druntime commit, revision 483


user: sean

msg:
Added new core.time.  Tested on OSX.  Still need to test on Windows.

http://www.dsource.org/projects/druntime/changeset/483

paths changed:
U   trunk/src/core/sync/condition.d
U   trunk/src/core/sync/config.d
U   trunk/src/core/sync/semaphore.d
U   trunk/src/core/thread.d
U   trunk/src/core/time.d

This commit also fixes a race issue between newly created threads and the GC.  I have one remaining issue to sort out, but not being able to debug on OSX is somewhat of a problem.  I'll look into it more on Linux this weekend, with luck.  The issue is that when I run the following app, sometimes I get a bus error, it hangs, I get a segfault, or there's no output.  It's something within std.concurrency, but it's hard to say what without a debugger.  Here's the code:

import std.concurrency;
import core.stdc.stdio;

void thr() {
  scope(failure) printf( "failure\n" );
  receive(
    (OwnerTerminated)
    {
        printf( "received\n" );
    }
 );
 printf( "done\n" );
}

void main() {
spawn(&thr);
}

And here's a backtrace:

Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00000074
[Switching to process 4169]
0x000025af in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb13onLinkDeadMsgMFKS3std11concurrency7MessageZb ()
(gdb) bt
#0  0x000025af in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb13onLinkDeadMsgMFKS3std11concurrency7MessageZb ()
#1  0x0000278b in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb12onControlMsgMFKS3std11concurrency7MessageZb ()
#2  0x00002800 in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb4scanMFKS3std11concurrency36__T4ListTS3std11concurrency7MessageZ4ListZb ()
#3  0x0000248d in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb ()
#4  0x00002346 in D3std11concurrency52__T7receiveTDFC3std11concurrency15OwnerTerminatedZvZ7receiveFDFC3std11concurrency15OwnerTerminatedZvZv ()
#5  0x0000229b in D3xxx3thrFZv ()
#6  0x00004138 in D3std11concurrency11__T6_spawnZ6_spawnFbPFZvZS3std11concurrency3Tid4execMFZv ()
#7  0x0000c5a6 in D4core6thread6Thread3runMFZv ()
#8  0x0000bb20 in thread_entryPoint ()
#9  0x960277fd in _pthread_start ()
#10 0x96027682 in thread_start ()

 Seems like I'm passing a NULL into the ctor for Message, but I don't see where.  That this appears to be timing related is even weirder.  Hopefully a real debugger will turn something up.

I think I've got it.  The collection that's running is probably when the runtime is terminating, so the spawned thread didn't start (and add itself to the thread list) until after thread_joinAll was called and things were shutting down.  Almost anything could have vanished by the time the receive operation is executed, and thus the weird behavior.  I'll have to go back to having Thread.start add the new thread to the list.  It's the way the code used to work until someone complained about a segfault that turned out to be a collection occurring when the new thread was still in the depths of the kernel startup code.  It's likely I'll have to add finer-grained state tracking, so a STARTING thread is suspended/resumed but not scanned, while a RUNNING thread is both suspended and scanned.  Sucks, but I don't see a way around it.

On Jan 7, 2011, at 10:24 PM, Sean Kelly wrote:

> This commit also fixes a race issue between newly created threads and the GC.  I have one remaining issue to sort out, but not being able to debug on OSX is somewhat of a problem.  I'll look into it more on Linux this weekend, with luck.  The issue is that when I run the following app, sometimes I get a bus error, it hangs, I get a segfault, or there's no output.  It's something within std.concurrency, but it's hard to say what without a debugger.  Here's the code:
> 
> import std.concurrency;
> import core.stdc.stdio;
> 
> void thr() {
>  scope(failure) printf( "failure\n" );
>  receive(
>    (OwnerTerminated)
>    {
>        printf( "received\n" );
>    }
> );
> printf( "done\n" );
> }
> 
> void main() {
> spawn(&thr);
> }
> 
> And here's a backtrace:
> 
> Program received signal EXC_BAD_ACCESS, Could not access memory.
> Reason: KERN_PROTECTION_FAILURE at address: 0x00000074
> [Switching to process 4169]
> 0x000025af in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb13onLinkDeadMsgMFKS3std11concurrency7MessageZb ()
> (gdb) bt
> #0  0x000025af in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb13onLinkDeadMsgMFKS3std11concurrency7MessageZb ()
> #1  0x0000278b in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb12onControlMsgMFKS3std11concurrency7MessageZb ()
> #2  0x00002800 in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb4scanMFKS3std11concurrency36__T4ListTS3std11concurrency7MessageZ4ListZb ()
> #3  0x0000248d in D3std11concurrency10MessageBox48__T3getTDFC3std11concurrency15OwnerTerminatedZvZ3getMFDFC3std11concurrency15OwnerTerminatedZvZb ()
> #4  0x00002346 in D3std11concurrency52__T7receiveTDFC3std11concurrency15OwnerTerminatedZvZ7receiveFDFC3std11concurrency15OwnerTerminatedZvZv ()
> #5  0x0000229b in D3xxx3thrFZv ()
> #6  0x00004138 in D3std11concurrency11__T6_spawnZ6_spawnFbPFZvZS3std11concurrency3Tid4execMFZv ()
> #7  0x0000c5a6 in D4core6thread6Thread3runMFZv ()
> #8  0x0000bb20 in thread_entryPoint ()
> #9  0x960277fd in _pthread_start ()
> #10 0x96027682 in thread_start ()
> 
> Seems like I'm passing a NULL into the ctor for Message, but I don't see where.  That this appears to be timing related is even weirder.  Hopefully a real debugger will turn something up.
> _______________________________________________
> D-runtime mailing list
> D-runtime at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/d-runtime

All three build started failing:

win32:
../dmd/src/dmd -c -d -o- -Isrc -Iimport -Hfimport\core\thread.di src\core\thread.d
src\core\thread.d(1084): Error: undefined identifier milliseconds
src\core\thread.d(1093): Error: function core.time.Duration.opCmp (in
const(Duration) rhs) const is not callable using argument types (_error_)
src\core\thread.d(1097): Error: 'val' is not a scalar, it is a Duration
src\core\thread.d(1097): Error: 'val' is not of arithmetic type, it is a Duration
src\core\thread.d(1099): Error: no property 'totalMilliseconds' for type 'Duration'

Linux 32 and 64:
../dmd/src/dmd -c -d -o- -Isrc -Iimport -Hfimport/core/time.di src/core/time.d
src/core/time.d(1294): Error: undefined identifier CLOCK_MONOTONIC
src/core/time.d(1860): Error: undefined identifier CLOCK_MONOTONIC




On 1/7/2011 10:09 PM, dsource.org wrote:
> druntime commit, revision 483
> 
> 
> user: sean
> 
> msg:
> Added new core.time.  Tested on OSX.  Still need to test on Windows.
> 
> http://www.dsource.org/projects/druntime/changeset/483
> 
> paths changed:
> U   trunk/src/core/sync/condition.d
> U   trunk/src/core/sync/config.d
> U   trunk/src/core/sync/semaphore.d
> U   trunk/src/core/thread.d
> U   trunk/src/core/time.d
> 
> _______________________________________________
> D-runtime mailing list
> D-runtime at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/d-runtime

On Saturday 08 January 2011 00:00:21 Brad Roberts wrote:
> All three build started failing:
> 
> win32:
> ../dmd/src/dmd -c -d -o- -Isrc -Iimport -Hfimport\core\thread.di
> src\core\thread.d src\core\thread.d(1084): Error: undefined identifier
> milliseconds
> src\core\thread.d(1093): Error: function core.time.Duration.opCmp (in
> const(Duration) rhs) const is not callable using argument types (_error_)
> src\core\thread.d(1097): Error: 'val' is not a scalar, it is a Duration
> src\core\thread.d(1097): Error: 'val' is not of arithmetic type, it is a
> Duration src\core\thread.d(1099): Error: no property 'totalMilliseconds'
> for type 'Duration'
> 
> Linux 32 and 64:
> ../dmd/src/dmd -c -d -o- -Isrc -Iimport -Hfimport/core/time.di
> src/core/time.d src/core/time.d(1294): Error: undefined identifier
> CLOCK_MONOTONIC src/core/time.d(1860): Error: undefined identifier
> CLOCK_MONOTONIC

For the monotonic clock,

    enum CLOCK_MONOTONIC    = 1;

needs to be added around line 140 of core.sys.posix.time (under CLOCK_REALTIME would probably the be the place to put it).

- Jonathan M Davis

On Saturday 08 January 2011 00:00:21 Brad Roberts wrote:
> All three build started failing:
> 
> win32:
> ../dmd/src/dmd -c -d -o- -Isrc -Iimport -Hfimport\core\thread.di
> src\core\thread.d src\core\thread.d(1084): Error: undefined identifier
> milliseconds
> src\core\thread.d(1093): Error: function core.time.Duration.opCmp (in
> const(Duration) rhs) const is not callable using argument types (_error_)
> src\core\thread.d(1097): Error: 'val' is not a scalar, it is a Duration
> src\core\thread.d(1097): Error: 'val' is not of arithmetic type, it is a
> Duration src\core\thread.d(1099): Error: no property 'totalMilliseconds'
> for type 'Duration'

I believe that line 1084 of core.thread needs to be changed to

            auto maxSleepMillis = dur!"msecs"(uint.max - 1);

Line 1095 needs to be changed to

                Sleep( cast(uint)
                       maxSleepMillis.total!"msecs"() );

Line 1099 needs to be changed to

            Sleep( cast(uint) val.total!"msecs"() );


I believe that that will fix those, but I haven't tested them. I really should figure out how to build druntime in wine.

- Jonathna M Davis

I got a new laptop the other day and haven't fully rebuilt my Windows partition or I'd have already tested this.  I'll see about getting DMD installed today.

On Jan 8, 2011, at 1:00 AM, Jonathan M Davis wrote:

> On Saturday 08 January 2011 00:00:21 Brad Roberts wrote:
>> All three build started failing:
>> 
>> win32:
>> ../dmd/src/dmd -c -d -o- -Isrc -Iimport -Hfimport\core\thread.di
>> src\core\thread.d src\core\thread.d(1084): Error: undefined identifier
>> milliseconds
>> src\core\thread.d(1093): Error: function core.time.Duration.opCmp (in
>> const(Duration) rhs) const is not callable using argument types (_error_)
>> src\core\thread.d(1097): Error: 'val' is not a scalar, it is a Duration
>> src\core\thread.d(1097): Error: 'val' is not of arithmetic type, it is a
>> Duration src\core\thread.d(1099): Error: no property 'totalMilliseconds'
>> for type 'Duration'
> 
> I believe that line 1084 of core.thread needs to be changed to
> 
>            auto maxSleepMillis = dur!"msecs"(uint.max - 1);
> 
> Line 1095 needs to be changed to
> 
>                Sleep( cast(uint)
>                       maxSleepMillis.total!"msecs"() );
> 
> Line 1099 needs to be changed to
> 
>            Sleep( cast(uint) val.total!"msecs"() );
> 
> 
> I believe that that will fix those, but I haven't tested them. I really should figure out how to build druntime in wine.
> 
> - Jonathna M Davis
> _______________________________________________
> D-runtime mailing list
> D-runtime at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/d-runtime