Efficient implementation of coroutines requires CPU-specific code in the RTL and possibly the compiler. However
http://www.chiark.greenend.org.uk/~sgtatham/coroutines.html suggests a way that coroutines can be implemented in a portable fashion in C, how can this be done in Object Pascal?
Mailing lists / fpc-pascal / 2015-03
Portable coroutines
7 messages · 4 people · –
- MMark Morgan Lloyd 4
- MMarco van de Voort 1
- MMichael Schnell 1
- UUnknown author 1
Reply by email Subscribe to this list Your subscription
Read-only archive. Use j and k to move between messages.
In our previous episode, Mark Morgan Lloyd said:
Show quoted text 5 lines
Seems more an oddity than a system, but it relies heavily on preprocessor and fallthrough case.
The preprocessor part can be done, just use whatever preprocessor (maybe even cpp) and then haul the resulting code through fpc.
But there is no fallthrough case (and personally I think that is a good thing)
Maybe look at the coroutine implementation in Go (Golang).
Maybe a better refrence. Maybe an optional switch in compiler so you are not required to have all CPU architectures working on day 1.
Intel and ARM seem nice...
md
Marco van de Voort wrote 19 lines
Show quoted text 6 lines
The issue isn't the syntax, it's that proper implementation of coroutines means switching stacks which is highly architecture-specific; adding something like that to the RTL is a non-starter.
In our previous episode, Mark Morgan Lloyd said: Seems more an oddity than a system, but…
Marco van de Voort wrote 15 lines
Generally agreed. I'm not particularly bothered about the syntax, although obviously having it fairly compact would be an advantage. I don't think it's possible to use a case/goto arrangement since this wouldn't be happy jumping into repeat or while loops, but it does look as though it's possible to use LongJmp() (insomnia has its advantages):
program coroutines;
uses SysUtils;
type TState= record
env: jmp_buf;
line: cardinal
end;
var state: TState;
once: boolean= false;
scratch: char;
function getchar(): char;
var sanity: jmp_buf;
begin
// Check that we're not trying to jump into exception blocks etc.
if state.line 0 then begin
SetJmp(sanity);
Assert(PtrUInt(state.env.sp) + PtrUInt(state.env.bp) =
PtrUInt(sanity.sp) + PtrUInt(sanity.bp),
'Bad SP or BP at xfer to line ' + IntToStr(state.line));
LongJmp(state.env, 1)
end;
SetJmp(state.env);
if state.line = 0 then begin
state.line := StrToInt( (*$I %LINE% *) );
exit('A')
end else
state.line := 0;..
repeat
SetJmp(state.env);
if state.line = 0 then begin
state.line := StrToInt( (*$I %LINE% *) );
exit(#$ff)
end else
state.line := 0;
once := true
until once
end { getChar } ;
begin
FillByte(state, SizeOf(state), 0);
scratch := getChar();
while scratch #$ff do begin
WriteLn(scratch);
scratch := getChar()
end
end.The sanity check at the start is obviously processor-specific, but it doesn't use anything hidden and can at a pinch be omitted. It's obviously excessively verbose, but it is comparatively regular and does appear to handle at least some loop types properly.
Efficient implementation of coroutines requires CPU-specific code in the RTL and possibly…
Mark Morgan Lloyd wrote 6 lines
Lazarus seems to supports a similar concept by using events and "Application.ProcessMessages". it might not be especially efficient, though.
-Michael
Michael Schnell wrote 10 lines
My tentative solution simplifies to the excerpt below, which I think is reasonably efficient.
..
// Check that we're not trying to jump into exception blocks etc. Valid
// line numbers are known to be > 0, the assertion is CPU-specific.
if state.line > 0 then begin
SetJmp(sanity);
Assert(PtrUInt(state.env.sp) shr 1 xor PtrUInt(state.env.bp) =
PtrUInt(sanity.sp) shr 1 xor PtrUInt(sanity.bp),
'Bad stack frame at xfer to line ' +
IntToStr(state.line - 1));
LongJmp(state.env, state.line)
end;
if SetJmp(state.env) = 0 then begin
state.line := StrToInt( (*$I %LINE% *) );
exit('A')
end;
if SetJmp(state.env) = 0 then begin
..Pity about the StrToInt(), which in principle could be resolved at compilation time.
