• Basile B. (48/48) Mar 30 2016 Hey, I have a class that wraps a process and implements two
• Adam D. Ruppe (5/6) Mar 30 2016 You are already using poll()... I'd just use read() directly on
• Basile B. (5/11) Mar 30 2016 It's probably the only way. I've also tried
• Steven Schveighoffer (3/14) Mar 31 2016 stdio.readf is buffered. It does not deal with async io properly I think...
• Basile B. (4/7) Mar 31 2016 Yes, I must use core.sys.posix.unistd.read, which was actually

Basile B. <b2.temp gmx.com> writes:

Hey, I have a class that wraps a process and implements two 
events. One of them is called after poll() for example if new 
data are available.

The event is called correctly but I don't know how exactly how to 
read the data:

- per buffer of fixed size ?
- using the stream information ?

If i try to read until nothing is available anymore, the event 
handler returns at the end of the process (instead of being 
called several times): eg this helper function of the process 
that the event handler can call to read:


     bool readOutput(T)(ref T t)
     if (isSomeString!T)
     {
         bool result;
         T buffer;
         while (true) // or (!output.eof)
         {
             auto cnt = output.readln(buffer);
             if (cnt)
             {
                 t ~= buffer;
                 result = true;
             }
             else break;
         }
         return result;
     }


the events are called in the callback of a thread-based timer:


     void check(Object notifier)
     {
         if (_ppid.stdout.eof)
         {
             _checker.stop; // the thrad-based timer
             if(_onTerminate)
                 _onTerminate(this);
         }
         else
         {
             // not sure if something else must be done to poll 
properly ?
	    pollfd pfd = { _ppid.stdout.fileno, POLLIN };
	    if (poll(&pfd, 1, 0) == 1 && _onOutputBuffer)
                 _onOutputBuffer(this); // the handler can call 
readOutput
         }
     }

There should be a way to know how many bytes are available ?

Adam D. Ruppe <destructionator gmail.com> writes:

On Thursday, 31 March 2016 at 00:50:16 UTC, Basile B. wrote:
 There should be a way to know how many bytes are available ?

You are already using poll()... I'd just use read() directly on 
the file number too. It will read as much as is available up to 
the max size of the buffer, but if it can't fill it all, it just 
does what it can and returns.

Basile B. <b2.temp gmx.com> writes:

On Thursday, 31 March 2016 at 01:12:50 UTC, Adam D. Ruppe wrote:
 On Thursday, 31 March 2016 at 00:50:16 UTC, Basile B. wrote:
 There should be a way to know how many bytes are available ?

 You are already using poll()... I'd just use read() directly on 
 the file number too. It will read as much as is available up to 
 the max size of the buffer, but if it can't fill it all, it 
 just does what it can and returns.

It's probably the only way. I've also tried 
core.stdc.stdio.readf() in a loop and it was also reading until 
the process termination. So I'll go for a single readf operation 
on a ubyte[].

Steven Schveighoffer <schveiguy yahoo.com> writes:

On 3/30/16 10:06 PM, Basile B. wrote:
 On Thursday, 31 March 2016 at 01:12:50 UTC, Adam D. Ruppe wrote:
 On Thursday, 31 March 2016 at 00:50:16 UTC, Basile B. wrote:
 There should be a way to know how many bytes are available ?

 You are already using poll()... I'd just use read() directly on the
 file number too. It will read as much as is available up to the max
 size of the buffer, but if it can't fill it all, it just does what it
 can and returns.

 It's probably the only way. I've also tried core.stdc.stdio.readf() in a
 loop and it was also reading until the process termination. So I'll go
 for a single readf operation on a ubyte[].

stdio.readf is buffered. It does not deal with async io properly I think.

-Steve

Basile B. <b2.temp gmx.com> writes:

On Thursday, 31 March 2016 at 13:48:46 UTC, Steven Schveighoffer 
wrote:
 stdio.readf is buffered. It does not deal with async io 
 properly I think.

 -Steve

Yes, I must use core.sys.posix.unistd.read, which was actually 
what A.D.Ruppe suggested, I think.