On Thu, Jun 18, 2015 at 12:57 PM, Patrick Palka <[email protected]> wrote:
> On Thu, Jun 18, 2015 at 12:31 PM, David Malcolm <[email protected]> wrote:
>> On Tue, 2015-06-09 at 13:31 -0400, Patrick Palka wrote:
>>> This patch improves the heuristics of the warning in a number of ways.
>>> The improvements are hopefully adequately documented in the code
>>> comments.
>>>
>>> The additions to the test case also highlight the improvements.
>>>
>>> I tested an earlier version of this patch on more than a dozen C code
>>> bases. I only found one class of bogus warnings yet emitted, in the
>>> libpng and bdwgc projects. These projects have a coding style which
>>> indents code inside #ifdefs as if this code was guarded by an if(), e.g.
>>>
>>> if (foo != 0)
>>> x = 10;
>>> else // GUARD
>>> y = 100; // BODY
>>>
>>> #ifdef BAR
>>> blah (); // NEXT
>>> #endif
>>
>> We have detect_preprocessor_logic which suppresses warnings when there's
>> preprocessor logic between BODY_EXPLOC and NEXT_STMT_EXPLOC, for cases
>> like this:
>>
>> if (flagA)
>> foo ();
>> ^ BODY_EXPLOC
>> #if SOME_CONDITION_THAT_DOES_NOT_HOLD
>> if (flagB)
>> #endif
>> bar ();
>> ^ NEXT_STMT_EXPLOC
>>
>> This currently requires that it be multiline preprocessor logic: there
>> must be 3 or more lines between BODY_EXPLOC and NEXT_STMT_EXPLOC for
>> this rejection heuristic to fire.
>
> Oh I now see why it requires 3 or more lines: one line each for the
> #if, #endif and for the
>
>>
>> Perhaps we could tweak or reuse that heuristic, perhaps if there's an
>> entirely blank (or all whitespace) line separating them (given that this
>> warning is all about the "look" of the code).
>
> That makes sense. What about just checking in
> detect_preprocessor_logic() if there is > 1 line (instead of >= 3
> lines) between the body and the next statement? When that's the case,
> then whatever is in between the start of the body must either be more
> of the body (if it's a multi-line compound statement) or whitespace.
> In either case we should not warn if the next statement is aligned
> with the body.
Actually, the body cannot be a compound statement because (with this
patch) we have already bailed out in that case. So if there is > 1
line between the body and the next statement then there must be
whitespace... unless there are just more #ifdefs, e.g.
if (foo)
bar ();
#ifdef A
#ifdef B
baz ();
#endif
#endif
We would want to warn here even though there are 2 lines, not 1 line,
in between the body and the next statement. So we still have to
directly check for a whitespace line, I think.
> Yet we will still rightfully warn on the following
> code:
>
> if (foo) // GUARD
> bar (); // BODY
> #ifdef BAZ
> baz (); // NEXT
> #endif
>
> because there is just one line between the body and the next
> statement. The user can add a line between the body and the next
> statement to suppress the warning if it's bogus.
I meant to say, a (empty) line between the body and the #ifdef.
>
>>
>>> These bogus warnings are pre-existing, however (i.e. not caused by this
>>> patch).
>>
>> (nods) Fixing the false positives from libpng/bdwgc sounds like a
>> separate issue and thus a separate patch then.
>>
>> [...snip...]
>>
>>
