Move to 1us resolution for latency measurements

[tycho]

A few fixes:

• The -z debug flag was implemented but not accepted by getopt_long. I added it.
• The -z debug flag only accepted a decimal argument, and since the argument is a bitfield, I made it possible to take a hex argument, i.e. 0xffff enables all debug bits.

And the most important part:

• Use 1us resolution for latency measurements. This fixes Change resolution to 1us rather than 10us #101.

[tycho]

Hm. Hold up on this for a bit. My SO_TIMESTAMPNS thing doesn't seem to work correctly yet.

[tycho]

Looks like SO_TIMESTAMPNS always uses CLOCK_REALTIME instead of CLOCK_MONOTONIC, so the numbers don't line up:

timestampns: 1538342854 527430214
clock_gettime: 5419 195062250
timestampns: 1538342855 528457016
clock_gettime: 5420 196056278

Bummer. CLOCK_MONOTONIC tends to have better quality. I'll make it use CLOCK_MONOTONIC only if the builder has opted out of using the SO_TIMESTAMPNS feature.

[luckman212]

@tycho Any time to check into the build errors so your PR can be merged?

[tycho]

Well, I assume you're referring to the Travis CI errors.

I've fixed those now:

• I was enqueuing the targets incorrectly. I used > instead of >= in ev_enqueue, which caused targets to be pushed to the front instead of back.
• The netdata test was broken because it used a regular expression for the result expecting a specific set of digits, but we're printing a higher resolution result now. I need to verify that it renders correctly in netdata now.

[coveralls]

Coverage decreased (-0.06%) to 79.337% when pulling ee76e49 on tycho:high-resolution-clock-sources into bc5017b on schweikert:develop.

[tycho]

The netdata output mode now appears to work properly. Helps if I tell it to use the right divisor.

[schweikert]

Hi Steven,

If you agree, I think we could move ahead with merging the code and target a fping 5.0 some times in the future. I will also create a branch for further 4.x releases. What do you think? Is the patch ready for merging?

[tycho]

Hi Steven,

If you agree, I think we could move ahead with merging the code and target a fping 5.0 some times in the future. I will also create a branch for further 4.x releases. What do you think? Is the patch ready for merging?

Yeah, I think it's ready to go.