Update for libatomic?

[andyj]

Rsyslog on TC 16 32-bit compile fails with an undefined reference to '__sync_fetch_and_add_8' when -march=i486 . The '__sync' functions have been deprecated for a while now in favor of '__atomic'. I mentioned this on an already open rsyslog issue, but it will be some time before it gets fixed. I think libatomic_ops could fix this, but with the current version loaded it still fails. The glib atomic functions have similar functions but without the '_8'. Updating -march to i686 from i486 allows it to compile, which brings up another question. What is the minimum architecture TC wants to support?

[Juanito]

The last time I had a similar problem libatomic solved things, but that might not always be the case.

We try to aim for i486 as the minimum architecture, but it's not always possible, in which case we should try for the next lowest - pentium-m or something like that?

[CNK]

We try to aim for i486 as the minimum architecture, but it's not always possible, in which case we should try for the next lowest - pentium-m or something like that?

Next lowest is Pentium AKA i586, not Pentium-M (which is beyond i686). I believe Slackware build packages for i586, for one.

[andyj]

i586 seems to work for the extensions that won't compile as i486, which right now are dhcp, rsyslog, and mariadb. I don't actually have that hardware, I'm doing all this in a VM on a 64-bit machine that is set up for booting different versions of TC. If VMware workstation has the ability to simulate specific CPU's I'm not aware of it.

[Juanito]

libatomic_ops updated in the x86 and x86_64 repos