Re: Bug in coerce_template_parms (patch)

[Kriang Lerdsuwanakij <lerdsuwa at scf-fs dot usc dot edu>]

Martin von Loewis wrote:
> 
> Did anybody notice problems in template parameter resolution?
> There seems to be a problem in coerce_template_parms, exposed e.g.
> by ttp26.C. Around line 1896, it checks whether the arglist is
> a vector. If it isn't, it treats it as a list.
> 
> Now, there is another check whether nargs==nparms. In ttp26.C, this is
> not the case because there are default parameters.  So the check
> fails, but the arguments are still passed as a vector. This goes
> unnoticed, as the code retrieves the argument using TREE_VALUE, which
> happens to be the right offset. I'm sure one can construct examples
> where the code breaks, though.
> 

> I found this by putting an assertion into TREE_VALUE that the node
> really is a TREE_LIST. I don't have a test case for this specific
> code. If nobody can reproduce the problem, I can try to find one.  For
> ttp26.C, the problem manifests in the 20th invocation of
> coerce_template_parms.
> 
> Thanks,
> Martin

Thanks for pointing that out.  It's a bug (and my fault :( ).  Below is
a small patch that fixes that.  When arglist is a vector with less
elements than required, it should be handled by inside the
'else if (is_tmpl_parm && i < nargs)' case.  is_tmpl_parm is an
indicator that vector may not satisfy the nargs==nparms condition.

Kriang

===========================

	* pt.c (coerce_template_parms): Fix thinko.

*** pt.c	1998/03/10 10:01:37	1.6
--- pt.c	1998/03/17 04:21:51
*************** coerce_template_parms (parms, arglist, i
*** 1904,1910 ****
  	  tree arg;
  	  tree parm = TREE_VEC_ELT (parms, i);
  
! 	  if (arglist)
  	    {
  	      arg = arglist;
  	      arglist = TREE_CHAIN (arglist);
--- 1904,1910 ----
  	  tree arg;
  	  tree parm = TREE_VEC_ELT (parms, i);
  
! 	  if (arglist && TREE_CODE (arglist) == TREE_LIST)
  	    {
  	      arg = arglist;
  	      arglist = TREE_CHAIN (arglist);