htop-0.5.2 dies an me when I let it run for some time. I have this behaviour on at least 5 x86 servers. The time to segfault varies. Sometimes 15 minutes are enough, but overnight (please don't argue about that being nonsense :-) I usually find it dead the next morning. All machines use +nptl. Investigating on alpha now. Reproducible: Sometimes Steps to Reproduce: 1. ssh into machine 2. start htop 3. wait Actual Results: htop segfaults Expected Results: htop keeps running x86, stable package versions only, except samba (3.0.20). 2.6.11 and 2.6.12 kernels with NPTL. Various RAM sizes from .5 to 2 Gig, uni- and dual-CPU machines.
Hm, I take a guess and hint to SMP. All affected machines have at least HyperThreading, and thus SMP support. I'll try with my old P3 and Athlons, and I don't remember htop dying on me there. More results to come.
What would be really useful here is a core dump and a backtrace. I suggest enabling debugging and seeing what function in htop is mis-handling the memory.
Well, the fun part is: when I build htop-0.5.2 with "debug", it bombs directly after calling with htop: ProcessList.c:579: ProcessList_scan: Assertion `cpuid == i - 1' failed. :)
Seriously. USE="debug" emerge htop gives me an htop-0.5.2 that does not work.
contacted the htop developer -- stay tuned...
Alright...
Hisham told me 0.5.4 will be released shortly which should fix those problems. So please stand by until it's released and added to the portage tree.
I find it mildly bewildering that this bug was closed weeks ago, and yet, we are still on 0.5.2 stable, which is definitely broken on *all* my SMP systems. This bug was closed but not resolved ...
have you tried the archmasked 0.5.4?
Since I use SMP systems for production and non-SMP for testing, no. I tend to draw a clear line between production and testing, so I didn't really want to. Guess I have to, hm?
$ echo '=sys-process/htop-0.5.4 ~arch' >>/etc/portage/package.keywords $ emerge -avt =sys-process/htop-0.5.4 What's the deal?
The deal was me not seeing that it's become ~x86 from hardmasked :) I usually don't bother with hardmasked things unless it's something where I'm personally committed the dig through the source. I'll report back.