Does g++ really need to compile main, if c++ code is involved?

Does g++ really need to compile main, if c++ code is involved?

[Eric Wolf]

Hello there!

We have a big bunch of C and C++ code in our product and compile
main with g++, because I read in

http://www.parashift.com/c++-faq/overview-mixing-langs.html

that this is needed.

Can we get rid of this condition? For example by not
using exceptions, which are not caught and not using global/static
non pod variables and not using class wide non pod variables?

I tried to find answers to that, by searching this mailing list
and the internet in general, but didn't find anything. Most
people don't seem to be aware of that "compile main with c++-compiler"
rule. If I missed some resource, please excuse.

Yours sincerely,

Eric Wolf

Re: Does g++ really need to compile main, if c++ code is involved?

[Tim Prince]

On 3/20/2014 4:27 AM, Eric Wolf wrote:
>
> We have a big bunch of C and C++ code in our product and compile
> main with g++, because I read in
>
> http://www.parashift.com/c++-faq/overview-mixing-langs.html
>
> that this is needed.
>
>
The proper compile mode will be selected regardless of whether you 
invoke gcc, g++, gfortran, ... e.g. according to the file name.  As your 
reference pointed out, that is not true of many other compilers.
The second point in that reference is valid:  g++ includes -lstdc++ 
implicitly, which the other languages do not, so that g++ references can 
be satisfied during linking.

-- 
Tim Prince

RE: Does g++ really need to compile main, if c++ code is involved?

[Andy Falanga (afalanga)]

> -----Original Message-----
> From: gcc-help-owner@gcc.gnu.org [mailto:gcc-help-owner@gcc.gnu.org] On
> Behalf Of Eric Wolf
> Sent: Thursday, March 20, 2014 2:28 AM
> To: gcc-help@gcc.gnu.org
> Subject: Does g++ really need to compile main, if c++ code is involved?
> 
> Hello there!
> 
> We have a big bunch of C and C++ code in our product and compile main
> with g++, because I read in
> 
> http://www.parashift.com/c++-faq/overview-mixing-langs.html
> 
> that this is needed.
> 
> Can we get rid of this condition? For example by not using exceptions,
> which are not caught and not using global/static non pod variables and
> not using class wide non pod variables?

It's not exceptions only for which this is required.  C++ performs "name-mangling" on the symbols which doesn't happen in C.  I'm not sure if I have enough context here to answer decisively, however, I think this FAQ link you've provided assumes that you're wishing to compile a C++ program and link with C sources.  It is possible to do the opposite.  Interestingly, the very FAQ you linked to says as much and that you should continue reading that section for how to perform this.  In fact, this link even states, "BTW there is another way to handle this whole thing: compile all your code (even your C-style code) using a C++ compiler."

I'll bet the rest of the FAQ in section 32 will be most interesting reading for you.

Andy

Re: Does g++ really need to compile main, if c++ code is involved?

[Eric Wolf]

Am 03/20/2014 04:24 PM, schrieb Andy Falanga (afalanga):
>> -----Original Message-----
>>
>> Can we get rid of this condition? For example by not using exceptions,
>> which are not caught and not using global/static non pod variables and
>> not using class wide non pod variables?
>
> It's not exceptions only for which this is required.  C++ performs
 > "name-mangling" on the symbols which doesn't happen in C.
 > I'm not sure if I have enough context here to answer decisively,
 > however, I think this FAQ link you've provided assumes that you're
 > wishing to compile a C++ program and link with C sources.  It is
 > possible to do the opposite.  Interestingly, the very FAQ you linked
 > to says as much and that you should continue reading that section for
 > how to perform this.  In fact, this link even states, "BTW there is
 > another way to handle this whole thing: compile all your code (even
 > your C-style code) using a C++ compiler."

I'm aware of name mangling. From C we only call C++-Functions, which 
have C compatible signatures and declarations wrapped in extern "C"
in the C++ code.

And we only call C functions from C++ code, whose declarations
are wrapped by extern "C" in the C++ code.

This all works well.

I really think, static initializations and exceptions are the problem,
because if I placed main in a C source file and compile it with gcc,
how are the initializations done? How are uncaught exceptions handled?

But there might be more, which I missed, so I asked here.

Using g++ for all the code is not an option. We want to use another
language, which insists on compiling the main function. So we really
want to ditch g++ for the file containing main.

> I'll bet the rest of the FAQ in section 32 will be most interesting reading for you.

I'm aware of the rest of the FAQ.

Maybe I should rephrase the question:

If I want to write a library in C++ with an C interface and I don't want 
to force my clients to use the g++ compiler, what bits do I have
to pay attention to in the C++ code?

Yours sincerely,

Eric Wolf

Re: Does g++ really need to compile main, if c++ code is involved?

[Jonathan Wakely]

On 20 March 2014 15:50, Eric Wolf wrote:
>
> I really think, static initializations and exceptions are the problem,
> because if I placed main in a C source file and compile it with gcc,
> how are the initializations done? How are uncaught exceptions handled?

Using the DWARF-based exception mechanism the compiler looks up the
stack for a suitable catch handler and if none is found (either
because the higher stack frames are C++ functions but don't catch the
exception, or because they are not C++ functions at all) then
std::terminate() is called at the throw site.

So I believe an uncaught exception will terminate the process
irrespective of what language main() is written in.

Global ctors/dtors are a separate issue, I'm not sure how they're
handled if main() isn't written in C++.

Re: Does g++ really need to compile main, if c++ code is involved?

[Ian Lance Taylor]

On Thu, Mar 20, 2014 at 9:38 AM, Jonathan Wakely <jwakely.gcc@gmail.com> wrote:
>
> Global ctors/dtors are a separate issue, I'm not sure how they're
> handled if main() isn't written in C++.

It doesn't matter.  As a GCC extension, you can write a global
constructor in C using __attribute__ ((constructor)), and the result
is equivalent to a C++ global constructor.  It follows that if global
constructors work at all, they will work whether main is written in C
or C++.

Ian

Re: Does g++ really need to compile main, if c++ code is involved?

[rbmj]

(Accidentally didn't reply all the first time, sorry)

On 20-Mar-14 12:38, Jonathan Wakely wrote:
> On 20 March 2014 15:50, Eric Wolf wrote:
>>
>> I really think, static initializations and exceptions are the problem,
>> because if I placed main in a C source file and compile it with gcc,
>> how are the initializations done? How are uncaught exceptions handled?
>
> Using the DWARF-based exception mechanism the compiler looks up the
> stack for a suitable catch handler and if none is found (either
> because the higher stack frames are C++ functions but don't catch the
> exception, or because they are not C++ functions at all) then
> std::terminate() is called at the throw site.
>
> So I believe an uncaught exception will terminate the process
> irrespective of what language main() is written in.

I agree with this.

>
> Global ctors/dtors are a separate issue, I'm not sure how they're
> handled if main() isn't written in C++.
>

Shouldn't these be handled in crt*.o by being placed in .dtors/.ctors? 
As long as these are linked in by gcc/g++ then everything should be fine 
(as a matter of fact the C compiler lets you hook into this 
functionality using attributes IIRC, so it needs to provide that as 
well...).

Re: Does g++ really need to compile main, if c++ code is involved?

[Florian Weimer]

* Eric Wolf:

> We have a big bunch of C and C++ code in our product and compile
> main with g++, because I read in
>
> http://www.parashift.com/c++-faq/overview-mixing-langs.html
>
> that this is needed.

This is target-dependent.  For example, GNU/Linux generally doesn't
have a problem anymore with a C main and C++ code elsewhere.

RE: Does g++ really need to compile main, if c++ code is involved?

[Kilian, Jens]

> I tried to find answers to that, by searching this mailing list
> and the internet in general, but didn't find anything. Most
> people don't seem to be aware of that "compile main with c++-compiler"
> rule. If I missed some resource, please excuse.

I remember only one compiler (HP aCC) which requires compiling main()
as C++.  As far as I know GCC never needed this.

Note that you will still need to link using g++, or pass all the
equivalent options to the linker.

Best regards,

	Jens.

Re: Does g++ really need to compile main, if c++ code is involved?

[Florian Weimer]

* Jens Kilian:

>> I tried to find answers to that, by searching this mailing list
>> and the internet in general, but didn't find anything. Most
>> people don't seem to be aware of that "compile main with c++-compiler"
>> rule. If I missed some resource, please excuse.
>
> I remember only one compiler (HP aCC) which requires compiling main()
> as C++.  As far as I know GCC never needed this.

Some systems need collect2 support for initializers:

  <http://gcc.gnu.org/onlinedocs/gccint/Collect2.html>

> Note that you will still need to link using g++, or pass all the
> equivalent options to the linker.

Doing this for the shared library should be sufficient, though.

RE: Does g++ really need to compile main, if c++ code is involved?

[Kilian, Jens]

> From: Florian Weimer [mailto:fw@deneb.enyo.de]
> * Jens Kilian:
> > I remember only one compiler (HP aCC) which requires compiling main()
> > as C++.  As far as I know GCC never needed this.
> 
> Some systems need collect2 support for initializers:
> 
>   <http://gcc.gnu.org/onlinedocs/gccint/Collect2.html>

I interpret this page to mean that GCC will always call the required __main()
from main(), even if compiling C code.  In this case you wouldn't need to
compile main() using g++ (not gcc).

As other posters have pointed out, this is really platform dependent.
There's a comment in 'libgcc/crtstuff.c' which may be relevant:

  Additionally, if the target system supports a .init section,
  this file allows us to support the linking of C++ code
  with a non-C++ main program.

Best regards,

	Jens.

Re: Does g++ really need to compile main, if c++ code is involved?

[Eric Wolf]

Am 03/24/2014 08:04 AM, schrieb Kilian, Jens:
>> From: Florian Weimer [mailto:fw@deneb.enyo.de]
>> * Jens Kilian:
>>> I remember only one compiler (HP aCC) which requires compiling main()
>>> as C++.  As far as I know GCC never needed this.
>>
>> Some systems need collect2 support for initializers:
>>
>>    <http://gcc.gnu.org/onlinedocs/gccint/Collect2.html>
>
> I interpret this page to mean that GCC will always call the required __main()
> from main(), even if compiling C code.  In this case you wouldn't need to
> compile main() using g++ (not gcc).

But that's strange. With this little example:

----- test.cpp ----------

#include <iostream>

class A {
public:
     int a;
     A() { a = 5; }
     void showA() { std::cout << a << std::endl; }
};

A a;

int main(int argc, char * argv[]) {
     a.showA();
     return 0;
}

----- test.cpp ----------

and doing

g++ -static -O0 -o test test.cpp
nm test | grep main

no __main pops up.

Could be inlined though?

But there is a
__libc_start_main
symbol, maybe __main got renamed?

Yours sincerely,

Eric Wolf

Re: Does g++ really need to compile main, if c++ code is involved?

[Ian Lance Taylor]

On Mon, Mar 24, 2014 at 12:13 AM, Eric Wolf <eric.wolf@abas.de> wrote:
> Am 03/24/2014 08:04 AM, schrieb Kilian, Jens:
>
>>> From: Florian Weimer [mailto:fw@deneb.enyo.de]
>>> * Jens Kilian:
>>>>
>>>> I remember only one compiler (HP aCC) which requires compiling main()
>>>> as C++.  As far as I know GCC never needed this.
>>>
>>>
>>> Some systems need collect2 support for initializers:
>>>
>>>    <http://gcc.gnu.org/onlinedocs/gccint/Collect2.html>
>>
>>
>> I interpret this page to mean that GCC will always call the required
>> __main()
>> from main(), even if compiling C code.  In this case you wouldn't need to
>> compile main() using g++ (not gcc).
>
>
> But that's strange. With this little example:

...

> no __main pops up.
>
> Could be inlined though?

You neglected to say what machine you were running on.  __main is only
used on some platforms.  In particular it is not used on GNU/Linux.


> But there is a
> __libc_start_main
> symbol, maybe __main got renamed?

No, that is a different symbol.

Ian

Re: Does g++ really need to compile main, if c++ code is involved?

[Florian Weimer]

* Ian Lance Taylor:

> You neglected to say what machine you were running on.  __main is only
> used on some platforms.  In particular it is not used on GNU/Linux.

Do you know how this behavior of collect2 is controlled?  I would like
to add this to the gccint documentation.

Re: Does g++ really need to compile main, if c++ code is involved?

[Ian Lance Taylor]

On Mon, Mar 24, 2014 at 11:02 AM, Florian Weimer <fw@deneb.enyo.de> wrote:
> * Ian Lance Taylor:
>
>> You neglected to say what machine you were running on.  __main is only
>> used on some platforms.  In particular it is not used on GNU/Linux.
>
> Do you know how this behavior of collect2 is controlled?  I would like
> to add this to the gccint documentation.

The existing documentation is at
http://gcc.gnu.org/onlinedocs/gccint/Initialization.html and the
following section
http://gcc.gnu.org/onlinedocs/gccint/Macros-for-Initialization.html .

In short, if use_collect2 is not set to yes in config.gcc, then
collect2 will not generate a __main function.  If use_collect2 is set
to yes, but if you define LD_INIT_SWITCH, then collect2 will also not
generate a call to a __main function.

Ian

Re: Does g++ really need to compile main, if c++ code is involved?

[David Brown]

On 20/03/14 13:10, Eric Wolf wrote:
> On 3/20/2014 7:29 AM, Tim Prince wrote:
>> On 3/20/2014 4:27 AM, Eric Wolf wrote:
>>>
>>> We have a big bunch of C and C++ code in our product and compile
>>> main with g++, because I read in
>>>
>>> http://www.parashift.com/c++-faq/overview-mixing-langs.html
>>>
>>> that this is needed.
>>
>> The proper compile mode will be selected regardless of whether you
>> invoke gcc, g++, gfortran, ... e.g. according to the file name. As
>> your reference pointed out, that is not true of many other compilers.
> 
> That's not the point.
> 
> I can easily move the main function into a file ending with .c. Linking
> will still succeed, but I assume, that things will start to go terribly
> wrong, when the c++ code throws an exception, which is caught nowhere,
> or if one tries to use a global non pod object.

Why do you /assume/ this?  Just try it and see.  It should be a simple
matter to write a few test files.

> 
> Here I need a catalog of things, I am not allowed to do in the c++
> portion of the code, so that nothing goes wrong.
> 
>> The second point in that reference is valid: g++ includes -lstdc++
>> implicitly, which the other languages do not, so that g++ references
>> can be satisfied during linking.
> 
> That's useful. So I can use ld to link again, as long as I include
> -lstdc++ Option. That's all?

It is normal practice to use gcc (or g++) to run the linker, rather than
calling it explicitly.  It makes it easier to get the write options
passed on, and makes everything a little more consistent.

> 
> Please excuse the missing References header. I subscribed the digest
> version of the mailing list and my mail programm is defective. I can't
> set the References header by hand.
> 
> Yours sincerely,
> 
> Eric Wolf
> 
> 

Have you tried googling for "difference between gcc and g++"?  There are
plenty of good answers here.

<http://stackoverflow.com/questions/172587/what-is-the-difference-between-g-and-gcc>

The key points are that "-lstdc++" is included by g++ when you use it
for the linker - you need to add it explicitly if using gcc to link.
And it seems that g++ will treat .c files as C++, while gcc treats them
as C files.


So the key issue is not whether you will use g++ or gcc, but whether you
can compile main() in a language other than C++ and still get features
such as static constructors and full exception handling.  I don't know
the answer to that one.

Re: Does g++ really need to compile main, if c++ code is involved?

[Eric Wolf]

On 3/20/2014 7:29 AM, Tim Prince wrote:
> On 3/20/2014 4:27 AM, Eric Wolf wrote:
>>
>>We have a big bunch of C and C++ code in our product and compile
>>main with g++, because I read in
>>
>>http://www.parashift.com/c++-faq/overview-mixing-langs.html
>>
>>that this is needed.
>
> The proper compile mode will be selected regardless of whether you invoke gcc, g++, gfortran, ... e.g. according to the file name. As your reference pointed out, that is not true of many other compilers.

That's not the point.

I can easily move the main function into a file ending with .c. Linking 
will still succeed, but I assume, that things will start to go terribly
wrong, when the c++ code throws an exception, which is caught nowhere,
or if one tries to use a global non pod object.

Here I need a catalog of things, I am not allowed to do in the c++ 
portion of the code, so that nothing goes wrong.

> The second point in that reference is valid: g++ includes -lstdc++ implicitly, which the other languages do not, so that g++ references can be satisfied during linking.

That's useful. So I can use ld to link again, as long as I include 
-lstdc++ Option. That's all?

Please excuse the missing References header. I subscribed the digest 
version of the mailing list and my mail programm is defective. I can't
set the References header by hand.

Yours sincerely,

Eric Wolf