Path: csiph.com!weretis.net!feeder8.news.weretis.net!eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: Tim Rentsch
Newsgroups: comp.lang.c
Subject: Re: Corrupted C99 _Bool?
Date: Thu, 20 Jul 2023 22:17:51 -0700
Organization: A noiseless patient Spider
Lines: 90
Message-ID: <86o7k5vo5c.fsf@linuxsc.com>
References: <2569e26b-c60a-4873-a759-ee00a50e2bf9n@googlegroups.com> <875y6oq10u.fsf@nosuchdomain.example.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Injection-Info: dont-email.me; posting-host="2d78163bfb891e7187c96a580a5a5784"; logging-data="3290646"; mail-complaints-to="abuse@eternal-september.org"; posting-account="U2FsdGVkX18sfTfjGiFlT2gLz9BiYLkYmXCh7J0QbV8="
User-Agent: Gnus/5.11 (Gnus v5.11) Emacs/22.4 (gnu/linux)
Cancel-Lock: sha1:MijxG0jB4I/8InBgd7yAN+alGMc= sha1:pyI7KiRnCeJNtF/tU36fKwTqzlA=
Xref: csiph.com comp.lang.c:171011
Keith Thompson writes:
> Opus writes:
>
>> On 12/07/2023 02:32, das...@gmail.com wrote:
>>
>>> C99 indicates that a _Bool is an unsigned integer type. When a
>>> variable of a type other than _Bool is assigned to a _Bool, C99
>>> specifies that the _Bool receives the value of 0 if the variable was
>>> 0, or 1 otherwise.
>>> But what does C99 say about a _Bool that is corrupted to a value
>>> other than 0 or 1 (perhaps through a pointer error in another part
>>> of the program)? Will this corrupted _Bool still function
>>> correctly?
>>
>> It's undefined behavior. Just like if you corrupt any value of any
>> other type.
>>
>>> For example:
>>> _Bool a = true;
>>> _Bool b; /* Assume b is corrupted to "2" somehow */
>>> if (a && b)
>>> {
>>> /* Would the compiler be able to use a bitwise */
>>> /* AND above, because it assumes all _Bools */
>>> /* must be 0 or 1, so that if (1 && 2) becomes */
>>> /* if (1 & 2) and evaluates false? */
>>> }
>>> Thanks for any clarification.
>>
>> If the memory holding a _Bool value gets corrupted, there is obviously
>> no way for the compiler to automatically correct it. How would it?
>>
>> It's UB, but the practical result will of course depend on the
>> target. A _Bool is held by an integer under the hood in practice.
>>
>> Compilers do force a value of 1 for any value other than 0 assigned to
>> it *during assignment*, and 0 otherwise. But again, only when an
>> assignment occurs (or a cast).
>
> More precisely, *conversion* from any scalar type to _Bool yields a
> value of 0 if the converted value is 0, 1 otherwise. Assignment
> performs an implicit conversion; so do argument passing, initialization,
> and a return statement. And of course a cast performs an explicit
> conversion.
>
>> For instance, the expression '(int)(_Bool) 5' yields 1.
>>
>> But if you alias a _Bool, say:
>>
>> _Bool b;
>>
>> *(int *) &b = 5;
>>
>> While UB, in practice in most implementations, b will be represented
>> internally as an int and here will hold the value 5. Not 1.
>
> _Bool is most likely smaller than int, so the assignment is likely to
> clobber bytes not belonging to the _Bool object.
>
>> This would cause a problem if you expect the value to be either 0 or 1
>> exclusively. For instance if you cast it back to an int and do some
>> artihmetic or bitwise operations with that.
>>
>> Regarding what would happen to '(a && b)' in your example, the '&&'
>> operator is NOT a bitwise operator. Even when dealing strictly with
>> _Bool's on each side of the operator, it can't be implemented with a
>> bitwise operation when compiled. Quoting the standard regarding the
>> '&&' operator:
>>
>> "The && operator shall yield 1 if both of its operands compare unequal
>> to 0; otherwise, it yields 0. The result has type int.
>> Unlike the bitwise binary & operator, the && operator guarantees left
>> to-right evaluation;
>> if the second operand is evaluated, there is a sequence point between
>> the evaluations of the first and second operands. If the first operand
>> compares equal to 0, the second operand is not evaluated."
>
> Yes, but if a and b are both of type _Bool, the compiler can
> assume that they both have a value of 0 or 1. [...]
Whether a compiler can safely make such an assumption depends on
implementation choices pertaining to how _Bool objects are
represented.
> A compiler *can* implement `a && b` as `a & b` if both are of type
> _Bool and neither is volatile.
Again, such a conclusion might or might not be valid, depending on
other aspects of the implementation.