Re: [DynInst_API:] Dyninst limitations for memory trace

[Josh Stone <jistone@xxxxxxxxxx>]

On 10/21/2012 10:24 PM, Legendre, Matthew P. wrote:
> I believe the "Unable to find function: dyn_pthread_self" error is
> frequently from a misset DYNINSTAPI_RT_LIB environment variable.  Make
> sure that's pointed to the full path to libdyninstAPI_RT.so.  A common
> mistake would be pointing it to libdyninstAPI.so.

On that note, I found it pretty cumbersome that this environment
variable needs to be set all the time, so I figured out a trick that you
may like to get it dynamically.  See guess_dyninst_rt() here:

http://sourceware.org/git/gitweb.cgi?p=systemtap.git;a=blob;f=stapdyn/dynutil.cxx;h=393a14231e35baf075149cef03076b799551c6e0;hb=HEAD#l30

That won't work for cross-arch mutatees (i.e. needing RT_m32), but
otherwise it does the job well, since native is most common for us.
Even if we could figure out m32, you're still stuck if you want a mix of
64 & 32-bit mutatees at once, since there's only one environment
variable to set!  (Hmm, can dyninst handle a "/path1:/path2" env?)


Josh