Re: template expansion bug or correct behaviour?

[Dave MacDonald <davidm at idirect dot com>]

I guess I wasn't clear in a previous message, so I'll start again...

I've tried compiling the PRCS configuration mgmt package using
egcs-971201, but it produces an error that I've been able to reproduce
with the following program (the #define SHOWERROR distinguishes between
a compilable and non-compilable version):

template <class Key, class Data>
class HashTable {
public:
#ifdef SHOWERROR
  HashTable(int (*func0)(const Key&, int),
            bool (*equal0)(const Key&, const Key&)){};
#else
  HashTable(int (*func0)(Key&, int),
            bool (*equal0)(Key&, Key&)){};
#endif
  ~HashTable(){};
};

int pathname_hash(const char*&, int)
{
    return 0;
}

bool pathname_equal(const char*&, const char*&)
{
  return true;
}

main(int, char**)
{
    HashTable<const char*, int> descriptors(pathname_hash,
pathname_equal);
}

For example:

~/tst# c++ --version
egcs-2.90.20 971201 (gcc2-970802 experimental)
~/tst# c++ -DSHOWERROR projdesc.cc
projdesc.cc: In function `int main(int, char **)':
projdesc.cc:26: no matching function for call to `HashTable<const char
*,int>::HashTable (int ()(const char *&, int), bool ()(const char *&,
const char *&))'
projdesc.cc:26: candidates are: HashTable<const char
*,int>::HashTable(const HashTable<const char *,int> &)
projdesc.cc:6:                 HashTable<const char
*,int>::HashTable(int (*)(const char *const &, int), bool (*)(const char
*const &, const char *const &))

However, if I compile without the preprocessor symbol SHOWERROR, (e.g.
"c++ projdesc.cc"), it compiles without complaint. That is, if I remove
the "const" modifiers from the paramaterized types in the constructor
definition, everything is OK. 

I'm running Linux 2.0.28 with gnulibc1 on a Pentium-133.

I've compiled PRCS on Solaris, HP-UX and AIX using the native platform
C++ compilers, and I get no complaint where egcs chokes. I've also used
gcc 2.7.2.2 to successfully compile PRCS.

My question relates to how egcs expanded the template instantiation

HashTable<const char *, int>

The error message indicates that egcs expanded the class's constructor
from

  HashTable(int (*func0)(const Key&, int),
         bool (*equal0)(const Key&, const Key&)){};
into

  HashTable(int (*func0)(const char * const&, int),
         bool (*equal0)(const char * const &, const char * const&)){};

A simplistic expansion of the template would give

  HashTable(int (*func0)(const const char *, int),
         bool (*equal0)(const const char *&, const const char *&)){};

which looks wrong, but I don't know how the standard says this should be
expanded.

Other compilers and previous gcc's either expand it differently or
recognize "const char *&" as being the same as "const char * const&". So
the real question is:

Is it a bug in how egcs expanded the template instantiation, or is PRCS
depending upon an old/incorrect implementation of C++?

Sorry this is so long, but I wanted to make sure I got my point across
this time.

-Dave.