Add g++-5 to ci

[ckormanyos]

No description provided.

[ckormanyos]

This first naive try to scope out GCC 5 on bionic may not work because not all the standards of the matrix are available.

[ckormanyos]

The first naive try is probably destined to fail since GCC 5 does supports neither -std=c++14 nor -std=c++17 in the test matrix.

[ckormanyos]

So this ran surprisingly well --- not perfect, but well. I think all the dedicated, consistent work on going standalone really paid off.

So... anyway, In a dedicated run, I added the basic Multiprecision g++-5 tests as shown here.

I had expected quite a few failures, but there was one single test failure in cpp_int_tests regarding a floating-point-exception. I can probably track down the root cause locally.

I don't know how to guage the excitement on getting g++-5 back running in CI. This can definitely wait for 1.80 (post 1.79), since there is a lot going on.

But it does, in fact, seem feasible to go for g++-5.

Thoughts?

Cc: @jzmaddock and @mborland and @NAThompson

[jzmaddock]

If it's straightforward then by all means go for it.

[ckormanyos]

If it's straightforward then by all means go for it.

While underway, I found a TODO for C++11.

I took a try at reworking the timer facility based on a TODO note I encountered. I opted for a C++11 design which results in both a stopwatch facility as well as the original interface to a class called timer. Both of these may be useful.

I am a bit unsure of the new design I made.

If anyone gets a chance (no hurry though), ... could you please taka a look at this? (Because <chrono> tends to be a bit of a wily one somethines...).

Cc: @jzmaddock and @mborland

[ckormanyos]

I think I found something. This might turn into a longer story. So I'll get this communication started. If I'm wrong please correct this.

I found that test_cpp_rational.cpp was generally running slowly (on PC locally) and also crashing in CI on g++-5 in GHA.

So I found this line. The numeric_limits::digits will give zero when compiling with __GNUC__ and using -std=c++11 (or 14, 17, etc.). So the number of random digits requested underflows and explodes.

Please note GHA tests are using something like -std=c++11 and not a GNU extension.

So in this PR, I have:

• Repaired test_cpp_rational.cpp. Use <random> and generally C++11-ify it.
• Completely reworked the timer facility to be C++11-ified.
• Add g++-5 tests to CI's GHA jobs.

I am not getting through the traffic jamon CI.

But I believe the underflow on digits I mention above might actually be a real factor that we might want to address for 1.79 itself.

@jzmaddock and @mborland please do comment on this or correct my thinking if I'm off on a tangent here...

Many thanks, guys.

[jzmaddock]

So I found this line. The numeric_limits::digits will give zero when compiling with GNUC and using -std=c++11 (or 14, 17, etc.). So the number of random digits requested underflows and explodes.

Ah :( Yes, double_limb_type is __int128 and numeric_limits is not specialized unless you compile with --gnu++XX, I'll check over the diff shortly, this is an easy one to get wrong and mess up as you've just discovered! :(

Good catch that! If Github had a big thumbs up emoji imagine it here :)

[jzmaddock]

Diff looks good to me.

[ckormanyos]

Diff looks good to me.

Thanks John (@jzmaddock)

CI on GHA is moving in the green direction. There was a hicup on drone.

I could [ci skip] a trivial commit to restart it or (when/if) GHA runs green. Or I could also simply merge to develop. But before deciding that, ... Could I get a second opinion on this.

To me, it looks like spurious runner problems on drone. What's your take?

[jzmaddock]

Yep spurious: couldn't connect to ubuntu.com for some reason.

[ckormanyos]

Yep spurious: couldn't connect to ubuntu.com for some reason.

Thanks John. So when I merge this, Multiprecision will officially support GCC5 on -std=c++11and -std=c++14. I activated a subset of the key tests (but the most important ones) on CI. So I think it's fair to say, after this PR, we have GCC5 again.

This will get in for 1.79.

I do have interest in finding out how it looks for 4.8. But i will not press or force this issue in February. I'll take a cursory glance, however, in a future PR and we can see if it's feasible.

Cc: @jzmaddock and @mborland and @NAThompson

[ckormanyos]

Failures in CI are:

• Spurious failure of drone
• Two server communication errors in GHA

Merging to develop