Compile under gcc fails, but VS2010 works

I can't figure out why the following code fails to compile under gcc, but works fine when plugged using Microsoft's compiler (using VS2010).

Code:

---

    TimeCalc time1;
 
    TimeCalc time2("16:28");
    TimeCalc time3("00:16");
 
    TimeCalc time4 = time3;
    TimeCalc time5 = "14:34";

---

Under GCC, when it gets to the time5 line, the compile fails with the following error:

Code:

---

... Lab0.cpp||In function 'int main()':|
... Lab0.cpp|250|error: no matching function for call to 'TimeCalc::TimeCalc(TimeCalc)'|
... Lab0.cpp|201|note: candidates are: TimeCalc::TimeCalc(TimeCalc&)|
... Lab0.cpp|186|note:                TimeCalc::TimeCalc(const char*)|
... Lab0.cpp|112|note:                TimeCalc::TimeCalc()|
||=== Build finished: 1 errors, 0 warnings (0 minutes, 0 seconds) ===|

---

The functions it's suggesting are:

line 112: the default constructor
line 186: the constructor taking a string as input
line 201: the copy constructor

The program compiles and runs fine under VS2010. When I use the debugger and step through the program, when it gets to time5, it uses the same constructor time2 and time3 use ( TimeCalc(const char* timeIn) ). I don't see why it would be failing under GCC, because it seems like a simple enough statement. I thought time2/time3 and time5 were equivalent ways to construct something?

Which compiler is wrong or right? Both compilers are running the stock installs, no extra messing around in the options. Is one or the other incomplete? Deviating? I don't have a preference right now of one over the other, I just want to know what the real error is, and what the proper way to handle it is.

Did you declare any constructor for TimeCalc to be explicit? Did you declare the copy constructor for TimeCalc?

EDIT:
Heh, I can duplicate your problem on g++ 4.4.4 when I declare the copy constructor to take a non-const reference. Just declare the parameter as a const reference.

Quote:

---

Originally Posted by laserlight

Did you declare any constructor for TimeCalc to be explicit? Did you declare the copy constructor for TimeCalc?

EDIT:
Heh, I can duplicate your problem on g++ 4.4.4 when I declare the copy constructor to take a non-const reference. Just declare the parameter as a const reference.

---

Thanks. I added const to the copy constructor and it works now (it shames me to learn I missed something so elementary), but what was the deal? I stepped through the file with the debugger in fine detail in both CodeBlocks and VS2010, and neither of them went to the copy constructor. In both, it called the constructor that takes a string parameter. So why did the copy constructor issue even make a difference?

When it comes to differences between gcc and MS, I'd tend to side with gcc because I can't imagine the open source community would deviated from accepted best practices and defined standards (whereas MS often seems to go off and do their own thing). Does VS just ignore the discrepancy between the const call and the non-const declaration?

Formally, given a choice of those constructors, the compiler must interpret the code as

1) invoking the const char * constructor to convert the string to a (temporarily created) TimeCalc
2) invoking a copy constructor to copy that temporary to another TimeCalc.

The C++ standard does not allow passing of a temporary by non-const reference. Hence the complaint by the compiler until you make a copy constructor that accepts a const reference.

Yes, it is true that a compiler is allowed to avoid creating temporaries in some circumstances (if the only way of detecting the temporary is by tracking constructor and destructor calls). But, even if the compiler does avoid creating temporaries, it is still required by the standard diagnose the code as if it is creating those temporaries.

VS2010 has got it wrong. It is presumably avoiding the creation of the temporaries. But it is also changing how the code is diagnosed.

Quote:

---

Originally Posted by grumpy

VS2010 has got it wrong.

---

I think it would be more precise to say VS2010 is not following the standards. Getting it wrong implies they did NOT do it on purpose; Microsoft has a past trend of not following the standards on purpose more often than by accident not following them.

Tim S.

Quote:

---

VS2010 has got it wrong. It is presumably avoiding the creation of the temporaries. But it is also changing how the code is diagnosed.

---

No, it just allows binding temporaries to non-const references. (And there might not be a way to turn off this compiler extension?)

Quote:

---

Originally Posted by anon

No, it just allows binding temporaries to non-const references. (And there might not be a way to turn off this compiler extension?)

---

Yes, this is an extension by Microsoft (and if you compile it, you will get warnings about it). It can be turned off in settings. Disable language extensions, it is called. Can't remember exactly where, but it should be easy to find.

There's a constructor that takes a const char *, why does some pointless temporary even come into it?

I believe this has already been explained.

Code:

---

TimeCalc time5 = "14:34";

---

means almost the same as

Code:

---

TimeCalc time5 = TimeCalc("14:34");

---

and

Code:

---

TimeCalc time5(TimeCalc("14:34"));

---

except that the first also requires the availability of a non-explicit const char* constructor.

It's just the rule of the language that the validity of such a construct is checked, even though in 100% of the cases the compiler will just optimize away the unnecessary constructor calls.

I suppose if you relax the rules to not bother about syntactic validity if something can be optimized away, you'd end up with the compiler's optimizing capabilities and settings determining whether a piece of code is valid or not (and just introduce more unnecessary special cases into the language specification which is complicated as it is).

Quote:

---

Originally Posted by anon

I believe this has already been explained.

Code:

---

TimeCalc time5 = "14:34";

---

means almost the same as

Code:

---

TimeCalc time5 = TimeCalc("14:34");

---

---

I thought a declaration/initialization of the form:

Code:

---

Foo x = y;

---

Was just another way of writing:

Code:

---

Foo x(y);

---

How the hell can I have that wrong?!

Apart from the (potential) introduction/elimination of temporaries by the compiler if y is not of type Foo, you were not wrong.

(Also assuming Foo has a non-explicit copy constructor and conversion constructors).

Thanks for the responses guys. These are the kinds of things I like to learn and know about.