travis: Add hlint to Travis-CI checks

[rbasso]

• Add hlint tests to the entire repository.
• Add and fix comments to match that.

Related to #355.

Motivation

It was briefly discussed in #355 that it would be interesting to run HLint in Travis-CI.

Thanks to @petertseng, we discovered that hlint is already available and whitelisted in Travis, but the version available in Ubuntu Precise is older than the ones we would like to run.

Initial experiments showed us that, installing it from source, the initial building times could be problematic. Another problem would be to assure that we are running a known version of HLint.

I opened this PR as an experiment, to see - after we finish #355 - if it is feasible to install and run it in "production". It also serves to check if we are really done with fixing the current code base.

The general idea is to install hlint using the same Stack snapshot used for testing the exercises, increasing Stack cache sharing while using a HLint version known to compile without problems.

I added HLint tests between configlet and the exercises' test. This way we can get feedback really fast if we push something without running hlint before.

Preliminary results

Installing hlint but not running it:

• Initial building time: 27m31s
• Rebuilding time: 1m50s
• Approximate cache size per snapshot: 700MB

I had problems in my first tests, but I was not able to study the logs.
As soon as I get better Internet connection I'll go back to it.

I think we'll be able to put that in production soon @petertseng. Thanks again!

[rbasso]

@petertseng, what do you think about this PR?

• Should we run hlint before the exercises, before each of the exercises, after each exercise or after all the exercises?
• Do you agree with using the hlint from the same Stack snapshot (for the current snapshot I think it is v1.9.35)?
• Do you think those initial build times are acceptable?

[rbasso]

Additional tests

I just run some more tests clearing the caches in my fork and temporarily disabling the error code from hlint.

Branch: master
    Building time (without cache)
        nightly-2016-07-17 :  14m12s
        nightly            :  14m36s

    Rebuild time (with cache)
        nightly-2016-07-17 :   1m26s
        nightly            :   1m50s

    Total cache size       : 1346.28MB

Branch: travis-hlint-no-exit-code

    Building time (without cache)
        nightly-2016-07-17 :  21m03s
        nightly            :  21m09s

    Rebuild time (with cache)
        nightly-2016-07-17 :   1m41s
        nightly            :   2m03s

    Total cache size       : 1384.49MB

Mean variation:
    Building time          : + 6m42s
    Rebuild time           : + 0m14s
    Tocal cache size       : + 38.21MB

It seems that the only relevant cost in installing and running hlint is the initial build time.

Considering that we don't usually clear the cache, those 7 additional minutes don't look so bad to me. I think we can still follow the plan of running it on Travis.

If in the future those 7 minutes prove to be critical, we can easily revert the change. 😄

[petertseng]

Do you think those initial build times are acceptable?
Considering that we don't usually clear the cache, those 7 additional minutes don't look so bad to me. I think we can still follow the plan of running it on Travis.

Yes, since it is cached we should be OK!

Do you agree with using the hlint from the same Stack snapshot (for the current snapshot I think it is v1.9.35)?

It is best if we do this, since it is best to be able to get the same result locally as that Travis will give us. So I support using Stack.

Should we run hlint before the exercises, before each of the exercises, after each exercise or after all the exercises?

I remember a related conversation at some point as to whether we want to run configlet first or last, but I don't have the conversation handy right now nor can I search for it. We wanted to fail fast then, right? hlint runs pretty fast, and faster than the individual tests, I think. That seems to support running it before. I don't actually feel strongly though.

[rbasso]

Now that we finished #355, I think we are ready to add hlint to Travis.

Everything seems fine in Travis-CI: https://travis-ci.org/exercism/xhaskell/builds/163280543

[rbasso]

Added commit updating README.md to explain that hlint now runs on Travis-CI.

[petertseng]

yup, looks good. One small change (I can change it when I have the time, but not right now)

[petertseng]

suppress

[rbasso]

Fixed!

[petertseng]

Should this [Travis-CI] have some (URL) after it so that it becomes a link?

[rbasso]

Ops! forgot to remove the brackets.

Should we add a links to https://travis-ci.org/exercism/xhaskell when mentioning Travis? I think we have two references to it.

[petertseng]

We do have the build status image, so it wouldn't strictly be necessary. You're right that we mention Travis multiple times in this file. If we want to link it I suggest we do something to only have to link to it once:

[e 1][ex]
[e 2][ex]

[ex]: http://example.com

e 1
e 2

[rbasso]

Created a PR removing the brackets to fix the problem.

I think that the README deserves a rewrite, but I'm without ideas for now.