dtrace discuss - Sep 2006 - USDT in C++, string problems

I can''t get probes with string/char* arguments to work in C++ and I
wonder if I''m missing something obvious... (NB running snv_40 at the
moment, so maybe I''m missing some recent fixes?)

If my probe definition uses ''string'' like this:

provider foo
{
	probe foo(string);
};

then the generated header for my probes refers to an undefined type
''string'':

/usr/sbin/dtrace -h -s foo.d -o foo_probes.h
/opt/SUNWspro/bin/CC -c foo.cxx -o foo.o
"foo_probes.h", line 18: Error: string is not defined.

if I define my probe with a char* argument instead, then everything builds fine
but when I run with the following script:

foo$target:::
{
	printf("%-15s: %-15s\n", probename, stringof(args[0]));
}

I get an error:

dtrace -s test.d -c ./bar
dtrace: script ''test.d'' matched 1 probe
Done.
dtrace: pid 2610 has exited
dtrace: error on enabled probe ID 1 (ID 51288: foo2610:bar:__1cDfoo6F_v_:foo):
invalid address (0x12000) in action #2

Note that the probe invocation looks like this:

    const char* hello = "hello";
    FOO_FOO((char*)hello);

So, I''m guessing that using string rather than char* is what I want to
do, but it looks as though the dtrace -h command isn''t handling string
correctly...?
 
 
This message posted from opensolaris.org

Matt Stupple stated:
< I can''t get probes with string/char* arguments to work in C++ and
I wonder if I''m missing something obvious... (NB running snv_40 at the
moment, so maybe I''m missing some recent fixes?)
< 
< If my probe definition uses ''string'' like this:
< 
< provider foo
< {
< 	probe foo(string);
< };
< 
< then the generated header for my probes refers to an undefined type
''string'':
< 
< /usr/sbin/dtrace -h -s foo.d -o foo_probes.h
< /opt/SUNWspro/bin/CC -c foo.cxx -o foo.o
< "foo_probes.h", line 18: Error: string is not defined.
< 
< if I define my probe with a char* argument instead, then everything builds
fine but when I run with the following script:
< 
< foo$target:::
< {
< 	printf("%-15s: %-15s\n", probename, stringof(args[0]));
< }

  You need to copy the string from the process/user space to kernel space
  with copyinstr().  See Chap. 33 on the dtrace guide:
http://docs.sun.com/app/docs/doc/817-6223/6mlkidlmd?a=view
  The link will show you why you''re getting the ''invalid
address'' error etc.

  Try:
foo$target:::
{
	printf("%-15s: %-15s\n", probename, stringof(copyinstr(arg0)));
}

  I''ve done something like this already with beer and python at:
http://blogs.sun.com/smg/entry/beer_python_and_stuff

  Where my provider is:
provider python {
        probe pythond__2(int, string);
};

   and I can use it like:
dtrace -o beer.out -n
''python$target:::pythond-2{@[stringof(copyinstr(arg1)),
arg0]=count();}'' -c ''../../python beer.py 200''

  Note the ''copyinstr'' there..

  Hope this help.s

Regards,
Sean.
.
  
< 
< I get an error:
< 
< dtrace -s test.d -c ./bar
< dtrace: script ''test.d'' matched 1 probe
< Done.
< dtrace: pid 2610 has exited
< dtrace: error on enabled probe ID 1 (ID 51288:
foo2610:bar:__1cDfoo6F_v_:foo): invalid address (0x12000) in action #2
< 
< Note that the probe invocation looks like this:
< 
<     const char* hello = "hello";
<     FOO_FOO((char*)hello);
< 
< So, I''m guessing that using string rather than char* is what I
want to do, but it looks as though the dtrace -h command isn''t handling
string correctly...?
<  
<  
< This message posted from opensolaris.org
< _______________________________________________
< dtrace-discuss mailing list
< dtrace-discuss at opensolaris.org

-- 
Sean.
.

> 
> You need to copy the string from the process/user
>  space to kernel space
>  with copyinstr().  See Chap. 33 on the dtrace guide:
> http://docs.sun.com/app/docs/doc/817-6223/6mlkidlmd?a=view
>
Magic - so I *was* missing something obvious, and all is working fine now.

Thanks,

Matt.

PS Am I right to think that the breakage in dtrace -h when using the
''string'' type rather than char* is a bug?
 
 
This message posted from opensolaris.org

On Fri, Sep 08, 2006 at 07:17:08AM -0700, Matt Stupple
wrote:
> PS Am I right to think that the breakage in dtrace -h when using the
> ''string'' type rather than char* is a bug?
Arguably, but the header file should really use the type ''char
*'' since
string (presumably) is just the D datatype. Regardless of how it handles it,
it could be a bit more graceful. Feel free to file a bug.

Adam

-- 
Adam Leventhal, Solaris Kernel Development       http://blogs.sun.com/ahl