Re: [lmi] sequence input editor -- how to get accepted keywords

[Greg Chicares]

On 2010-06-26 18:41Z, Vaclav Slavik wrote:
> 
> What is the problem with gcc 3.4.5 exactly?

It is
  warning: control reaches end of non-void function
as in the following simple testcase:

enum N {one, two};

int test_main()
{
    volatile N n = one;
    switch(n)
        {
        case one: return 1;
        case two: return 2;
        }
}

Adding 'default:' resolves that, but effectively inhibits '-Wswitch'.

> That it interprets -Wswitch
> differently from what
> http://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html says, namely that
> 
>         -Wswitch
>                 Warn whenever a switch statement has an index of
>                 enumerated type and lacks a case for one or more of the
>                 named codes of that enumeration. (The presence of a
>                 default label prevents this warning.) case labels
>                 outside the enumeration range also provoke warnings when
>                 this option is used (even if there is a default label).
>                 This warning is enabled by -Wall.
> 
> ?

It does what the manual says, AFAICT.

> As far as I can tell, -Wall shouldn't cause any warning if there's a
> 'default:' label, should it?

Correct. To summarize (assuming '-Wswitch'), if 'default:' is present:
  No:  warning about missing switch cases
  Yes: warning about returning without a value
And if it's not present:
  Yes: warning about missing switch cases
  No:  warning about returning without a value
If I add 'default:', then I don't get the (bogus) return-without-value
warning, and I do get the switch warning when an enum case is absent.

> I have trouble imagining any situation when this -- or -Wswitch-default
> for that matter -- would be a good idea. The whole point of using
> 'default:' is to express "all other values should be handled the same".
> Likewise, the whole point of not including it is to express "there's no
> default, this switch should special-case all possibilities". Per the
> documentation, -Wswitch helps with exactly that: it warns about missing
> labels, unless there's an all-catching 'default:' label.

I like that convention. But MinGW gcc-3.4.5 and prior don't let me rely
on it with '-Werror', and that's why I've generally followed a different
convention, striving to put this in every switch:
  default: throw [something];
Really, though, that means I'm trying to do the compiler's job for it
(and, sadly, only at run time) so I don't mind abandoning that practice.

Here's an alternative that detects missing enum cases with '-Wswitch'
and doesn't produce
  warning: control reaches end of non-void function
with the compiler I'm using:

int test_main()
{
    volatile N n = one;
    switch(n)
        {
        case one: return 1;
        case two: return 2;
        }
    throw "Unreachable";
}

What do you think?