www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - Threads and concurrency

reply Charles <Charles_member pathlink.com> writes:
While on the subject of Threads would someone please hit me w/ the clue bat b/c
I can't seem to get the following working. This example should just call
mthread.run since I've overloaded it . But that's not what happens. It just call

the mthread constructor and finishes. I'd also appreciate any links to some
beginner Thread examples in D. I'm sure the locks code would come in handy at
some point but I'm not there yet.

int main () {
mthread m = new Thread();
m.run(); // <== this should call mthread.run but it just calls the constructor

return 0;
}


module mthread;
import std.thread;

class mthread : Thread {
this () {
super();
std.string.writefln("S thread cons");
std.string.writefln("E thread cons");
}

~this() { }

int run () { foo(); }

void foo () {
std.string.writefln("inside foo");
}
void bar () {
std.string.writefln("inside bar");
}
}
Nov 08 2005
next sibling parent reply Charles <Charles_member pathlink.com> writes:
And if I can get a couple simple Thread examples working I'll post a (hopefully
useful) D Thread tutorial ...

In article <dkrbiu$17v1$1 digitaldaemon.com>, Charles says...

While on the subject of Threads would someone please hit me w/ the clue bat b/c
I can't seem to get the following working. This example should just call
mthread.run since I've overloaded it . But that's not what happens. It just call

the mthread constructor and finishes. I'd also appreciate any links to some
beginner Thread examples in D. I'm sure the locks code would come in handy at
some point but I'm not there yet.

int main () {
mthread m = new Thread();
m.run(); // <== this should call mthread.run but it just calls the constructor

return 0;
}


module mthread;
import std.thread;

class mthread : Thread {
this () {
super();
std.string.writefln("S thread cons");
std.string.writefln("E thread cons");
}

~this() { }

int run () { foo(); }

void foo () {
std.string.writefln("inside foo");
}
void bar () {
std.string.writefln("inside bar");
}
}
Nov 08 2005
parent jicman <jicman_member pathlink.com> writes:
Anything on this Thread Tutorial? :-)

Can anyone point me to a nice D thread sample?  I need to start using more CPU.

thanks.

jose

Charles says...

And if I can get a couple simple Thread examples working I'll post a (hopefully
useful) D Thread tutorial ...

In article <dkrbiu$17v1$1 digitaldaemon.com>, Charles says...

While on the subject of Threads would someone please hit me w/ the clue bat b/c
I can't seem to get the following working. This example should just call
mthread.run since I've overloaded it . But that's not what happens. It just call

the mthread constructor and finishes. I'd also appreciate any links to some
beginner Thread examples in D. I'm sure the locks code would come in handy at
some point but I'm not there yet.

int main () {
mthread m = new Thread();
m.run(); // <== this should call mthread.run but it just calls the constructor

return 0;
}


module mthread;
import std.thread;

class mthread : Thread {
this () {
super();
std.string.writefln("S thread cons");
std.string.writefln("E thread cons");
}

~this() { }

int run () { foo(); }

void foo () {
std.string.writefln("inside foo");
}
void bar () {
std.string.writefln("inside bar");
}
}
Nov 30 2005
prev sibling next sibling parent reply Sean Kelly <sean f4.ca> writes:
Charles wrote:

 While on the subject of Threads would someone please hit me w/ the clue bat b/c
 I can't seem to get the following working. This example should just call
 mthread.run since I've overloaded it . But that's not what happens. It just
call
 
 the mthread constructor and finishes.





 int main () {
 mthread m = new Thread();



I'm surprised this is even allowed.  You're constructing an instance of 
the base class, not mthread.  Try this:

mthread m = new mthread();


Sean
Nov 08 2005
parent reply Charles <Charles_member pathlink.com> writes:
In article <dkrd8s$19km$1 digitaldaemon.com>, Sean Kelly says...


I'm surprised this is even allowed.  You're constructing an instance of 
the base class, not mthread.  Try this:

mthread m = new mthread();


Sean



Sean, thanks for the reply. I've corrected the code to allocate an mthread()
and I still get the same output. Here's my makefile,main.d and mthread.d .
I don't think I can think of a simpler example than this ...

D       = gdc
main: main.d mthread.o
$(D) -o main main.d mthread.o 
mthread.o: mthread.d
$(D) -o mthread.o -c mthread.d
c clean:
rm -f ./*.o

import mthread;
int main () {
mthread m = new mthread();
m.start();
return 0;
}

module mthread;
import std.thread;

class mthread : Thread {
this () {
super();
std.string.writefln("S thread cons");
std.string.writefln("E thread cons");
}
~this() { }
int run () { foo(); }
void foo () {
std.string.writefln("inside foo");
}
void bar () {
std.string.writefln("inside bar");
}
}
Nov 08 2005
next sibling parent reply Sean Kelly <sean f4.ca> writes:
Try this:


import std.thread;
import std.c.stdio;

class mthread : Thread
{
     int run ()
     {
         foo();
         return 0;
     }

     void foo ()
     {
         printf( "inside foo\n" );
     }
}

void main()
{
     mthread m = new mthread();
     m.start();
     m.wait();
}


The problem was tha you didn't have the m.wait() line at the end of main 
so the program was ending before your thread had started.


Sean
Nov 08 2005
parent Charles <Charles_member pathlink.com> writes:
Thanks!!

In article <dkrfmi$1c1a$1 digitaldaemon.com>, Sean Kelly says...

Try this:


import std.thread;
import std.c.stdio;

class mthread : Thread
{
     int run ()
     {
         foo();
         return 0;
     }

     void foo ()
     {
         printf( "inside foo\n" );
     }
}

void main()
{
     mthread m = new mthread();
     m.start();
     m.wait();
}


The problem was tha you didn't have the m.wait() line at the end of main 
so the program was ending before your thread had started.


Sean
Nov 08 2005
prev sibling parent reply Charles <Charles_member pathlink.com> writes:
okay, I solved my own problem ... I forgot to call wait() ... christ.
Here my new main.d

import mthread;
int main () {
mthread m;
for( int i;i<10;i++) {
m = new mthread(i);
m.start();
}
m.wait();            
return 0;
}

In article <dkre89$1alk$1 digitaldaemon.com>, Charles says...

In article <dkrd8s$19km$1 digitaldaemon.com>, Sean Kelly says...


I'm surprised this is even allowed.  You're constructing an instance of 
the base class, not mthread.  Try this:

mthread m = new mthread();


Sean



Sean, thanks for the reply. I've corrected the code to allocate an mthread()
and I still get the same output. Here's my makefile,main.d and mthread.d .
I don't think I can think of a simpler example than this ...

D       = gdc
main: main.d mthread.o
$(D) -o main main.d mthread.o 
mthread.o: mthread.d
$(D) -o mthread.o -c mthread.d
c clean:
rm -f ./*.o

import mthread;
int main () {
mthread m = new mthread();
m.start();
return 0;
}

module mthread;
import std.thread;

class mthread : Thread {
this () {
super();
std.string.writefln("S thread cons");
std.string.writefln("E thread cons");
}
~this() { }
int run () { foo(); }
void foo () {
std.string.writefln("inside foo");
}
void bar () {
std.string.writefln("inside bar");
}
}
Nov 08 2005
parent reply Georg Wrede <georg.wrede nospam.org> writes:
Charles wrote:

 okay, I solved my own problem ... I forgot to call wait() ... christ.
 Here my new main.d
 
 import mthread;
 int main () {
 mthread m;
 for( int i;i<10;i++) {
 m = new mthread(i);
 m.start();
 }
 m.wait();            
 return 0;
 }



I hope the above code was just "a sketch" thrown in, since I don't think 
it is quite right?

((Somebody correct me if I'm wrong here: ))

In the for loop:
  - you create a thread
  - you start the thread
  - you abandon the thread by assigning a new one to m
  - thus you can't reach any of the previously created threads

Luckily, ;-( the previous threads don't get reaped by the GC, because 
std.thread itself has to keep pointers to them. (And you can always 
issue Thread.getAll() to get all the running threads, but it's not a 
substitute for properly being able to e.g. pause, wait, or resume a 
specific known thread.)

After the loop, main waits only for the last created thread. Now, in 
real code this would be dangerous because there's no guarantee that it 
finishes last.

So, while the above code probably "works", it should be corrected by 
somebody better versed (than me) in threads.



 In article <dkre89$1alk$1 digitaldaemon.com>, Charles says...
 

In article <dkrd8s$19km$1 digitaldaemon.com>, Sean Kelly says...



I'm surprised this is even allowed.  You're constructing an instance of 
the base class, not mthread.  Try this:

mthread m = new mthread();


Sean



Sean, thanks for the reply. I've corrected the code to allocate an mthread()
and I still get the same output. Here's my makefile,main.d and mthread.d .
I don't think I can think of a simpler example than this ...

D       = gdc
main: main.d mthread.o
$(D) -o main main.d mthread.o 
mthread.o: mthread.d
$(D) -o mthread.o -c mthread.d
c clean:
rm -f ./*.o

import mthread;
int main () {
mthread m = new mthread();
m.start();
return 0;
}

module mthread;
import std.thread;

class mthread : Thread {
this () {
super();
std.string.writefln("S thread cons");
std.string.writefln("E thread cons");
}
~this() { }
int run () { foo(); }
void foo () {
std.string.writefln("inside foo");
}
void bar () {
std.string.writefln("inside bar");
}
}
Nov 09 2005
parent reply Sean Kelly <sean f4.ca> writes:
Georg Wrede wrote:

 Charles wrote:
 

 okay, I solved my own problem ... I forgot to call wait() ... christ.
 Here my new main.d

 import mthread;
 int main () {
 mthread m;
 for( int i;i<10;i++) {
 m = new mthread(i);
 m.start();
 }
 m.wait();            return 0;
 }



 
 I hope the above code was just "a sketch" thrown in, since I don't think 
 it is quite right?
 
 ((Somebody correct me if I'm wrong here: ))
 
 In the for loop:
  - you create a thread
  - you start the thread
  - you abandon the thread by assigning a new one to m
  - thus you can't reach any of the previously created threads
 
 Luckily, ;-( the previous threads don't get reaped by the GC, because 
 std.thread itself has to keep pointers to them. (And you can always 
 issue Thread.getAll() to get all the running threads, but it's not a 
 substitute for properly being able to e.g. pause, wait, or resume a 
 specific known thread.)
 
 After the loop, main waits only for the last created thread. Now, in 
 real code this would be dangerous because there's no guarantee that it 
 finishes last.



Yup.  The code should probably do something like this instead:

foreach( Thread t; Thread.getAll() )
     t.wait();


Sean
Nov 09 2005
parent reply "Ben Hinkle" <ben.hinkle gmail.com> writes:
 After the loop, main waits only for the last created thread. Now, in real 
 code this would be dangerous because there's no guarantee that it 
 finishes last.



 Yup.  The code should probably do something like this instead:

 foreach( Thread t; Thread.getAll() )
     t.wait();



where "something like" doesn't actually call Thread.getAll(), right? 
Glancing over std.thread getAll returns a slice of the static thread array 
so its contents may change as the foreach loops over it - ie some elements 
might be null if that thread happened to stop by the time the foreach got to 
it. Also Thread.wait errors when you try to wait on yourself so since getAll 
includes the current thread that loop will error at some point.
Nov 09 2005
parent Sean Kelly <sean f4.ca> writes:
Ben Hinkle wrote:

After the loop, main waits only for the last created thread. Now, in real 
code this would be dangerous because there's no guarantee that it 
finishes last.



Yup.  The code should probably do something like this instead:

foreach( Thread t; Thread.getAll() )
    t.wait();



 where "something like" doesn't actually call Thread.getAll(), right? 
 Glancing over std.thread getAll returns a slice of the static thread array 
 so its contents may change as the foreach loops over it - ie some elements 
 might be null if that thread happened to stop by the time the foreach got to 
 it.



Doh.  I'd forgotten about this particularly wonderful aspect of Phobos 
thread behavior.


 Also Thread.wait errors when you try to wait on yourself so since getAll 
 includes the current thread that loop will error at some point. 



So perhaps this then?

foreach( Thread t; Thread.getAll() )
	if( t && t != Thread.getThis() )
		t.wait();

A single pass still won't handle threads that are created during the 
loop, but the original example didn't do that anyway.


Sean
Nov 09 2005
prev sibling parent reply Derek Parnell <derek psych.ward> writes:
On Tue, 8 Nov 2005 23:15:42 +0000 (UTC), Charles wrote:


 While on the subject of Threads would someone please hit me w/ the clue bat b/c
 I can't seem to get the following working...



I don't blame you...the documentation is very unhelpful. I've been playing
with thread for an hour now and I can't get to do much useful work. Has
anyone got some novice-oriented documentation on D threads?

-- 
Derek
(skype: derek.j.parnell)
Melbourne, Australia
"A learning experience is one of those things that says,
 'You know that thing you just did? Don't do that.'" - D.N. Adams
1/12/2005 5:35:32 PM
Nov 30 2005
next sibling parent reply Sean Kelly <sean f4.ca> writes:
Derek Parnell wrote:

 On Tue, 8 Nov 2005 23:15:42 +0000 (UTC), Charles wrote:
 

 While on the subject of Threads would someone please hit me w/ the clue bat b/c
 I can't seem to get the following working...



 I don't blame you...the documentation is very unhelpful. I've been playing
 with thread for an hour now and I can't get to do much useful work. Has
 anyone got some novice-oriented documentation on D threads?



I'll outline how threads are used in Ares and I expect 95% of it will 
apply to Phobos as well.  There are two basic ways to handle 
multithreading in D, subclassing Thread or passing a function pointer to 
a new thread object on construction.  Subclassing can be a risky 
proposition in C++ and other languages without garbage collection 
because the Thread object may be destroyed before the thread completes. 
  As D has GC however, this isn't an issue, and I find subclassing to be 
the most natural way to deal with multithreading in most cases.  The 
function pointer method tend to be quite useful when adapting existing 
code or when you aren't entirely sure whether you will need to 
parallelize a certain function or not.  Here is a short example of how 
to create, run, and wait for threads in D:


import std.thread;
import std.c.stdio;

class MyThread : Thread
{
     // the signature of run() might be different in Phobos--
     // I believe it returns an int or uint
     void run()
     {
         printf( "Hello from MyThread\n" );
     }
}

void threadFunc()
{
     printf( "Hello from threadFunc\n" );
}

void main()
{
     int i = 5;

     void threadDelegate()
     {
         printf( "threadDelegate can see i = %i\n", i );
     }

     MyThread mt = new MyThread();
     Thread   tf = new Thread( &threadFunc );
     Thread   td = new Thread( &threadDelegate );

     // threads must be started explicitly
     mt.start();
     tf.start();
     td.start();

     // you must wait for the threads to complete if you want to
     // be sure their output is displayed
     mt.join(); // join() is called waitFor() in Phobos
     tf.join();
     td.join();
}

C:\code\d>dmd test
C:\bin\dmd\bin\..\..\dm\bin\link.exe test,,,user32+kernel32/noi;

C:\code\d>test
Hello from MyThread
Hello from threadFunc
threadDelegate can see i = 5



When a thread in started, it is guaranteed to see any data changes made 
by the starting thread before the start() call, so threadDelegate is 
guaranteed to see i as 5.  If you intend to subsequently change the 
value of i however, accesses should be done within a synchronized block. 
  The purpose of 'volatile' in D is basically to preserve code order 
when lock-free concurrency is desired, so be aware that it provides no 
memory visibility guarantees like 'volatile' in C++ is (mistakenly) said 
to.  That's all I can think of offhand.  If you have any more specific 
questions, I'd be happy to answer them.


Sean
Nov 30 2005
next sibling parent reply "Kris" <fu bar.com> writes:
I'll just add that subclassing Thread allows one to naturally create their 
own TLS (thread local storage). The alternate method, where TLS is supported 
via an int[] or something similar,  requires some less-than-clear 
bit-stuffing to achieve the equivalent.

- Kris



"Sean Kelly" <sean f4.ca> wrote in message 
news:dmm954$2e4t$1 digitaldaemon.com...

 Derek Parnell wrote:

 On Tue, 8 Nov 2005 23:15:42 +0000 (UTC), Charles wrote:


 While on the subject of Threads would someone please hit me w/ the clue 
 bat b/c
 I can't seem to get the following working...



 I don't blame you...the documentation is very unhelpful. I've been 
 playing
 with thread for an hour now and I can't get to do much useful work. Has
 anyone got some novice-oriented documentation on D threads?



 I'll outline how threads are used in Ares and I expect 95% of it will 
 apply to Phobos as well.  There are two basic ways to handle 
 multithreading in D, subclassing Thread or passing a function pointer to a 
 new thread object on construction.  Subclassing can be a risky proposition 
 in C++ and other languages without garbage collection because the Thread 
 object may be destroyed before the thread completes. As D has GC however, 
 this isn't an issue, and I find subclassing to be the most natural way to 
 deal with multithreading in most cases.  The function pointer method tend 
 to be quite useful when adapting existing code or when you aren't entirely 
 sure whether you will need to parallelize a certain function or not.  Here 
 is a short example of how to create, run, and wait for threads in D:


 import std.thread;
 import std.c.stdio;

 class MyThread : Thread
 {
     // the signature of run() might be different in Phobos--
     // I believe it returns an int or uint
     void run()
     {
         printf( "Hello from MyThread\n" );
     }
 }

 void threadFunc()
 {
     printf( "Hello from threadFunc\n" );
 }

 void main()
 {
     int i = 5;

     void threadDelegate()
     {
         printf( "threadDelegate can see i = %i\n", i );
     }

     MyThread mt = new MyThread();
     Thread   tf = new Thread( &threadFunc );
     Thread   td = new Thread( &threadDelegate );

     // threads must be started explicitly
     mt.start();
     tf.start();
     td.start();

     // you must wait for the threads to complete if you want to
     // be sure their output is displayed
     mt.join(); // join() is called waitFor() in Phobos
     tf.join();
     td.join();
 }

 C:\code\d>dmd test
 C:\bin\dmd\bin\..\..\dm\bin\link.exe test,,,user32+kernel32/noi;

 C:\code\d>test
 Hello from MyThread
 Hello from threadFunc
 threadDelegate can see i = 5



 When a thread in started, it is guaranteed to see any data changes made by 
 the starting thread before the start() call, so threadDelegate is 
 guaranteed to see i as 5.  If you intend to subsequently change the value 
 of i however, accesses should be done within a synchronized block. The 
 purpose of 'volatile' in D is basically to preserve code order when 
 lock-free concurrency is desired, so be aware that it provides no memory 
 visibility guarantees like 'volatile' in C++ is (mistakenly) said to. 
 That's all I can think of offhand.  If you have any more specific 
 questions, I'd be happy to answer them.


 Sean
Nov 30 2005
parent reply Sean Kelly <sean f4.ca> writes:
Kris wrote:

 I'll just add that subclassing Thread allows one to naturally create their 
 own TLS (thread local storage). The alternate method, where TLS is supported 
 via an int[] or something similar,  requires some less-than-clear 
 bit-stuffing to achieve the equivalent.



Agreed.  It can be appropriate for some algorithms but I wouldn't 
suggest it as the default storage choice.  I've also heard the Posix 
implementation of TLS is a tad slow, though Windows TLS should be quite 
decent.  For what it's worth, Thread.getThis() in Ares is implemented 
using TLS to avoid the need to access the global thread list.


Sean
Dec 01 2005
parent "Kris" <fu bar.com> writes:
"Sean Kelly" <sean f4.ca> wrote

  For what it's worth, Thread.getThis() in Ares is implemented using TLS to 
 avoid the need to access the global thread list.



Nice.
Dec 01 2005
prev sibling parent Derek Parnell <derek psych.ward> writes:
On Wed, 30 Nov 2005 23:35:56 -0800, Sean Kelly wrote:


 Derek Parnell wrote:

 On Tue, 8 Nov 2005 23:15:42 +0000 (UTC), Charles wrote:
 

 While on the subject of Threads would someone please hit me w/ the clue bat b/c
 I can't seem to get the following working...



 I don't blame you...the documentation is very unhelpful. I've been playing
 with thread for an hour now and I can't get to do much useful work. Has
 anyone got some novice-oriented documentation on D threads?



 I'll outline how threads are used in Ares and I expect 95% of it will 
 apply to Phobos as well. 



But back in Phobos-land ... I managed to make something that I understand
what is going on. There are two files here...

===== mthread.d ===========
module mthread;
import std.thread;
import std.stdio;
import std.c.windows.windows;

class mthread : Thread
{
    private{
        static int g_id;
        static const uint OneSecond = 1000;
        int m_id;
        int m_V;
    }

    this (int V)
    {
        m_id = ++g_id;
        m_V = V;

        // Register callback function with Thread Central.
        super(&thread_main);

        writefln("create thread #%d", m_id);
    }

    ~this()
    {
        writefln("destroy thread #%d", m_id);
    }

    int id()
    {
        return m_id;
    }

    int init_value()
    {
        return m_V;
    }

    int thread_main ()
    {
        writefln("in  thread_main #%d", m_id);
        for(int i = 0; i < m_V; i++)
        {
            Sleep(OneSecond); // Do something very important
            yield();  // Let someone else have some play time.
        }
        writefln("out thread_main #%d", m_id);
        return 0;
    }

}

===== test.d  =============
import mthread;
import std.math;

int main ()
{
    mthread m[];
    uint allDone = 0;
    uint doneFlag;

    // Create some threads of different durations.
    m ~= new mthread(5);
    m ~= new mthread(7);
    m ~= new mthread(2);
    m ~= new mthread(4);

    doneFlag = cast(uint)pow(2.0l, m.length)-1;

    // Start all the threads running
    foreach(mthread t; m)
    {
        writefln("Starting #%d for %d seconds", t.id, t.init_value);
        t.start();
    }

    // Hang around until they all finish.
    while (allDone != doneFlag)
    {
        foreach(int i, mthread t; m)
        {
            if (t.getState == Thread.TS.TERMINATED)
            {
                uint old;
                old = allDone;
                allDone |= cast(uint)pow(2.0l,i);
                if (old != allDone)
                    writefln("Finished #%d after %d seconds", t.id,
t.init_value);
            }
        }
    }

    writefln("done");


    // Ensure that they clean up.
    foreach(int i, mthread t; m)
    {
        writefln("final wait #%d", t.id);
        t.wait();
        delete t;
    }

    return 0;
}

===========================

And I ran "build test -run" to get ...

C:\temp>build test -run
Path and Version : y:\util\build.exe v2.9(1197)
  built on Wed Aug 10 11:03:42 2005
y:\dmd\bin\..\..\dm\bin\link.exe
test+mthread,test.exe,,user32+kernel32,test.def
/noi;
create thread #1
create thread #2
create thread #3
create thread #4
Starting #1 for 5 seconds
Starting #2 for 7 seconds
Starting #3 for 2 seconds
Starting #4 for 4 seconds
in  thread_main #1
in  thread_main #2
in  thread_main #3
in  thread_main #4
out thread_main #3
Finished #3 after 2 seconds
out thread_main #4
Finished #4 after 4 seconds
out thread_main #1
Finished #1 after 5 seconds
out thread_main #2
Finished #2 after 7 seconds
done
final wait #1
destroy thread #1
final wait #2
destroy thread #2
final wait #3
destroy thread #3
final wait #4
destroy thread #4

C:\temp>

-- 
Derek
(skype: derek.j.parnell)
Melbourne, Australia
"A learning experience is one of those things that says,
 'You know that thing you just did? Don't do that.'" - D.N. Adams
2/12/2005 3:17:07 PM
Dec 01 2005
prev sibling parent J C Calvarese <technocrat7 gmail.com> writes:
In article <137ig7kvpvpus$.1a7ft4fbi03n0$.dlg 40tude.net>, Derek Parnell says...

On Tue, 8 Nov 2005 23:15:42 +0000 (UTC), Charles wrote:


 While on the subject of Threads would someone please hit me w/ the clue bat b/c
 I can't seem to get the following working...



I don't blame you...the documentation is very unhelpful. I've been playing
with thread for an hour now and I can't get to do much useful work. Has
anyone got some novice-oriented documentation on D threads?



I haven't tried running this recently, but I got this to work in the past:
http://trac.dsource.org/projects/tutorials/wiki/ThreadsExample

jcc7
Dec 02 2005