Include cmath explicitly.

[mahge]

This is a "retry" of #1097. To see if there is some transient issue with the PR itself.

Purpose

cmath is not included directly.

• The files use std::isnan and std::isinf from cmath.
• It breaks on old compilers (e.g. gcc-5) because the header might not be included indirectly.

Approach

Include cmath explicitly.

[lochel]

It keeps failing unfortunately. @adrpo do you have an idea how to get things back running again?

[adrpo]

Looking at ps aux | grep hudson on ripper2 seems this one is stuck:

/usr/bin/perl -w ../rtest -v -nolib tlm3d.lua
/var/lib/jenkins1/ws/OMSimulator_PR-1100//install/linux/bin//OMSimulator tlm3d.lua

I'll gdb it to see where is stuck.

[lochel]

Okay, otherwise we might have to disable that particular tlm test?

[adrpo]

That test is using 20 terrabytes of memory?!?!!?! Geez :)

1361045 hudson 20 0 20.0t 1.0g 8480 S 159.5 1.6 20:55.28 /var/lib/jenkins/ws/OMSimulator_PR-1100//install/linux/bin//OMSimulator tlm3d.lua

I can't gdb to it, it won't let me. Maybe one could run valgrind on it to see where it goes totally wrong.

[adrpo]

I killed the test as it used all the memory from that host.

[sjoelund]

I wonder if it's related to the tests hanging on the library coverage runs.

Also, is there no ulimit for virtual memory?

[adrpo]

I think is the asan version of the test as far as I can see.

[adrpo]

No idea about the limits, but it seems no.

[adrpo]

After I added ulimit -v to 6GB a lot of tests are failing!?

[adrpo]

It really should not need that much memory I guess.

[adrpo]

Any idea why this happens? Is asan using that much memory?

[mahge]

The issue has been "worked-around" for now by disabling address sanitizer tests by default with #1106. So the original PR #1097 is now merged.

This can be closed or kept open until the issue is resolved for good.