dtrace discuss - Sep 2006 - committing multiple speculations in a single probe

While using DTrace to track down a problem recently, we came across an
unexpected restriction: the compiler will not permit different
speculations to be committed in the same instance of a probe identifier.

For instance, consider the following useless D script:

	#!/usr/sbin/dtrace -s

	#pragma D option nspec=2
	
	BEGIN {
	        spec1 = speculation();
	        spec2 = speculation();
	}
	
	END {
	        commit(spec1);
	        commit(spec2);
	}

	# ./spec.d
	dtrace: failed to compile script ./spec.d: line 12: commit( ) may
	not follow commit( )

However, if the script is changed to:

	#!/usr/sbin/dtrace -s

	#pragma D option nspec=2
	
	BEGIN {
	        spec1 = speculation();
	        spec2 = speculation();
	}
	
	END {
	        commit(spec1);
	}
	END {
	        commit(spec2);
	}

... then it compiles without issue.  Is this restriction intentional?

Thanks!
-- 
meem

[ Apologies if this is a duplicate; I''m having some MTA issues. ]

While using DTrace to track down a problem recently, we came across an
unexpected restriction: the compiler will not permit different
speculations to be committed in the same instance of a probe identifier.

For instance, consider the following useless D script:

        #!/usr/sbin/dtrace -s

        #pragma D option nspec=2

        BEGIN {
                spec1 = speculation();
                spec2 = speculation();
        }

        END {
                commit(spec1);
                commit(spec2);
        }

        # ./spec.d
        dtrace: failed to compile script ./spec.d: line 12: commit( ) may
        not follow commit( )

However, if the script is changed to:

        #!/usr/sbin/dtrace -s

        #pragma D option nspec=2

        BEGIN {
                spec1 = speculation();
                spec2 = speculation();
        }

        END {
                commit(spec1);
        }
        END {
                commit(spec2);
        }

... then it compiles without issue.  Is this restriction intentional?

Thanks!
-- 
meem

> However, if the script is changed to:
> 
> 	#!/usr/sbin/dtrace -s
> 
> 	#pragma D option nspec=2
> 	
> 	BEGIN {
> 	        spec1 = speculation();
> 	        spec2 = speculation();
> 	}
> 	
> 	END {
> 	        commit(spec1);
> 	}
> 	END {
> 	        commit(spec2);
> 	}
> 
> ... then it compiles without issue.  Is this restriction intentional?
I don''t know if I''d say "intentional" -- more like
"unfortunate."  The
issues here are very grotty low-level details, but suffice it to say that
it''s not trivial to allow multiple commit()s in a single clause without
opening up a can of worms (or perhaps a duffle bag of snakes is a more apt
metaphor).  If this particular case is really bothering you, we might be
able to find a way to address it -- but in general commit() doesn''t
play
well with other actions in the same clause, and it''s likely that there
will always be restrictions of one sort or another...

	- Bryan

--------------------------------------------------------------------------
Bryan Cantrill, Solaris Kernel Development.       http://blogs.sun.com/bmc

> > ... then it compiles without issue.  Is this restriction intentional?
 > 
 > I don''t know if I''d say "intentional" -- more
like "unfortunate."  The
 > issues here are very grotty low-level details, but suffice it to say that
 > it''s not trivial to allow multiple commit()s in a single clause
without
 > opening up a can of worms (or perhaps a duffle bag of snakes is a more apt
 > metaphor).  If this particular case is really bothering you, we might be
 > able to find a way to address it -- but in general commit()
doesn''t play
 > well with other actions in the same clause, and it''s likely that
there
 > will always be restrictions of one sort or another...

Understood.  Given that there''s a viable workaround, it''s not
a bother.
However, would it be possible to explicitly state in the DTrace Guide that
committing multiple speculations in the same clause is unsupported?

-- 
meem