Re: globbing in the repeat-count word gives "illegal character"

[Daniel Shahaf <d.s@daniel.shahaf.name>]

Mikael Magnusson wrote on Tue, 17 Mar 2020 19:37 +0100:
> On 3/17/20, Daniel Shahaf <d.s@daniel.shahaf.name> wrote:
> > [[[
> > $ Src/zsh -f
> > % cd "$(mktemp -d)"
> > % repeat ? true
> > zsh: bad math expression: illegal character: \M-W
> > %
> > ]]]
> >
> > [[[
> > frame #1: 0x0000000000487eac zsh`matheval(s="\x97") at math.c:1479
> >    1476     x = mathevall(s, MPREC_TOP, &junk);
> >    1477     mtok = xmtok;
> >    1478     if (*junk)  
> > -> 1479         zerr("bad math expression: illegal character: %c", *junk);  
> >    1480     return x;
> >    1481 }
> >    1482
> > (lldb) p *junk
> > (char) $0 = '\x97'
> > ]]]
> >
> > [[[
> > % ag 0x97 | cat
> > Src/zsh.h:191:#define Quest             ((char) 0x97)
> > %
> > ]]]  
> 
> Not limited to globbing characters,
> 
> % repeat { false
> zsh: bad math expression: illegal character: \M-O

And what about this one? —
.
    % repeat 2*2 pwd
    zsh: bad math expression: illegal character: \M-G

What's the expected behaviour on this one?  Should it report a different
error, or perform globbing, or take «2*2» as a valid arithmetic
expression and run 4 iterations of the loop body?

Code-wise, execrepeat() calls singsub() when the "has tokens" flag is set:

   487	int
   487	execrepeat(Estate state, UNUSED(int do_exec))
   488	{
   ⋮
   500	    lastval = 0;
   501	    tmp = ecgetstr(state, EC_DUPTOK, &htok);
   502	    if (htok)
   503		singsub(&tmp);
   504	    count = mathevali(tmp);
   505	    if (errflag)
   506		return 1;

When singsub() returns, there are still tokens in the string (Quest,
Inbrace, or Star, respectively).  Is that expected, or should singsub()
have converted those tokens to their ASCII equivalents?

Is the fix just to call untokenize() after singsub()?  Some singsub()
callers do this, but some don't, and I haven't figured out the rule (if
there is one).  I tried this and it seemed to work, but…

Cheers,

Daniel