https://gcc.gnu.org/bugzilla/show_bug.cgi?id=44949
Manuel López-Ibáñez changed:
What|Removed |Added
Keywords||diagnostic
Status|UNCONFIRMED |NEW
Last reconfirmed||2016-04-03
CC||manu at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #2 from Manuel López-Ibáñez ---
(In reply to Marc Glisse from comment #1)
> Created attachment 24935 [details]
> hack
>
> It is an easy enough hack to call the Wparentheses code for &= and others
> (the warning says "&" instead of "&=", it isn't a clean patch) so I can test
> on real code. And it seems painful. It warns on code like: a |= b & c; where
> it is quite clear we could never intend (a|=b). Restricting the warning
> to the case where the result of the expression is used should improve
> things, except that I don't know how to access that information...
I think the problem cannot be solved like this because assignment operators
have a different precedence than their non-assignment equivalents. That is,
i&=c|b;
i = i & c|b;
are not equivalent. Do you want to warn about the above?
It would probably help you to create a very long testcase listing everything
you wish to warn and not warn about, then start from a new function instead
from warn_about_parentheses. If it turns out that some parts can be shared,
then great. (In any case, the warnings in warn_about_parentheses are very
clumsy and obscure: Clang provides a better explanation about what is wrong and
how to fix it, so starting from scratch seems better).
It may also be worth looking at the point where we give:
case MODIFY_EXPR:
if (!TREE_NO_WARNING (expr)
&& warn_parentheses)
{
warning (OPT_Wparentheses,
"suggest parentheses around assignment used as truth
value");
TREE_NO_WARNING (expr) = 1;
}
break;
since it may be the point to catch most of:
if (i&=2 == 0)
but you may need to revert the hack done here:
if (code == NOP_EXPR)
ret.original_code = MODIFY_EXPR;
else
{
TREE_NO_WARNING (ret.value) = 1;
ret.original_code = ERROR_MARK;
}
and devise a way to keep the info that this was originally more than an
assignment operator (set ret.original_code to BIT_AND_EXPR, etc ?).
