AmigaDOS threads (was: Re: killing thread ?)

[Irmen de Jong]

Dennis Lee Bieber wrote:
>         Or, to use Amiga terms, the bottom of the hierarchy was an Exec task; 
> AmigaDOS processes were tasks with extra data in the task control block 
> -- primarily entries for stdin, stdout, and some other AmigaDOS level 
> data (Exec was the task switch/memory allocation/IPC [messages] kernel 
> library; I/O and other stuff were libraries built using Exec 
> capabilities. AmigaDOS was the "user level" view, the shell commands 
> and file concepts.)

You're right ofcourse. Though I always refer to the Amiga's OS with
"AmigaDOS".   Others use "Kickstart", which was actually the name
for the 512Kbyte ROM that contained the core OS. ;-)

>         In a way, the Exec message port and signal bits were a close match to 
> DEC VMS mailboxs (with the above "zero-copy" constraint -- what was 
> passed was a list node with the address of the message data) and 
> process local event flags (local, as you could not share a signal bit 
> between tasks -- though there likely was some scheme by which one task 
> could locate/modify the bits of another).

I don't know much about VMS, but I do know that a lot of nice stuff
was borrowed from VMS (and Unix) and used in the creation of AmigaDOS.
( 'Assigns' , anyone?)

>         Of course, even signal bits were only effective if the program 
> explicitly checked them -- a common problem for neophyte coders is that 
> <ctrl-c> had no effect on a process bound loop as the signal check was 
> done as part of the C I/O library.
> 
>         while (1)
>                 x = y;
> 
> could not be broken from the command line as it never tests the signal 
> bits.

You hit the nail on the head.
Thanks for pointing this out again.
Because of this requirement, the SIGNAL solution for aborting a thread,
proposed elsewhere in this discussion, would never work on the Amiga
if the thread doesn't check for the signal explicitly.

Though I now remember the following: it was possible to set a special
function for a designated "interrupt" signal. (there is a field for
this in the Exec task structure). I even used this in
AmigaPython to catch and process the CTRL-C (or "break") signal,
if I'm not mistaken. So it might be possible after all!

Well, it doesn't matter anymore. AmigaPython died at version 2.0,
without thread support. ;-)


Irmen de Jong