RE: Why Bison prefer reduce to shift in dangling else?

[Hans Aberg]

At 22:13 -0500 2003/02/24, John Madsen wrote:
>> Thank you. I added %nonassoc THEN %nonassoc ELSE to my
>> grammar for the Tiger langauge. Attached. The bison still
>> chooses reduce rather than shift. The first line of the
>> tiger.output reads "Conflict in state 104 between rule 25 and
>> token ELSE resolved as shift." but under state 104, i found
>>
>>      state 104
>>
>>              exp  ->  IF vexp THEN exp .   (rule 25)
>>              exp  ->  IF vexp THEN exp . ELSE exp   (rule 26)
>>
>>              ELSE    shift, and go to state 123
>>
>>              $default        reduce using rule 25 (exp)
>>
>> It is frustrating. Any help will be appreciated.
>>
>> Lunjin Lu
>
>I think you're misinterpreting what Bison is telling you.  It *is*
>choosing shift on ELSE.  The last two lines from state 104 mean that if
>it sees ELSE next, it shifts, otherwise ($default) it reduces.

Right: it looks as a misunderstanding.

If the order of the two %nonassoc is changed, ELSE should become reduce.

But the normal in computer languages is to have it as shift.

>  Also,
>what version of Bison are you using?  When I ran your grammar through
>1.875a, I got no conflicts.

It should not have anything with that to do.

  Hans Aberg