 
W. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-06-30 15:04 -------
And as a last note: If I fix the code like so: 
 
template<template<class, class> class Q1, class GT1, class Head1, 
           class Tail1, class DU1, class U1, 
         template<class, class> class Q2, class GT2, class Head2, 
           class Tail2, class DU2, class U2, class DT> 
  struct GenerateVariable 
    <Q1<Quantity<GT1, Typelist<Head1, Tail1>, DU1, DT>, U1>, 
                Q2<Quantity<GT2, Typelist<Head2, Tail2>, DU2, DT>, U2> > 
  { 
     private: 
 
       typedef typename Dimension<GT1>::template Add<Dimension<GT2> >::Class 
                                                                   AddT; 
       typedef NonPrefixable<GenericClass<AddT>, GenericUnit> AddU; 
     //... 
  }; 
 
then it compiles cleanly. Note that the 'template' keyword in the declaration 
of AddT is necessary. I removed the erroneous third template parameter 
in the declaration of AddU, but I have of course no clue what you originally 
intended there. 
 
W. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-invalid-code


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |3.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[bernd dot speiser at uni-tuebingen dot de]

------- Additional Comments From bernd dot speiser at uni-tuebingen dot de  2004-07-01 11:30 -------

(In reply to comment #8)

I'm the original writer of this code, so I'll comment on the third parameter.

> And as a last note: If I fix the code like so:
>
> template<template<class, class> class Q1, class GT1, class Head1,
>            class Tail1, class DU1, class U1,
>          template<class, class> class Q2, class GT2, class Head2,
>            class Tail2, class DU2, class U2, class DT>
>   struct GenerateVariable
>     <Q1<Quantity<GT1, Typelist<Head1, Tail1>, DU1, DT>, U1>,
>                 Q2<Quantity<GT2, Typelist<Head2, Tail2>, DU2, DT>, U2> >
>   {
>      private:
>
>        typedef typename Dimension<GT1>::template Add<Dimension<GT2> >::Class
>                                                                    AddT;
>        typedef NonPrefixable<GenericClass<AddT>, GenericUnit> AddU;
>      //...
>   };
>
> then it compiles cleanly. Note that the 'template' keyword in the declaration
> of AddT is necessary. I removed the erroneous third template parameter
> in the declaration of AddU, but I have of course no clue what you originally
> intended there.

NonPrefixable really takes THREE parameters, with DT being the last one and
having a default value of double. Thus, if you leave it off, as in this example,
then the last parameter to NonPrefixable is just taken as double.

I understand that 3.4 handles the missing `template' keyword now correctly, right?
We'll test this, and report back what we find.

Still, the question remains, why the -pedantic flag results in NOT crashing??
Bernd



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[kai dot ludwig at uni-tuebingen dot de]

------- Additional Comments From kai dot ludwig at uni-tuebingen dot de  2004-07-01 13:23 -------
(In reply to comment #7)
> OK, here's something small: 
> ---------------- 
> template <typename> struct O { 
>     template <typename> struct I { 
>         struct II {}; 
>     }; 
> }; 
>  
> template <typename T> struct X { 
>     typedef typename O<T>::I<int>::II type1; 
>     typedef O<type1>                  type2; 
> }; 
> ---------------------- 
> It still crashes with -pedantic, so something is different. Note that 
> the crash also goes away if I fix the code: one has to write 
>   typedef typename O<T>::template I<int>::II type1; 
>  
> However, this is fixed in 3.4 and mainline, where we get an error message 
> about this, rather than an ICE. 

Indeed that's it. After introducing the 'template' keyword as mentioned above
into the Variable.h source, g++-3.3.3 works as expected.

Thank you Wolfgang,
Kai


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295

[Bug c++/16295] -pedantic option resolves compiler error "sorry, unimplemented: ..."

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-07-01 14:22 -------
> Still, the question remains, why the -pedantic flag results in NOT 
> crashing?? 
 
I have no clue, but since this case was fixed already, it's hard to tell. 
One could go about and try to find a _small_ testcase for 3.3.x in 
which it ICEs unless we give -pedantic. Then we could figure out what 
is happening there and and see if the same patch could also be 
applied to mainline. However, since development generally only happens 
on mainline, I doubt that anyway has the energy to do things like this. 
Let's just wait until we get a similar bug against mainline, and then 
reconsider whether we need to do the necessary investigations. 
 
W. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16295