Re: toy parser committed to cvs

[William Cohen <wcohen at redhat dot com>]

Frank Ch. Eigler wrote:
> Hi -
>
> The beginnings of a possible parser for systemtap are now in
> CVS (module "src").  Some of the grammar is taken from awk,
> though the outer "probe" level wrapping is new.  It's fairly
> tame hand-written C++.
>
> I intend to start adding some semantic checking next, as a
> guideline to the elaboration/translation phases, plus some
> syntax to let the dtr-flavoured jprobes declarations through.
>
> - FChE

LL(1) parsers are good. They are much easier to debug than LR parsers 
generated by YACC. Meaningful error messages are also easier to produce. 
Is there any particular reason that tools such as antlr are not being 
used to generate the parser?

In dtr there were declarations for the original functions. What are you 
thinking about for the syntax for the jprobes systemtap probes? 
Explicitly listing the arguments passed into the function and hide this 
in the elaboration for common cases?

kernel:function("function_to_instrument"):
arguments(type1 arg1, type2 arg2,...)
{
	/* probe code */
}

As far as code generation, it looks like there would be some differences 
in the wrapper around the probe code depending on the instrumentation 
scheme, e.g. difference end of probe whether jprobe or kprobe.

-Will