Re: [Tinycc-devel] tcc 0.9.23 shift generation bug

[Zdenek Pavlas]

Fabrice Bellard wrote:
> Zdenek Pavlas wrote:
>
> > Signed shifts of signed integers should not give unsigned results.
> > This fix is probably a better one.
> >
> > --- tcc.c.orig  2005-06-18 00:09:15.000000000 +0200
> > +++ tcc.c       2005-06-28 12:17:44.472171840 +0200
> > @@ -5628,7 +5628,8 @@
> >          t = VT_INT;
> >          /* convert to unsigned if it does not fit in an integer */
> >          if ((t1 & (VT_BTYPE | VT_UNSIGNED)) == (VT_INT | VT_UNSIGNED) ||
> > -            (t2 & (VT_BTYPE | VT_UNSIGNED)) == (VT_INT | VT_UNSIGNED))
> > +            (t2 & (VT_BTYPE | VT_UNSIGNED)) == (VT_INT | VT_UNSIGNED) &&
> > +            (op != TOK_SAR))
> >              t |= VT_UNSIGNED;
> >      std_op:
> >          /* XXX: currently, some unsigned operations are explicit, so
>
> OK for the bug, but this fix is still incorrect : the shift operations 
> should be handled separately because they keep the type of the first 
> operand.

Isn't this enough?

--- tcc.c.orig  2005-06-18 00:09:15.000000000 +0200
+++ tcc.c       2005-06-28 12:17:44.472171840 +0200
@@ -5628,7 +5628,8 @@
         t = VT_INT;
         /* convert to unsigned if it does not fit in an integer */
         if ((t1 & (VT_BTYPE | VT_UNSIGNED)) == (VT_INT | VT_UNSIGNED) ||
-            (t2 & (VT_BTYPE | VT_UNSIGNED)) == (VT_INT | VT_UNSIGNED))
+            (t2 & (VT_BTYPE | VT_UNSIGNED)) == (VT_INT | VT_UNSIGNED) &&
+            (op != TOK_SAR) && (op != TOK_SHL))
             t |= VT_UNSIGNED;
     std_op:
         /* XXX: currently, some unsigned operations are explicit, so

--
Zdenek Pavlas