[PATCHv2] implement -Wold-style-cast

[Alp Toker]

On 26/11/2013 06:29, Sean Silva wrote:
>
>
>
> On Tue, Nov 26, 2013 at 12:37 AM, Alp Toker <alp at nuanti.com 
> <mailto:alp at nuanti.com>> wrote:
>
>
>     On 22/11/2013 15:54, Ondřej Hošek wrote:
>
>         I've been trying to implement the relevant fix-it, but apart
>         from not
>         being able to find the end of the expression being cast (to
>         insert the
>         closing paren), the pathological case
>
>         On Fri, Nov 22, 2013 at 2:13 AM, Dmitri Gribenko
>         <gribozavr at gmail.com <mailto:gribozavr at gmail.com>> wrote:
>
>             (for example, static cast of a const cast)
>
>         might pose even more of a challenge, because I have to find
>         and print
>         the "halfway" type:
>
>         int i = 42;
>         const int *cpi = &i;
>         float *fi = (float *)cpi;
>         // -> reinterpret_cast<float *>(const_cast<int *>(cpi));
>
>         This already spells trouble with regard to code formatting, and it
>         gets worse if I have a multi-level pointer type where some of the
>         levels are const and some aren't.
>
>
>     Hi Ondřej,
>
>     I suspect the complexity of converting complex / multi-level
>     C-style casts to C++-style casts just isn't worth the effort.
>
>
> Agreed. This almost borders on a stylistic choice better-suited for 
> clang-tidy.

Sean, you hit the nail on the head and I've reconsidered my suggestions.

The conversion would definitely benefit from the kind of rewrite 
facilities that already exist in clang-tidy / clang-modernize so it 
seems better to do there. This would solve difficulties encountered with 
matching parentheses too. So that's an open task if someone wants to 
work on it.

In light of this, and considering also that we don't usually make an 
effort to produce FixIts for warning-level diagnostics in Sema, the 
original patch to the parser by Ondřej looks fine to me. Dmitri, is that 
OK with you?

Alp.


>
> -- Sean Silva
>
>
>     Even if you get it to work, the results wouldn't be pretty. It's
>     better to encourage the developer to find an idiomatic solution by
>     restructuring their code. Thanks for investigating this though.
>
>     I'd say it's fine just to provide FixIt replacements for the
>     common cases like straight const_cast<> that are easy to determine
>     using existing CastOperation machinery.
>
>
>
>         (Also, I have to move the warning from Parser to Sema because only
>         then do I have the type information to generate a fix-it, so I
>         don't
>         know at all how well it will work with templates.)
>
>
>     It's also fine to work on this incrementally if you could prepare
>     a patch similar to the original one but moving the check to the
>     right place in Sema, along with a TODO about the FixIt. (It
>     shouldn't be added to Parse because that's a clear dead-end.)
>
>     I don't know what user expectations are for -Wold-style-cast, but
>     it might be worth silently permitting the idiomatic C-style
>     CK_ToVoid used to suppress unused variable warnings as a special
>     case. Do you have an opinion on that?
>
>     Alp.
>
>
>
>
>         Cheers,
>         ~~ Ondra
>         _______________________________________________
>         cfe-commits mailing list
>         cfe-commits at cs.uiuc.edu <mailto:cfe-commits at cs.uiuc.edu>
>         http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
>
>     -- 
>     http://www.nuanti.com
>     the browser experts
>
>
>     _______________________________________________
>     cfe-commits mailing list
>     cfe-commits at cs.uiuc.edu <mailto:cfe-commits at cs.uiuc.edu>
>     http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
>

-- 
http://www.nuanti.com
the browser experts