[cfe-commits] [PATCH] Implicit fall-through between switch labels

David Blaikie dblaikie at gmail.com
Thu May 3 12:03:50 PDT 2012

On Thu, May 3, 2012 at 11:35 AM, Richard Smith <richard at metafoo.co.uk> wrote:
> Hi Alex,
> Thanks for your persistence on this! I've committed your patch as r156086.
> I've also committed two of your three fallthrough fixes for clang in
> r156082. The third one looked correct, and already had a "// Fall through"
> comment, so I decided to leave it alone.
> In testing your patch, I did find one issue. We produce a
> 'clang::fallthrough attribute is unreachable' diagnostic for this code:
>   switch (2) {
>     case 1:
>       f();
>       [[clang::fallthrough]];
>     case 2:
>       break;
>   }
> I think we should suppress the diagnostic in this case.

This is the edge of a large bunch of issues with the current
unreachable code calculation that you might want to just generally
steer clear of (by not warning about unreachable fallthroughs that
would fall through to a case), if possible, until it's improved.

This like:

if (sizeof(int) == 4) { stuff(); [[clang::fallthrough]]; }

will currently be classified as unreachable code, as well as
expressions using non-type template parameters (or anything much that
is evaluated at compile-time - macro expansions, etc).

While the particular case Richard has brought up here probably needs a
special case even if we improve/fix the general unreachable code
problems (continue to assume the condition expression is
non-constant/variable for the purposes of this warning in particular)
I'm not sure whether that'll be sufficient to keep the noise down on
this warning if you're flagging anything Clang considers unreachable -
but it might be enough given how narrow these cases are (unreachable
code based on a compile-time constant condition within a switch's case
that has a fallthrough attribute).

- David

> On Thu, May 3, 2012 at 5:02 AM, Alexander Kornienko <alexfh at google.com>
> wrote:
>> Hi Richard,
>> Following your privately posted comments, I've made some changes in
>> punctuation for diagnostics, added documentation to
>> docs/LanguageExtensions.html and rebuilt patch against current HEAD. Here it
>> is.
>> I've also attached a small patch which removes unnecessary fall-throughs
>> in 3 widely-used header files. Though it is a separate patch, it's
>> definitely related to the main one.
>> Thanks for your review!
>> On Wed, May 2, 2012 at 12:42 AM, Alexander Kornienko <alexfh at google.com>
>> wrote:
>>> Hi Richard,
>>> Here's an updated patch.
>>> On Tue, May 1, 2012 at 5:28 AM, Richard Smith <richard at metafoo.co.uk>
>>> wrote:
>>>> Hi Alex,
>>>> The 'unreachable block' test isn't quite right; a block can have
>>>> (unreachable) predecessors and still be unreachable:
>>>> switch (x) {
>>>> // oops, deleted "case 0:"
>>>>   if (a)
>>>>     f();
>>>>   [[clang::fallthrough]]; // no warning here!
>>>> case 1:
>>>>   break;
>>>> }
>>>> That said, since -Wunreachable-code warns on (most) such cases, I don't
>>>> think that is necessarily a problem. It might be worth updating the comment
>>>> in the code to indicate that we only catch trivial cases, though.
>>> I wrote in one of previous letters that this detects only trivially
>>> unreachable blocks. Now I added comment in the code. I think that this
>>> specific diagnostic can be helpful if -Wunreachable-code isn't in use.
>>>> Style nits: Please update the capitalization of FallthroughMapper's
>>>> members and local variables to match the coding standard. Functions should
>>>> start with a lowercase letter, and variables and data members should start
>>>> with a capital letter. Also, please start comments with a capital letter,
>>>> and end them with a full stop when they are full sentences.
>>> Done. The only question is: is naming style enforced somehow?
>>> RecursiveASTVisitor.h, for example, uses different conventions.
>>>> On Mon, Apr 30, 2012 at 5:40 AM, Alexander Kornienko <alexfh at google.com>
>>>> wrote:
>>>>> On Fri, Apr 27, 2012 at 3:47 AM, Jordy Rose <jediknil at belkadan.com>
>>>>> wrote:
>>>>>> >> FWIW, Richard convinced me that this must be spelled
>>>>>> >> 'clang::fallthrough' for the immediate future. We should probably focus on
>>>>>> >> that form.
>>>>>> >
>>>>>> > Done. It turned out though, that namespaces for attributes were not
>>>>>> > used until now, so I implemented my own vision: I just add namespace  as a
>>>>>> > prefix to spelling with triple underscore delimiter. Better ideas?
>>>>>> The right thing might be to implement real namespacing now, but why
>>>>>> not just use ::, as in the written form? The parser only expects attributes
>>>>>> to be made of identifiers, but our table can use any characters we want.
>>>>> These strings are used as parts of C++ identifiers in clang code, so we
>>>>> can't use arbitrary characters inside. I found underscores to be the most
>>>>> appropriate option here. Regarding correct implementation of namespacing in
>>>>> attributes, I strongly believe it's an independent piece of functionality,
>>>>> which requires a separate design discussion. So I would leave my temporary
>>>>> implementation (which is quite flexible to be used at the moment).
>>>> Since you're planning on working on fixing this next, I think the
>>>> temporary implementation is OK. Please add a FIXME next to it, though.
>>> Added.
>>>>>> > +    typedef std::vector<const Stmt*> StmtVec;
>>>>>> > +    bool GetFallThroughSourceStmts(const CFGBlock &B, StmtVec
>>>>>> > &Unannotated,
>>>>>> > +                                   StmtVec &Annotated) {
>>>>>> Should this be a SmallVector? The argument would then be a
>>>>>> SmallVectorImpl<const Stmt*>, and only the actual stack-allocated vector
>>>>>> needs to have a size.
>>>>>> Also, do we need the annotated set? If not, we could leave it out for
>>>>>> now, and only put it back in if we want to try for some intelligence in the
>>>>>> suggestions.
>>>>> I changed this to return just counts of annotated and unannotated
>>>>> fall-through source locations
>>>> I notice you don't use UnannotatedCnt outside GetFallThroughSourceCount.
>>>> I suggest you drop that argument, and replace UnannotatedCnt with a bool
>>>> inside the function. AnnotatedCnt could likewise be replaced by a bool
>>>> FoundAnyFallthroughAttributes or similar.
>>> Removed UnannotatedCnt from parameters, but left AnnotatedCnt as int.
>>> Don't think this will hurt performance and code readability, but this code
>>> can be easier changed to provide more detailed diagnostics.
>>>> I'd also suggest renaming that function, perhaps to
>>>> CheckFallThroughIntoBlock.
>>> You mean checkFallThroughIntoBlock? ;) Done.
>>>>>> > +      SourceLocation L = Label->getLocStart();
>>>>>> > +      S.Diag(L, diag::note_insert_fallthrough_fixit) <<
>>>>>> > +        FixItHint::CreateInsertion(L, "[[clang::fallthrough]];");
>>>>>> This will give a proper diagnostic, but not a very nice placement of
>>>>>> the attribute. Clients like Xcode actually apply fixit hints to the user's
>>>>>> code, and really we want this on a previous line. Can we figure out how to
>>>>>> insert this in the right place, rather than suggest something the user will
>>>>>> have to manually edit afterwards?
>>>>> Proper formatting should be a function of an IDE, not a compiler. If
>>>>> Xcode uses clang's fix-it hints for changing code, it could also deal with
>>>>> white-space handling.
>>>>> I've inserted a line-break at the end of fix-it hint, but I'm not
>>>>> really sure if even this is a duty of the compiler. And AFAIK there are no
>>>>> other examples of fix-it hints, which insert new statements or deal with
>>>>> code formatting in any other way.
>>>> We do have existing fixits which insert spaces and newlines to keep the
>>>> code tidy, in cases where "tidy" is obvious, but I agree that it can't be
>>>> the responsibility of the fixit to reformat the code according to whatever
>>>> coding style is relevant. This fixit seems like a bit of a grey area, since
>>>> "keep the case label at the same indentation, and put the attribute at the
>>>> same indentation as the previous line" is likely to cover almost all cases
>>>> perfectly, but I think what you've got now is a reasonable tradeoff. And I
>>>> don't want for us to need to worry about what happens if 'case' or 'default'
>>>> isn't the first token on the line :-)
>>> I even had to remove '\n' from my fix-it hints, 'cause \n's caused some
>>> assertion fails inside diagnostic rendering. I suggest start using this
>>> diagnostic and then figure out what fix-it spelling is the best here.
>>>> Other than the superficial issues above, I think this is looking good.
>>>> Thanks!
>>>> Richard
>> --
>> Best regards,
>> Alexander Kornienko
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits

More information about the cfe-commits mailing list