On 06/25/2018 02:00 AM, Bruno Haible wrote:
> Hi Jim, Paul,
>
>> parse-datetime.y: In function 'parse_datetime2':
>> parse-datetime.y:1791:19: error: jump skips variable initialization \
>> [-Werror=jump-misses-init]
>
> First this occurred in af_alg.c with pretty standard Linux kernel style
> with gotos. [1] Now here.
>
> What is the point of this warning?
>
> I've read [2][3].
>
> Test case:
> ===========================================================
> extern void bar (int);
> int foo (int a)
> {
> if (a == 1) goto one;
> int b = 4;
> bar (b);
> one:
> return a;
> }
> ===========================================================
>
> $ gcc-version 8.1.0 -Wall -Wjump-misses-init -c foo.c
> foo.c: In function 'foo':
> foo.c:4:15: warning: jump skips variable initialization [-Wjump-misses-init]
> if (a == 1) goto one;
> ^~~~
> foo.c:7:2: note: label 'one' defined here
> one:
> ^~~
> foo.c:5:7: note: 'b' declared here
> int b = 4;
> ^
>
> In C++ such a warning makes sense, for types that have a destructor,
> because the destructor of 'b' would run even through its constructor
> has not run.
>
> But here
> 1. The variable 'b' is of type 'int', which has a trivial destructor.
> 2. We are in C, no C++, where ISO C types don't have destructors anyway.
> (And AFAICS there are no GCC extensions to C that would provide
> destructors either.)
>
> Is the point to help people reduce the scope of the variables?
> ===========================================================
> extern void bar (int);
> int foo (int a)
> {
> if (a == 1) goto one;
> {
> int b = 4;
> bar (b);
> }
> one:
> return a;
> }
> ===========================================================
>
> Is the point to force people to hoist the variable declarations
> at the top of the function (back to the BSD style of the 1980ies),
> like Paul did in [1]?
> ===========================================================
> extern void bar (int);
> int foo (int a)
> {
> int b;
> if (a == 1) goto one;
> b = 4;
> bar (b);
> one:
> return a;
> }
> ===========================================================
>
> Is the point to help people write code that can be compiled with a C++
> compiler?
> If so, then there's no reason for it to be enabled on gnulib code.
>
> Is the point merely "the standard says that a warning is possible, so let's
> implement it"?
>
> Can someone answer this?
Your (first) example shouldn't trigger the warning. But this should:
===========================================================
extern int bar (int);
int foo (int a)
{
if (a == 1) goto one;
int b = bar (a);
one:
return b;
}
===========================================================
I can only assume that triggering the warning for your example is a side
effect. It might be worth a report.
BTW, I have been effected by a bug like in my example and at that time
opened a request for such a warning flag :-)
Regards, Tim
signature.asc
Description: OpenPGP digital signature
