fche changed the topic of #systemtap to: http://sourceware.org/systemtap; email systemtap@sourceware.org if answers here not timely, conversations may be logged
<fche>
serhei, and do I remember correctly that your tool associated these test cases with the previous relocation-related change that kerneltoast is about to revise?
<serhei>
which commit id was that change in?
<fche>
d1b554876c68f3
<serhei>
interesting
<serhei>
that's associated with a regression in a different set of testcases
<kerneltoast>
funny we caught this on the same day
* serhei
doesn't mind giving up the credit here
<serhei>
we have the buildbots but the scripting for making the results easily visible testcase-by-testcase hasn't started coming together until very recently
<serhei>
currently the primary product is a mondo 40MB of HTML tables that threatens to crash my firefox if I have too many copies open. Working on filtering that down a bit
<serhei>
the html I posted is 'only' 440kb, still a bit big for bugzilla attachment
<serhei>
I'm not sure how firefox's <table> rendering manages to translate MB's of HTML into GB's of swap space but it does ..
<serhei>
anyways
<kerneltoast>
probably the same way chromium translates a new tab into GB's of swap space
<serhei>
chromium doesn't even try to render my entire html, just displaying the 'sad tab' after a while
<kerneltoast>
serhei, speaking of testing, are the buildbots doing parallel testing yet?
<serhei>
no, they're not
<kerneltoast>
a lot of bugs were caught just running the stap testsuite in parallel
* serhei
finds the prior instances this was discussed in my transcript
<serhei>
ah, and removing superfluous testcases too. there were 64 (of a total 460) whose results don't seem to change over the time period I looked, some *may* be superfluous
<serhei>
hmm, parallel testing over one testsuite run, or multiple testsuite runs in parallel?
<kerneltoast>
multiple
<serhei>
ah
<kerneltoast>
i don't know of any bugs right now
* serhei
notices GDB testsuite supports parallel testsuite execution with -j, not sure we did
<kerneltoast>
i guess the low hanging fruit was fixed
<kerneltoast>
but who knows what's still out there
<tromey>
to make -j work for gdb, I took the Makefile bits from gcc, but also instrumented the gdb testsuite so I could use inotify to watch file creations, to find races where different tests wrote to the same files
<serhei>
thanks
<serhei>
hmm, attempting to run stap make installcheck with -j currently gives a ton of
<serhei>
AIL: abort: TEST 7: abort() in timer.profile (more concurrency and no globals): stdout: string should be "", but got "make[3]: Entering directory '/opt/stap-checkout/stap_build/testsuite'
<serhei>
so more crucial than file interference would be the makefile bits to keep testcases from seeing each other's make output in stdout
<serhei>
the parallel testing method with two parallel installcheck runs would just be a matter of adding e.g.
<serhei>
cd /usr/share/systemtap-testsuite && make installcheck &
<serhei>
to my stap testing script
<serhei>
probably with more output suppression
tromey has quit [Quit: ERC (IRC client for Emacs 27.1)]