On 03/12/2013, at 7:42 AM, srean wrote: > Will some automatic leak checker like valgrind, memprof help, when you try it > on the webserver while loading a page that triggers the leak. The page on > hunting leaks in Nde was quite instructive. Valgrind
( DYLD_LIBRARY_PATH=build/release/host/lib/rtl valgrind --dsymutil=yes build/release/host/bin/flx --test=b uild/release src/tools/webserver --port=1235 2>&1) ^CShell terminated by signal SIGINT ==1894== ==1894== HEAP SUMMARY: ==1894== in use at exit: 3,753,611 bytes in 57,310 blocks ==1894== total heap usage: 246,835 allocs, 189,525 frees, 19,623,040 bytes allocated ==1894== ==1894== LEAK SUMMARY: ==1894== definitely lost: 775,742 bytes in 17,165 blocks ==1894== indirectly lost: 2,214,036 bytes in 29,183 blocks ==1894== possibly lost: 2,365 bytes in 73 blocks ==1894== still reachable: 761,468 bytes in 10,889 blocks ==1894== suppressed: 0 bytes in 0 blocks ==1894== Rerun with --leak-check=full to see details of leaked memory ==1894== ==1894== For counts of detected and suppressed errors, rerun with: -v ==1894== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0) Not even sure if it's grinding webserver as opposed to flx. Ok, so running on webserver binary executable .. ==1902== Conditional jump or move depends on uninitialised value(s) ==1902== at 0x1002195D8: pthread_rwlock_init (in /usr/lib/libSystem.B.dylib) That's the OSX problem. There's a way to work around it. ==1902== ==1902== HEAP SUMMARY: ==1902== in use at exit: 205,147 bytes in 3,702 blocks ==1902== total heap usage: 12,066 allocs, 8,364 frees, 2,487,653 bytes allocated ==1902== ==1902== LEAK SUMMARY: ==1902== definitely lost: 12,360 bytes in 505 blocks ==1902== indirectly lost: 62,376 bytes in 1,320 blocks ==1902== possibly lost: 56,600 bytes in 1,140 blocks ==1902== still reachable: 73,811 bytes in 737 blocks ==1902== suppressed: 0 bytes in 0 blocks ==1902== Rerun with --leak-check=full to see details of leaked memory ==1902== ==1902== For counts of detected and suppressed errors, rerun with: -v ==1902== Use --track-origins=yes to see where uninitialised values come from ==1902== ERROR SUMMARY: 40 errors from 4 contexts (suppressed: 0 from 0) Basically it's too much work. Valgrind is just plain wrong in too many cases. For example, some of the "errors" it sees are a result of the GC scanning the stack. It's true some of the scanned addresses are uninitialised. It's a conservative collector with respect to the stack so this doesn't matter. Actually it's tricky on the x86_64 because the idiot API has a thing called a RED ZONE. That's a block of memory on the wrong side of the stack pointer that you're ALLOWED to put data in. Felix doesn't scan that as such. It saves the stack pointer with a hack, and machine registers in a jmpbuf using setjmp. Also I cannot stop the webserver properly. Ctrl-C doesn't kill it under valgrind until I try to load a page. So I'm not sure if FLX_FINALISE works for it (I think clean up on termination doesn't happen if killed by a Ctrl-C). On the server we now have: Memory usage: 32% PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 8006 felixweb 20 0 216m 16m 3136 S 0.0 3.4 0:14.79 webserver So the 3.4% is where I left it. It's not leaking when no one is using it at the moment. Still strings.fdoc is causing this in a few page loads: 8006 felixweb 20 0 258m 56m 3136 S 0.0 11.5 0:22.70 webserver So memory use up to 11.5% now. -- john skaller skal...@users.sourceforge.net http://felix-lang.org ------------------------------------------------------------------------------ Rapidly troubleshoot problems before they affect your business. Most IT organizations don't have a clear picture of how application performance affects their revenue. With AppDynamics, you get 100% visibility into your Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro! http://pubads.g.doubleclick.net/gampad/clk?id=84349351&iu=/4140/ostg.clktrk _______________________________________________ Felix-language mailing list Felix-language@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/felix-language
