[class.union] looks ok

@jwakely, how so? There is [class.union.anon] nested inside.

Oh I misread the directive as only talking about a subclause with a single .1 paragraph. But of course ISO doesn't even number paragraphs, it's talking about subclause numbers. Sorry.

N.B. the problem here is really that we have hanging paragraphs. If the content of [class.union] was in [class.union.xxx] then that would be a sibling of [class.union.anon], and we wouldn't be breaking this rule.

In the ISO world, where hanging paragraphs don't exist, a .1 subclause with no .2 sibling would mean the parent has no text of its own and has an "only child". That's obviously pointless and the content of the only child could be moved up into the empty parent.

With hanging paragraphs the parent is not actually empty, and an only child can make sense.

Editorial meeting: This is duplicative of the hanging paragraphs issue #1730.

cplusplus/papers#533 plans to remove [class.this], making [class.mfct.non-static]'s paragraphs non-hanging.

I'm using this Perl script to find these:

$d = 0;                                                        
@secs = {}; @seclabels = {};
while (<>) {
  if ($_ =~ /\\rSec(\d)/) {
    if ($1 < $d && $secs[$d] == 1) { print "isolated subclause $seclabels[$d] in $seclabels[$d-1]\n"; }
    $d = $1;                       
    $sec = $_;
    chomp $sec;
    $secs[$d] += 1;
    $seclabels[$d] = $sec;
    $secs[$d + 1] = 0;
  }
}