When compiling a code with DMC I get the following message:

Internal error: cod2 3692
--- errorlevel 1

I have looked on the Digital Mars website for clues as to the meaning of the code, but cannot find any.  The code successfully compiles with GCC GNU 4.1.1 and  MSVC++ 7.1.

Any suggestions as to where to look within the code would be helpful and appreciated.

Gregg wrote:
> Any suggestions as to where to look within the code would be helpful and appreciated.


http://www.digitalmars.com/bugs.html

I have a certain familiarity with DMC++ (and other compiler) ICE bugs. Why
not post the code (or an effective snapshot of it) and I'll see if I
recognise anything.

Matthew

"Gregg" <no@spam.com> wrote in message news:eib632$1pd9$1@digitaldaemon.com...
> When compiling a code with DMC I get the following message:
>
> Internal error: cod2 3692
> --- errorlevel 1
>
> I have looked on the Digital Mars website for clues as to the meaning of the code, but cannot find any.  The code successfully compiles with GCC GNU 4.1.1 and  MSVC++ 7.1.
>
> Any suggestions as to where to look within the code would be helpful and appreciated.

Attached is about as simple an example of the problem as I can get.  As noted in my original post, it compiles (and links and runs) successfully with GCC GNU 4.1.1 and MSVC++ 7.1 and compiles successfully with Comeau 4.3.8 (using their online test tool).

When I run the code in its attached form I get the following compile time error message:

Internal error: cod1 1293
--- errorlevel 1

When class A is defined in separate source and header files, compiling the source file yields the error message I mentioned in my earlier post.

The compile command I am using is:

dmc -Ic:\dm\stlport\stlport TestInline.cpp

The problem seems to involve combinations of the use of inline, static, and std::pair.  I've seen this before and have work arounds, but that becomes cumbersome when dealing with existing library code (e.g., special versions of library code for use w/ DMC).

Thanks again for any help.


Matthew wrote:
> I have a certain familiarity with DMC++ (and other compiler) ICE bugs. Why
> not post the code (or an effective snapshot of it) and I'll see if I
> recognise anything.
> 
> Matthew
> 
> "Gregg" <no@spam.com> wrote in message news:eib632$1pd9$1@digitaldaemon.com...
>> When compiling a code with DMC I get the following message:
>>
>> Internal error: cod2 3692
>> --- errorlevel 1
>>

Gregg wrote:
> Attached is about as simple an example of the problem as I can get.

Thanks for doing this. It helps.

Hi

I think you've done an excellent job of paring it down for Walter's purposes - i.e. to be able to replicate, identify and fix the bug - but probably a little too much for the kind of workarounds I'm familiar with. ;-)

If Walter can get this fixed soonish, that's probably your best bet. If not, or if you can't wait for a new beta, I'd be happy to take a look at something a little more substantial. From what you've presented, several things come to mind, but they may not be appropriate in the real code. FWIW:

 - use a custom template pair instead of std::pair
 - use a non-template pair instead of std::pair
 - use a static reference and a static ptr+auto_ptr, e.g.

    typedef std::pair<int, int>    pair_t;
    static std::auto_ptr<pair_t>   b_(new pair_t(1, 1));
    static pair_t                            &b = *b_.get();

    (Note: this has thread-safety implications.)

 - move the static into another function / scope, e.g. into a local
namespace

That's all I can think of right now. HTH




"Gregg" <no@spam.com> wrote in message news:eidasr$p4v$1@digitaldaemon.com...
> Attached is about as simple an example of the problem as I can get.  As noted in my original post, it compiles (and links and runs) successfully with GCC GNU 4.1.1 and MSVC++ 7.1 and compiles successfully with Comeau 4.3.8 (using their online test tool).
>
> When I run the code in its attached form I get the following compile time error message:
>
> Internal error: cod1 1293
> --- errorlevel 1
>
> When class A is defined in separate source and header files, compiling the source file yields the error message I mentioned in my earlier post.
>
> The compile command I am using is:
>
> dmc -Ic:\dm\stlport\stlport TestInline.cpp
>
> The problem seems to involve combinations of the use of inline, static, and std::pair.  I've seen this before and have work arounds, but that becomes cumbersome when dealing with existing library code (e.g., special versions of library code for use w/ DMC).
>
> Thanks again for any help.
>
>
> Matthew wrote:
> > I have a certain familiarity with DMC++ (and other compiler) ICE bugs.
Why
> > not post the code (or an effective snapshot of it) and I'll see if I
> > recognise anything.
> >
> > Matthew
> >
> > "Gregg" <no@spam.com> wrote in message news:eib632$1pd9$1@digitaldaemon.com...
> >> When compiling a code with DMC I get the following message:
> >>
> >> Internal error: cod2 3692
> >> --- errorlevel 1
> >>
>
>


----------------------------------------------------------------------------
----


> #include <utility>
>
> class A {
>   int getSomething();
> };
>
> inline
> int A::getSomething() {
>   static std::pair<int, int>  b(1, 1);
>   return 0;
> }
>
> int main() {
>   return 0;
> }
>

Matthew wrote:
>  - use a custom template pair instead of std::pair
>  - use a non-template pair instead of std::pair
>  - use a static reference and a static ptr+auto_ptr, e.g.
> 
>     typedef std::pair<int, int>    pair_t;
>     static std::auto_ptr<pair_t>   b_(new pair_t(1, 1));
>     static pair_t                            &b = *b_.get();
> 
>     (Note: this has thread-safety implications.)
> 
>  - move the static into another function / scope, e.g. into a local
> namespace

All good suggestions to be sure.  I've opted for the simplest change to the existing code: comment out the inline statement.  I'm hoping this change stands the least chance of breaking something else.

> All good suggestions to be sure.  I've opted for the simplest change to the existing code: comment out the inline statement.  I'm hoping this change stands the least chance of breaking something else.


:-)

Good luck!