www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - writeln() assertion failed in hybrid x64

reply "Szymon Gatner" <noemail gmail.com> writes:
Hey,

I am trying to build hybrid (C++, D) application (more here: 
http://forum.dlang.org/thread/ugkpqprobonorbdunxsx forum.dlang.org) 
but I am now getting assertion failure from within writeln(). 
writeln() is called from a D function that has C++ linkage:

D definition:

extern (C++) void printSomething()
{
   writeln("hello from D");
}


usage from C++:

extern "C++" void printSomething();

int main()
{
   DRuntime druntime; // rt_init(), rt_term()
   printSomething();
}

this causes run-time assertion in fprintf() called from within 
writeln():

int __cdecl fprintf (
         FILE *str,
         const char *format,
         ...
         )
/*
  * 'F'ile (stream) 'PRINT', 'F'ormatted
  */
{
     va_list(arglist);
     FILE *stream;
     int buffing;
     int retval=0;

     _VALIDATE_RETURN( (str != NULL), EINVAL, -1); <=== assetion 
here

[...]
}

meaning that str arg passed is null. writelns()'s call site:

         enforce(fprintf(.stdout._p.handle, "%.*s\n",
                         cast(int) args[0].length, args[0].ptr) >= 
0);

so for some reason .stdout._p.handle is null.
Sep 03 2014
parent reply "Szymon Gatner" <noemail gmail.com> writes:
On Wednesday, 3 September 2014 at 09:55:55 UTC, Szymon Gatner 
wrote:

 Hey,

 I am trying to build hybrid (C++, D) application (more here: 
 http://forum.dlang.org/thread/ugkpqprobonorbdunxsx forum.dlang.org) 
 but I am now getting assertion failure from within writeln(). 
 writeln() is called from a D function that has C++ linkage:

 D definition:

 extern (C++) void printSomething()
 {
   writeln("hello from D");
 }


 usage from C++:

 extern "C++" void printSomething();

 int main()
 {
   DRuntime druntime; // rt_init(), rt_term()
   printSomething();
 }

 this causes run-time assertion in fprintf() called from within 
 writeln():

 int __cdecl fprintf (
         FILE *str,
         const char *format,
         ...
         )
 /*
  * 'F'ile (stream) 'PRINT', 'F'ormatted
  */
 {
     va_list(arglist);
     FILE *stream;
     int buffing;
     int retval=0;

     _VALIDATE_RETURN( (str != NULL), EINVAL, -1); <=== assetion 
 here

 [...]
 }

 meaning that str arg passed is null. writelns()'s call site:

         enforce(fprintf(.stdout._p.handle, "%.*s\n",
                         cast(int) args[0].length, args[0].ptr)

= 0);


 so for some reason .stdout._p.handle is null.


Any ideas why this happens?
Sep 03 2014
parent reply "Jorge A. S." <donotspamme nowhereland.com> writes:
I'm having an error related to yours: when I call writeln 
function in a closed stdout I will get a segfault message. 
Example:

import std.stdio;

void main() {
   stdout.close();
   write("hello\n");
}

The code above will crash with segfault buf the following code 
will raise an exception instead:

import std.stdio;

void main() {
   stdout.close();
   stdout.write("hello\n");
}

In one of the specializations of the write function in the 
std.stdio (the call site that you showed in your post) no check 
for closed stdout (when stdout._p is null) is done. I can't say 
if this is a bug in the write function or the desired behaviour 
(I'm a novice here).

On Wednesday, 3 September 2014 at 18:48:00 UTC, Szymon Gatner 
wrote:

 On Wednesday, 3 September 2014 at 09:55:55 UTC, Szymon Gatner 
 wrote:

 Hey,

 I am trying to build hybrid (C++, D) application (more here: 
 http://forum.dlang.org/thread/ugkpqprobonorbdunxsx forum.dlang.org) 
 but I am now getting assertion failure from within writeln(). 
 writeln() is called from a D function that has C++ linkage:

 D definition:

 extern (C++) void printSomething()
 {
  writeln("hello from D");
 }


 usage from C++:

 extern "C++" void printSomething();

 int main()
 {
  DRuntime druntime; // rt_init(), rt_term()
  printSomething();
 }

 this causes run-time assertion in fprintf() called from within 
 writeln():

 int __cdecl fprintf (
        FILE *str,
        const char *format,
        ...
        )
 /*
 * 'F'ile (stream) 'PRINT', 'F'ormatted
 */
 {
    va_list(arglist);
    FILE *stream;
    int buffing;
    int retval=0;

    _VALIDATE_RETURN( (str != NULL), EINVAL, -1); <=== assetion 
 here

 [...]
 }

 meaning that str arg passed is null. writelns()'s call site:

        enforce(fprintf(.stdout._p.handle, "%.*s\n",
                        cast(int) args[0].length, args[0].ptr)

= 0);


 so for some reason .stdout._p.handle is null.


 Any ideas why this happens?
Sep 04 2014
next sibling parent reply ketmar via Digitalmars-d-learn <digitalmars-d-learn puremagic.com> writes:
On Thu, 04 Sep 2014 15:10:21 +0000
"Jorge A. S. via Digitalmars-d-learn"
<digitalmars-d-learn puremagic.com> wrote:


 In one of the specializations of the write function in the=20
 std.stdio (the call site that you showed in your post) no check=20
 for closed stdout (when stdout._p is null) is done. I can't say=20
 if this is a bug in the write function or the desired behaviour=20
 (I'm a novice here).


it's definitely a bug. writeln() should throw, not segfaulting.
Sep 04 2014
parent reply "Szymon Gatner" <noemail gmail.com> writes:
On Thursday, 4 September 2014 at 15:25:59 UTC, ketmar via 
Digitalmars-d-learn wrote:

 On Thu, 04 Sep 2014 15:10:21 +0000
 "Jorge A. S. via Digitalmars-d-learn"
 <digitalmars-d-learn puremagic.com> wrote:


 In one of the specializations of the write function in the 
 std.stdio (the call site that you showed in your post) no 
 check for closed stdout (when stdout._p is null) is done. I 
 can't say if this is a bug in the write function or the 
 desired behaviour (I'm a novice here).


 it's definitely a bug. writeln() should throw, not segfaulting.


Shouldn't writeln() work tho instead of throwing or segfaulting?
Sep 04 2014
parent reply "Marc =?UTF-8?B?U2Now7x0eiI=?= <schuetzm gmx.net> writes:
On Thursday, 4 September 2014 at 17:57:47 UTC, Szymon Gatner 
wrote:

 On Thursday, 4 September 2014 at 15:25:59 UTC, ketmar via 
 Digitalmars-d-learn wrote:

 On Thu, 04 Sep 2014 15:10:21 +0000
 "Jorge A. S. via Digitalmars-d-learn"
 <digitalmars-d-learn puremagic.com> wrote:


 In one of the specializations of the write function in the 
 std.stdio (the call site that you showed in your post) no 
 check for closed stdout (when stdout._p is null) is done. I 
 can't say if this is a bug in the write function or the 
 desired behaviour (I'm a novice here).


 it's definitely a bug. writeln() should throw, not segfaulting.


 Shouldn't writeln() work tho instead of throwing or segfaulting?


No, if stdout is closed, how should it work? Throwing is the 
correct behaviour.
Sep 04 2014
parent reply "Szymon Gatner" <noemail gmail.com> writes:
On Thursday, 4 September 2014 at 18:22:55 UTC, Marc Schütz wrote:

 On Thursday, 4 September 2014 at 17:57:47 UTC, Szymon Gatner 
 wrote:

 On Thursday, 4 September 2014 at 15:25:59 UTC, ketmar via 
 Digitalmars-d-learn wrote:

 On Thu, 04 Sep 2014 15:10:21 +0000
 "Jorge A. S. via Digitalmars-d-learn"
 <digitalmars-d-learn puremagic.com> wrote:


 In one of the specializations of the write function in the 
 std.stdio (the call site that you showed in your post) no 
 check for closed stdout (when stdout._p is null) is done. I 
 can't say if this is a bug in the write function or the 
 desired behaviour (I'm a novice here).


 it's definitely a bug. writeln() should throw, not 
 segfaulting.


 Shouldn't writeln() work tho instead of throwing or 
 segfaulting?


 No, if stdout is closed, how should it work? Throwing is the 
 correct behaviour.


Ah sorry, I totaly missed close() part.

Any idea why it is null in my case tho?
Sep 04 2014
next sibling parent reply "Kagamin" <spam here.lot> writes:
Maybe some module constructor wasn't run due to linking mess. So 
it remains uninitialized.
Sep 04 2014
parent reply "Szymon Gatner" <noemail gmail.com> writes:
On Thursday, 4 September 2014 at 20:38:38 UTC, Kagamin wrote:

 Maybe some module constructor wasn't run due to linking mess. 
 So it remains uninitialized.


Is there a way I can check if module c-tor run? rt_init() 
returned no error.
Sep 04 2014
parent reply "Kagamin" <spam here.lot> writes:
https://github.com/D-Programming-Language/druntime/blob/master/
rc/rt/dmain2.d#L270 
well, this sucks.
Sep 04 2014
parent reply "Szymon Gatner" <noemail gmail.com> writes:
On Thursday, 4 September 2014 at 20:57:41 UTC, Kagamin wrote:

 https://github.com/D-Programming-Language/druntime/blob/master/
rc/rt/dmain2.d#L270 
 well, this sucks.


Is there a way I can call module c-tors explicitly?

I was under impression that D(dmd) was suppose to work with 
VisualC++ in x64bit mode.
Sep 05 2014
parent "Kagamin" <spam here.lot> writes:
It's not a module ctor, this code is executed much earlier. You 
can write a function, which will initialize standard streams, and 
call it from the C code before rt_init.
Sep 05 2014
prev sibling parent "Kagamin" <spam here.lot> writes:
https://github.com/D-Programming-Language/druntime/blob/master/src/co
e/stdc/stdio.d#L457 
see? It's null. Hmm... where is it initialized?
Sep 04 2014
prev sibling parent reply "hane" <han.ehit.suzi.0 gmail.com> writes:
On Thursday, 4 September 2014 at 15:10:22 UTC, Jorge A. S. wrote:

 I'm having an error related to yours: when I call writeln 
 function in a closed stdout I will get a segfault message. 
 Example:

 import std.stdio;

 void main() {
   stdout.close();
   write("hello\n");
 }

 The code above will crash with segfault buf the following code 
 will raise an exception instead:

 import std.stdio;

 void main() {
   stdout.close();
   stdout.write("hello\n");
 }

 In one of the specializations of the write function in the 
 std.stdio (the call site that you showed in your post) no check 
 for closed stdout (when stdout._p is null) is done. I can't say 
 if this is a bug in the write function or the desired behaviour 
 (I'm a novice here).


With DMD of git master, the latter code seems to raise Exception 
(stdio.d(2247): Enforcement failed) instead of segfault.
Sep 05 2014
parent "hane" <han.ehit.suzi.0 gmail.com> writes:
On Friday, 5 September 2014 at 07:22:23 UTC, hane wrote:

 On Thursday, 4 September 2014 at 15:10:22 UTC, Jorge A. S. 
 wrote:

 I'm having an error related to yours: when I call writeln 
 function in a closed stdout I will get a segfault message. 
 Example:

 import std.stdio;

 void main() {
  stdout.close();
  write("hello\n");
 }

 The code above will crash with segfault buf the following code 
 will raise an exception instead:

 import std.stdio;

 void main() {
  stdout.close();
  stdout.write("hello\n");
 }

 In one of the specializations of the write function in the 
 std.stdio (the call site that you showed in your post) no 
 check for closed stdout (when stdout._p is null) is done. I 
 can't say if this is a bug in the write function or the 
 desired behaviour (I'm a novice here).


 With DMD of git master, the latter code seems to raise 
 Exception (stdio.d(2247): Enforcement failed) instead of 
 segfault.


Oops, I meant 'former' of course.
Sep 05 2014