Instead, the generated binary will abort at run-time and give misleading message:


int f() {
return 0;  // gdc Error: AssertError Failure rtn.d(3); dmd Aborted
}

int main(char[][] args) {
f();
return 0;
}

>int f() {
>  // return 0;  // gdc Error: AssertError Failure rtn.d(3); dmd Aborted
---^^

need to comment out line 3, to see the error.

>}
>
>int main(char[][] args) {
>f();
>return 0;
>}

On Fri, 26 Aug 2005 07:18:51 +0000 (UTC), z@gg.com wrote:

> Instead, the generated binary will abort at run-time and give misleading message:
> 
> int f() {
> return 0;  // gdc Error: AssertError Failure rtn.d(3); dmd Aborted
> }
> 
> int main(char[][] args) {
> f();
> return 0;
> }

This is not a bug.

If you define a routine to return a value, but you don't actually return anything, the function will crash at run time. The compiler does not attempt to detect this situation, instead it generates a hidden assert at the end of the function.

-- 
Derek
(skype: derek.j.parnell)
Melbourne, Australia
26/08/2005 6:04:14 PM

Derek Parnell wrote:
> On Fri, 26 Aug 2005 07:18:51 +0000 (UTC), z@gg.com wrote:
> 
>> Instead, the generated binary will abort at run-time and give misleading
>> message:
>>
>> int f() {
[implemented correction]
>> // return 0;  // gdc Error: AssertError Failure rtn.d(3); dmd Aborted
>> }
>>
>> int main(char[][] args) {
>> f();
>> return 0;
>> }
> 
> This is not a bug. 

Nonsense.

http://www.digitalmars.com/d/statement.html#return

"At least one return statement is required if the function specifies a
return type that is not void."

Stewart.

-- 
-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GCS/M d- s:- a->--- UB@ P+ L E@ W++@ N+++ o K- w++@ O? M V? PS- PE- Y? PGP- t- 5? X? R b DI? D G e++>++++ h-- r-- !y
------END GEEK CODE BLOCK------

My e-mail is valid but not my primary mailbox.  Please keep replies on the 'group where everyone may benefit.

On Fri, 26 Aug 2005 12:09:41 +0100, Stewart Gordon wrote:

> Derek Parnell wrote:
>> On Fri, 26 Aug 2005 07:18:51 +0000 (UTC), z@gg.com wrote:
>> 
>>> Instead, the generated binary will abort at run-time and give misleading message:
>>>
>>> int f() {
> [implemented correction]
>>> // return 0;  // gdc Error: AssertError Failure rtn.d(3); dmd Aborted
>>> }
>>>
>>> int main(char[][] args) {
>>> f();
>>> return 0;
>>> }
>> 
>> This is not a bug.
> 
> Nonsense.
> 
> http://www.digitalmars.com/d/statement.html#return
> 
> "At least one return statement is required if the function specifies a return type that is not void."

But 'required' by what? I suspect it is the run-time application that requires it rather than the compiler itself.

-- 
Derek Parnell
Melbourne, Australia
27/08/2005 12:27:36 AM

Derek Parnell wrote:
> On Fri, 26 Aug 2005 12:09:41 +0100, Stewart Gordon wrote:
<snip>
>> http://www.digitalmars.com/d/statement.html#return
>>
>> "At least one return statement is required if the function specifies a
>> return type that is not void."
> 
> But 'required' by what?

The D language.  And hence any correct D compiler.

> I suspect it is the run-time application that
> requires it rather than the compiler itself.

If that's so, then you could equally state that

"The body of all asserts is required to evaluate to non-zero."

By which the average person is likely to interpret what?

Stewart.

-- 
-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GCS/M d- s:- a->--- UB@ P+ L E@ W++@ N+++ o K- w++@ O? M V? PS- PE- Y? PGP- t- 5? X? R b DI? D G e++>++++ h-- r-- !y
------END GEEK CODE BLOCK------

My e-mail is valid but not my primary mailbox.  Please keep replies on the 'group where everyone may benefit.

In article <dengj7$1is0$1@digitaldaemon.com>, Stewart Gordon says...
>
>Derek Parnell wrote:
>> On Fri, 26 Aug 2005 12:09:41 +0100, Stewart Gordon wrote:
><snip>
>>> http://www.digitalmars.com/d/statement.html#return
>>>
>>> "At least one return statement is required if the function specifies a return type that is not void."
>> 
>> But 'required' by what?
>
>The D language.  And hence any correct D compiler.
>
>> I suspect it is the run-time application that
>> requires it rather than the compiler itself.
>
>If that's so, then you could equally state that
>
>"The body of all asserts is required to evaluate to non-zero."
>

As long as your not talking about the compiler checking every possible path for a return statement (something that probably no compiler has ever gotten 100% right anyhow), then I would have to agree that the compiler should at least check for a return statement inside the body of a function.

Take a look in the archives - IMO, Walter has good reasoning for the compiler not forcing a return for all paths. But checking that a function actually has a valid return statement if it's expected to return something should be part of the reference compiler because 1) it will save programmers a lot of "oopses", 2) it's gotta be readily do-able to verify 100% of the time and 3) the spec. above really is unequivocal on the matter.

- Dave


>By which the average person is likely to interpret what?
>
>Stewart.
>
>-- 
>-----BEGIN GEEK CODE BLOCK-----
>Version: 3.1
>GCS/M d- s:- a->--- UB@ P+ L E@ W++@ N+++ o K- w++@ O? M V? PS- PE- Y?
>PGP- t- 5? X? R b DI? D G e++>++++ h-- r-- !y
>------END GEEK CODE BLOCK------
>
>My e-mail is valid but not my primary mailbox.  Please keep replies on the 'group where everyone may benefit.

>> return 0;  // gdc Error: AssertError Failure rtn.d(3); dmd Aborted

>The compiler does not
>attempt to detect this situation, instead it generates a hidden assert at
>the end of the function.

*EVEN* in that case, a more meaningfull message should be generated, e.g.

"NoReturnValueAssertError"

Which is much more human readable.

z@gg.com wrote:
>>>return 0;  // gdc Error: AssertError Failure rtn.d(3); dmd Aborted

> *EVEN* in that case, a more meaningfull message should be generated, e.g.
> 
> "NoReturnValueAssertError"
> 
> Which is much more human readable.
> 
> 
> 

When you compile with -w the compiler will let you know of missing return statements:

warning - <filename>(<line #>): function <function name> no return at end of function.

Walter added this as a concession after much debate over the default behaviour. So if you want to catch these cases at compile time, use -w and be happy.