07:22 <irker513> systemtap: mcermak systemtap.git:master * release-4.3-29-gf1a9bb064 / tapset/linux/networking.stp tapset/linux/scsi.stp testsuite/buildok/pretty.stp testsuite/semok/cast.stp testsuite/semok/pretty.stp testsuite/semok/sizeof.stp testsuite/semok/thirtyeight.stp testsuite/semok/thirtysix.stp testsuite/systemtap.base/bitfield.stp testsuite/systemtap.base/target_set.stp testsuite/systemtap.context/usymbols.exp: Tapset and testsuite updat

15:48 <linus2> fche: stap can't probe openmp code,right?

16:06 <fche> he linus2

16:07 <fche> did start looking at the problem this morning, and it turns out it's something related to another problem we've been tracking

16:07 <fche> so we'll work it out

16:07 <fche> the brief answer is yeah something is not quite right, likely a stap bug

16:09 <wcohen> fche, you suspect that is is related to https://sourceware.org/bugzilla/show_bug.cgi?id=25549 ?

16:09 <fche> yes.

16:56 <linus2> fche: good, look forward your result

20:12 <fche> hey wcohen are you ready for a tentative patch for this?

20:12 <fche> https://paste.centos.org/view/e3d8f439

20:16 <fche> some of it is just extra diagnostics

20:16 <fche> the key part is the relaxation in query_dwarf_func()

20:16 <wcohen> yes I saw the q->sess.verbose checks.

20:19 <fche> for that f30 ld lto binary, it gets us apprx. 30x the .statement() probe locations than before

20:20 <wcohen> fche, so the significant change is the conditional that determine whether to return DWARF_CB_OK?

20:21 <fche> 02223 decl_file != "" && // do not skip decl_file-free DIEs; could be artificial/LTO?

20:21 <fche> that line

20:21 <wcohen> that's the one I was looking at.

20:37 <wcohen> fche, just tried out the patch: https://paste.centos.org/view/28782947 about 15x the probe points with the patch for the lto code.

20:38 <fche> interestingly .function("*") probes match before & after, which I suspect is good

20:48 <wcohen> also tried the things out with the toy openmp hellow world https://paste.centos.org/view/a3886391 and it is showing statements for the #pragma omp parallel and printf.

21:31 <fche> yes, that was what I tested first.

21:31 <fche> ok, any concerns about merging this code?

21:31 <fche> I'll run some testsuite fragments here, but otherwise?