Re: [PING] Walways-true:Warn if a comparison is always true or always false due to the limited range of the data type

["Manuel López-Ibáñez" <lopezibanez at gmail dot com>]

On 04 Apr 2007 09:49:05 -0500, Gabriel Dos Reis <gdr@cs.tamu.edu> wrote:
> "Manuel López-Ibáñez" <lopezibanez@gmail.com> writes:
>
> | On 22/03/07, Dirk Mueller <dmueller@suse.de> wrote:
> | > On Thursday, 22. March 2007, Manuel López-Ibáñez wrote:
> | >
> | > > I think that after 18 days, a ping is due.
> | > > PING: http://gcc.gnu.org/ml/gcc-patches/2007-03/msg00207.html
> | >
> | > Again my usual comment: I think -Walways-true is a bad name for a warning that
> | > warns about expressions being always false.
> | >
> | > I think -Wlogical or -Wlogical-invariant or similar would be a better name.
> | >
> |
> | I agree that it is a bad name but not because of the reason you point.
> | I think it is bad because this warning should not try to warn about
> | anything else than conditions that are always true (or false) because
> | of the limited
> | range of the data type. What you are trying to detect is whether you
> | are using the wrong type in a comparison (or perhaps the wrong
> | variable). In that context, Walways-true (and Wlogical) are too vague
> | names but I prefer the former since Wlogical is even vaguer.
> |
> | Alternative names:
>
>   -Wtautology?
>   -Wunsigned-rel?

I think I forgot to put the emphasis in  the proper place: conditions
that are always true (or false) *because of the limited range of the
data type*.

-Wtautology
You may have a tautology in a logical expression, like if (1 == 1) or
something like if (!x || x), but this warning should not handle those.

-Wunsigned-rel
Again, this is not neccesarily about just unsigned. I think shorts and
enums are handled by the warning as well.

Don't you like Wtype-bounds or Wtype-range? I am feeling already
biased. I think it would be great to be able to get some feedback from
users. Like:

How do you think that the option that produces this message:
"condition is always true because of the limited range of the data type"
is called?

[  ] Wtautology
[  ] Wtype-range
[  ] Wtype-bounds
[  ] Wunsigned-rel
[  ] Wlogical-op
Other: [                     ]

Perhaps we could send this to gcc-help or some user list...

> Please, send me links to other patches I've not formally reviewed --
> use my address at integrable-solutions.net; they have again messed up
> mails here. :-(

http://gcc.gnu.org/ml/gcc-patches/2007-03/msg01435.html
http://gcc.gnu.org/ml/gcc-patches/2007-02/msg01102.html

And if you had some time, it would be very helpful to get some
feedback from you about the thread here
http://gcc.gnu.org/ml/gcc-patches/2007-03/msg00171.html. The patch is
useful for the C front-end as it is and I think it is useful as well
for the C++ front-end despite it doesn't warn for anonymous
namespaces. But I don't think I can solve that since it is an
orthogonal issue. See the thread for more info.

Thanks,

Manuel.