Test repeat in 10 sec: increase check frequency tenfold

[rht]

No description provided.

[jbenet]

@rht thanks!

[rht]

Ok will at least replace with test_seq.

The tests in sharness are meant to be passed by the node & python implementation, right?
The granular check didn't cut the test time that much, so now I am in favor for integer sleep (and so no boilerplate dependency for other implementation). Unless absolutely necessary.

[rht]

I wonder if there is a checker for sh for POSIX compliance.

[chriscool]

In Git test_seq uses Perl but we don't want to depend on Perl I think.

[rht]

e.g. debian posh

[chriscool]

Unfortunately it looks like http://www.shellcheck.net/ doesn't check for POSIX compliance.

[rht]

Also shellcheck is less portable than perl (no binary build in https://github.com/koalaman/shellcheck).

I think any of posh, dash, bsh is sufficient to check for POSIX compliance.
With dash being the most available everywhere (I can't find posh in yum/rpm/homebrew. I didn't check for bsh though).

[jbenet]

The best way if we really want to sleep for less than a second might be to write or find a go program that sleeps for the specified amount of time. It might use: http://golang.org/pkg/time/#Sleep

Let's do this. The fractional seconds add up. I'd like to have our tests wait as little time as we can make them.

[rht]

What about perl -MTime::HiRes -e 'sleep(0.1)' + test_seq?
Since sharness already assumes that perl exists.
And 0.1s is an order of magnitude longer than the load time concern in http://stackoverflow.com/questions/7757814/how-to-sleep-a-random-fraction-of-a-second-in-posix-shell.

And what about reusing the test for other lang implementation?

[chriscool]

Where does sharness assumes that perl exists?

Greping for perl, I found only:

test/3nodetest/bin/save_logs.sh:STRIP="perl -pe 's/\e\[?.*?[\@-~]//g'"

which is not part of sharness.

[rht]

Ah right, it is git test that uses perl, while sharness is pure shell.

Hence any util that is not fully portable is to be written in go?
Including seq (and curl)?

[chriscool]

@rht yeah we often write some small utils in go, when we need them and there is no easy way to have them in POSIX shell.
About seq we can implement test_seq in POSIX shell using expr.
About curl, for now we accept that we depend on it, but the long term goal is to replace it with a go binary.

[chriscool]

See PR #1329 for test_seq in POSIX shell using expr.

[chriscool]

About sleeping for some milliseconds there is now, shamelessly stolen from go-random:

https://github.com/chriscool/go-sleep

[jbenet]

@chriscool 👍

[chriscool]

By the way after sleeping a bit on it myself :-) I think I will rename it soon like "go-millisleep".

[jbenet]

haha, i like it. (maybe can use seconds but allow fractions. (i.e. parse it as a time.Duration))

[rht]

Which one is usually preferable: sleep 0.1, gosleep 0.1 (like golint), or go sleep 0.1, or go-sleep/go sleep (as in git-clone/git clone)?

[jbenet]

I'd say:

sleep 0.1

But this may clash with other tools so another name is ok. Don't use "go sleep" as that should only be a tool part of the go distribution 

[cryptix]

Like this a lot.

I opened PR chriscool/go-sleep#1 to use time.ParseDuration() instead. go-sleep 2s is much easier to approach for new people IMHO.

[rht]

go sleep 0.1!# This is all for the want of the machines to sleep less and less. For the same reason we want an ever faster in-time compiler.

[chriscool]

I merged @cryptix's PR so now it is sleep 100ms, but I think I will rename the command to go-sleep.

[chriscool]

I renamed it go-sleep as it is the most logical given the name of the repo.