linux-kernel - Re: [PATCH] block/blk-iocost (gcc13): cast enum members to int in prints

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <d833ad15-f458-d43d-cab7-de62ff54a939@kernel.org>
Date:   Tue, 1 Nov 2022 06:46:56 +0100
From:   Jiri Slaby <jirislaby@...nel.org>
To:     Tejun Heo <tj@...nel.org>, Christoph Hellwig <hch@...radead.org>
Cc:     linux-kernel@...r.kernel.org, Martin Liska <mliska@...e.cz>,
        Josef Bacik <josef@...icpanda.com>,
        Jens Axboe <axboe@...nel.dk>, cgroups@...r.kernel.org,
        linux-block@...r.kernel.org
Subject: Re: [PATCH] block/blk-iocost (gcc13): cast enum members to int in
 prints

On 31. 10. 22, 18:57, Tejun Heo wrote:
> On Mon, Oct 31, 2022 at 05:24:28AM -0700, Christoph Hellwig wrote:
>> On Mon, Oct 31, 2022 at 12:45:20PM +0100, Jiri Slaby (SUSE) wrote:
>>> Cast the enum members to int when printing them.
>>>
>>> Alternatively, we can cast them to ulong (to silence gcc < 12) and use %lu.
>>> Alternatively, we can move VTIME_PER_SEC away from the enum.
>>
>> Yes, either split the enum or just use a define.  But casts are a big
>> code smell and should be avoided if there is a reasonable alternative.
> 
> enums are so much better for debugging and other instrumentation stuff. The
> only requirement for the enum types is that they're big enough to express
> all the members and we can use whatever printf format letter which matches
> the type in use. The problem here is that the compiler behavior is different
> depending on the compiler version, which kinda sucks.

Yes. The real problem is that using anything else then an INT_MIN <= x 
<= INT_MAX _constant_ in an enum is undefined in ANSI C < 2x (in 
particular, 1 << x is undefined too). gcc manual defines unsigned int on 
the top of that as defined too (so this holds for our -std=g*).

> I suppose the most reasonable thing to do here is just splitting them into
> separate enum definitions. Does anyone know how this behavior change came to
> be?

C2x which introduces un/signed long enums. See the bug I linked in the 
commit log:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36113

The change is also turned on in < C2x on purpose. AIUI, unless there is 
too much breakage. So we'd need to sort it out in (rather distant) 
future anyway (when we come up to -std=g2x).

> Do we know whether clang is gonna be changed the same way?

In C2x, Likely. In < C2x, dunno what'd be the default.

thanks,
-- 
js
suse labs