FP: Route contains fewer than 2 points

[yakra]

This is truly weird. Observed in a benchmark run on bsdlab:

grep -rn 'Route contains fewer than 2 points:' *
siteupdate_896c-bsdlab-2t2p.log:7:alba albsh ERROR: Route contains fewer than 2 points: alb.a001 (0 total points)
siteupdate_896c-bsdlab-2t2p.log:36:1: Route contains fewer than 2 points: alb.a001 (0 total points)

alb.a001 is of course perfectly fine, and has 53 waypoints. args: -v -T 4 -C (-C doing nothing), with nmp_merged files turned off.

Commit: 896ca4a, which is a merge of 3282921 & 9076682. This happened during ReadWptThread, which is one of the earliest tasks siteupdate does. All the changes between 3282921 & 9076682 affected tasks done after this, except for minor changes to processing args, and Args.h has one less static variable. IOW, if this is a bug in siteupdate, it's got to be something already in production code.

Something as off the wall as this, which has been fine through many thousands of runs, I wanna say is a single-event upset. ;P I believe that siteupdate is valgrind-clean, with nothing untoward reported.

• [x] To make sure that is indeed the case, a valgrind run is near completion on lab2. Clean.
• [ ] For good measure, I'll start one on bsdlab too after posting this. Result: No invalid reads or writes (which would lead to such an error), but there were some memory leaks (which wouldn't). Forgot to compile with -g. Doing it again. Weird that we get this here but not in Linux.
• [x] ...and also leave siteupdate running in a loop for as many iterations as it takes until the bug is reproduced. ...12 times at once, in a tmux session. I have a feeling this could take a while. 5700 runs were all fine. Unsurprising. Aborted. May give it another go in the future.

[yakra]

Valgrind results are... weird. Nothing looks related to my code; nothing like the results from back in the day before I implemented proper memory management & class destructors. Weird stuff like:

• pthread management
• basic_streambuf ctor
• calls to ctime
• stack unwinding & throwing exceptions

O HAY, know what? I messed up my FreeBSD install a week or so ago, when trying to use pkg to upgrade or install... something. Something like

Newer FreeBSD version for package zstd:
To ignore this error set IGNORE_OSVERSION=yes
- package: 1302001
- running kernel: 1300139
Ignore the mismatch and continue? [y/N]:

and I was like "Sure, y not?"

Now, I frequently see stuff like ld-elf.so.1: /lib/libc.so.7: version FBSD_1.7 required by /usr/local/bin/emacs not found

Seeing if I can use freebsd-update to restore the system to good working order. If not, I'll just do a reinstall.

In the meantime, I'm just rerunning siteupdate_896c at 2 threads one more time to replace the offending siteupdate.log with one that runs to completion (and performs graph generation) instead of aborting. I'll just cross my fingers and home the benchmark data themselves aren't grossly thrown off by the glitches in the system; I don't want to spend an entire day re-running siteupdate. :P

[yakra]

freebsd-update helped somewhat. Emacs works again, as goes wget, which IIRC was also hosed before. Ran valgrind again, with the same results. :( /lib/libc.so.7 is dated May 28 15:43. Interestingly, this is one of the files in the stack trace in valgrind.log. Other so files have the same date & time, except for a valgrind-specific one. Does this mean something?

@jteresco, could I convince you to pkg install valgrind on noreaster?