Fix: Assignment to const results in a "no overload" error for operator=

[rljacobson]

On some platforms, an assignment to a const mpz_class results in a "no overload" error for operator= (assignment operator). The error is triggered when it is not true that SIZEOF_UNSIGNED_LONG == 8, e.g. on 32-bit architectures.

The context is small enough to reproduce in its entirety here:

DagNode*
SuccSymbol::makeNatDag64(uint64_t nat)
{
  DagNode* zero = zeroTerm.getDag();
#if SIZEOF_UNSIGNED_LONG == 8
  return (nat == 0) ? zero : (new S_DagNode(this, static_cast<unsigned long>(nat), zero));
#else
  //
  //	Hack for 32-bit machines.
  //
  unsigned long low = nat & 0xFFFFFFFF;
  unsigned long high = nat >> 32;
  const mpz_class number(high);
  number = (number << 32) + low;
  return (nat == 0) ? zero : (new S_DagNode(this, number, zero));
#endif
}

It looks like just a copy+paste error, because number doesn't need to be const here. Excruciating details are given in the next paragraph, which you're welcome to skip over.

The issue is in the #else preprocessor branch, where it is assumed that unsigned long is 32 bits. The "hack" is to load the value of the 64-bit parameter nat into an mpz_class number 32 bits at a time by first loading the high 32 bits into the lower 32 bits of number, shifting number left by 32, and then loading the low 32 bits of nat into the now zeroed lower 32 bits of number. These steps require at least two assignments to number, which is prohibited by the const.

The fix is to remove the const.

[rljacobson]

Out of curiosity, are there many people running Maude on 32-bit platforms?

[stevenmeker]

I've made your suggested change in Alpha152. I'm reluctant to accept pull requests on this repository in case it breaks my script for updating it from a private development repository.

That last bug report I have with respect to i386 was on 11/28/20 by a Debian maintainer rather than a user. People have built Maude on a Raspberry Pi but that is hardly recommended for serious use.