Did anyone else ever have any thoughts on this issue? I ran across
the same issue myself recently in a situation where the insertion of
the unnecessary ops was a moderate inconvenience.
Ben
On Mar 24, 2009, at 1:33 PM, Elnatan Reisner wrote:
> In cabs2cil's doExp function, the case for &&, ||, and ! includes
> comments saying that the results must be normalized to 0 or 1. Then,
> in
> the case where the expression is UnOp(LNot,_,_), there is a comment
> saying that the result is already normalized. Isn't this also true for
> binary operators that evaluate to boolean values? That is, couldn't we
> replace line 3827 in cabs2cil.ml (revision 10610):
> | CEExp (se, (UnOp(LNot, _, _) as e)) ->
> with
> | CEExp (se, ((UnOp(LNot, _, _)|BinOp((Eq|Ne|Le|Lt|Ge|Gt|LAnd|
> LOr),_,_,_)) as e)) ->
>
> Right now, pattern matching falls through to the next case, causing
> some
> superfluous occurrences of '!= 0' to get inserted into code. For
> example, this code
>
> int main() {
> int x = 0;
> return 1 && x == 4;
> }
>
> gets transformed into
>
> int main(void)
> { int x ;
> {
> x = 0;
> return ((x == 4) != 0);
> }
> }
>
> This might not come up so often---it seems that this only happens when
> an && or || has a constant value as an operand, or if you set CIL's
> useLogicalOperators flag to true---but if it is simple to stop
> generating unnecessary extra code, why not do it?
>
> Speaking of useLogicalOperators, here is a second question:
> When useLogicalOperators is set to true, cabs2cil.ml's doCondExp
> function keeps &&s and ||s (rather than transforming them to explicit
> conditionals) if neither operand has side-effects. However, notice
> that
> the *first* operand will be evaluated regardless of whether short-
> circuiting occurs; this means that we only need to check if the
> *second*
> operand has side-effects to see if we can keep the && or ||. This
> means
> we can change lines 4474-4478 in cabs2cil.ml:
>
> | CEExp(se1, e1'), CEExp (se2, e2') when
> - !useLogicalOperators && isEmpty se1 && isEmpty se2 ->
> - CEExp (empty, BinOp(LAnd,
> + !useLogicalOperators && isEmpty se2 ->
> + CEExp (se1, BinOp(LAnd,
> makeCast e1' intType,
> makeCast e2' intType, intType))
>
> and similarly for the LOr case.
>
> Is my reasoning correct? Or is there a reason *not* to keep &&s and
> ||s
> in such cases?
>
------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
CIL-users mailing list
CIL-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/cil-users
