PS: Just noticed, UWA ? small world :) I studied back at Curin Uni back in the 90's
------------------------------------------------------------------------------------------------------ Jason Richard Groothuis bSc(compSci) ------------------------------------------------------------------------------------------------------ > From: [email protected] > To: [email protected]; [email protected] > Date: Tue, 21 Apr 2015 17:39:36 +0930 > Subject: Re: [boinc_dev] Bug in the BOINC Wrapper > > Hi Kevin, While waiting for official answers (which may differ from what I > suggest here), barring any other problems yours may be a case similar to what > I see with seti@home applications when built against unmodified boincapi. > For the Windows case I'd suggest first try linking to the microsoft supplied > COMMODE.OBJ, documented > at:https://msdn.microsoft.com/en-us/library/9yky46tz.aspx, which should at > least enable the commit to disk feature under Visual studio toolchain. If > using another toolchain where that is not provided, opening the files with > the 'c' flag (in addition to the 'a, b or whatever) should enable commit, and > I'd guess might be available under other OSes as well (though require a tweak > to boincapi sources) > Note that the above I'd consider workarounds, for long standing issues with > process termination and threading, which runs afoul of modern threaded > C-Runtimes. These won't fix the aggressive exit handling strategies > employed, but at least attempt to bypass the intentional OS (desktop > optimisation and power management related) delays that can be from fractions > of a second to 10's of second or more under high contention. > In our case (as far as we can tell) our state/checkpoints and other files > are mostly/usually correct, though some suspicious behaviour is observed from > time to time. More dominant for us are truncated/missing stderr.txt content > (making debugging difficult) which is demonstrably tied in with Boincapi's > exit handling. > ------------------------------------------------------------------------------------------------------ > Jason Richard Groothuis > bSc(compSci) > > ------------------------------------------------------------------------------------------------------ > > > > From: [email protected] > > To: [email protected] > > Date: Tue, 21 Apr 2015 14:32:59 +0800 > > Subject: Re: [boinc_dev] Bug in the BOINC Wrapper > > > > And we’re having the same issue with the PIE Android version which we > > compiled from the BOINC source > > > > Regards > > Kevin > > > > > > > > On 21 Apr 2015, at 14:19, Kevin VINSEN > > <[email protected]<mailto:[email protected]>> wrote: > > > > No - I’m wrong… > > > > n will be 3 which is != 2 so it will continue on... > > > > We are having a lot of problems with Windows 7 Clients not reading the > > check point file. We can see them writing to wrapper_checkpoint_text, but > > they fail to read it correctly. > > > > Regards > > Kevin > > > > > > > > On 21 Apr 2015, at 13:34, Kevin VINSEN > > <[email protected]<mailto:[email protected]>> wrote: > > > > I think I’ve found a bug in the BOINC wrapper for checkpointing > > > > Line 973 of wrapper.c is the read_checkpoint function. > > > > > > int read_checkpoint(int& ntasks_completed, double& cpu, double& rt) { > > int nt; > > double c, r; > > > > ntasks_completed = 0; > > cpu = 0; > > FILE* f = fopen(CHECKPOINT_FILENAME, "r"); > > if (!f) return ERR_FOPEN; > > int n = fscanf(f, "%d %lf %lf", &nt, &c, &r); > > fclose(f); > > if (n != 2) return 0; > > ntasks_completed = nt; > > cpu = c; > > rt = r; > > return 0; > > } > > > > The fscanf is scanning 3 parameters not 2. > > > > Regards > > Kevin > > > > > > > > > > > > _______________________________________________ > > boinc_dev mailing list > > [email protected] > > http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev > > To unsubscribe, visit the above URL and > > (near bottom of page) enter your email address. > > _______________________________________________ > boinc_dev mailing list > [email protected] > http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev > To unsubscribe, visit the above URL and > (near bottom of page) enter your email address. _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
