Can't access argument of process marker

Can't access argument of process marker

[Wenji Huang]

Hi,

I got the failure when running test semok/thirtysix.stp. The error
is triggered by the following script,

$ stap -e 'probe process("stap").mark("*")? { 
println(@defined($arg1)?$arg1:$nosuchvar) }'
semantic error: invalid argument number: identifier '$nosuchvar' at 
<input>:1:66
         source: probe process("stap").mark("*")? { 
println(@defined($arg1)?$arg1:$nosuchvar) }
 
           ^
Pass 2: analysis failed.  Try again with another '--vp 01' option.

It seems arguments of marker aren't accessible. Moreover, I think
it's not very accurate to complain "invalid argument number". In fact, 
the variable doesn't follow the naming rule, it's better to make
a little modification and give some hints like $argN.


Regards,
Wenji

Re: Can't access argument of process marker

[Josh Stone]

On 06/24/2010 11:04 PM, Wenji Huang wrote:
> Hi,
> 
> I got the failure when running test semok/thirtysix.stp. The error
> is triggered by the following script,
> 
> $ stap -e 'probe process("stap").mark("*")? { 
> println(@defined($arg1)?$arg1:$nosuchvar) }'
> semantic error: invalid argument number: identifier '$nosuchvar' at 
> <input>:1:66
>          source: probe process("stap").mark("*")? { 
> println(@defined($arg1)?$arg1:$nosuchvar) }
>  
>            ^
> Pass 2: analysis failed.  Try again with another '--vp 01' option.
> 
> It seems arguments of marker aren't accessible.

I saw this too, and I think it's just a failed operation of @defined for
SDT.  There are two overrides:

sdt_kprobe_var_expanding_visitor::visit_defined_op - just calls the
parent class... harmless, but unnecessary.

sdt_uprobe_var_expanding_visitor::visit_defined_op - leaves the
defined_op untouched, without checking its argument.  This means later
it expands to "undefined" --> 0.

I believe that neither class really needs to customize visit_defined_op,
but I haven't had a chance to try removing them yet.  Stan, do you
recall why these were needed?

> Moreover, I think it's not very accurate to complain "invalid
> argument number". In fact, the variable doesn't follow the naming
> rule, it's better to make a little modification and give some hints
> like $argN.

I agree, and there's also a comment that it needs to check for the $arg
prefix in the first place.


Josh

Re: Can't access argument of process marker

[Frank Ch. Eigler]

Josh Stone <jistone@redhat.com> writes:

> [...]
> sdt_kprobe_var_expanding_visitor::visit_defined_op - just calls the
> parent class... harmless, but unnecessary.
>
> sdt_uprobe_var_expanding_visitor::visit_defined_op - leaves the
> defined_op untouched, without checking its argument.  This means later
> it expands to "undefined" --> 0.
> [...]

One piece of background that may help explain this is that some of the
sdt.h options did not encode the actual number of arguments into the
.probe metadata, so stap couldn't actually tell how many arguments
there are supposed to be.  Perhaps this is why it was thought
appropriate to short-circuit @defined() there.

In any case, this shortage of metadata is fixed with SDT_V2, which
stap usually forces on for itself.  The translator-side code may not
have caught up.

- FChE