Added HAProxy configuration language

[NickMRamirez]

Adding a new language for HAProxy configuration files, e.g. haproxy.cfg. The ".cfg" file extension is already used by the INI language, but I have added samples to both sample folders to differentiate them.

Checklist:

• I am associating a language with a new file extension.

  • The new extension is used in hundreds of repositories on GitHub.com
    • Search results for each extension:
  • I have included a real-world usage sample for all extensions added in this PR:
    • Sample source(s):
      • [URL to each sample source, if applicable]
    • Sample license(s):
  • I have included a change to the heuristics to distinguish my language from others using the same extension.

• [X ] I am adding a new language.

  • [X ] The extension of the new language is used in hundreds of repositories on GitHub.com.

    • Search results for each extension:

    Search for haproxy.cfg filename:
    https://github.com/search?q=filename%3Ahaproxy.cfg&type=Code
    Search for .cfg that is not haproxy.cfg:
    https://github.com/search?q=roundrobin+extension%3Acfg+-filename%3Ahaproxy.cfg

  • [X ] I have included a real-world usage sample for all extensions added in this PR:

    • Sample source(s):
      • https://github.com/NickMRamirez/linguist/tree/haproxy/samples/HAProxy
      • https://github.com/NickMRamirez/linguist/blob/haproxy/samples/INI/ms.cfg
    • Sample license(s):

  • [X ] I have included a syntax highlighting grammar: https://github.com/abulimov/atom-language-haproxy

  • I have updated the heuristics to distinguish my language from others using the same extension.

• I am fixing a misclassified language

  • I have included a new sample for the misclassified language:
    • Sample source(s):
      • [URL to each sample source, if applicable]
    • Sample license(s):
  • I have included a change to the heuristics to distinguish my language from others using the same extension.

• I am changing the source of a syntax highlighting grammar

  • Old: https://github-lightshow.herokuapp.com/
  • New: https://github-lightshow.herokuapp.com/

• I am updating a grammar submodule

• I am adding new or changing current functionality

  • I have added or updated the tests for the new or changed functionality.

[pchaigno]

Thanks for the contribution!

I've left a few comments inline.

[pchaigno]

Here is an overview of how syntax highlighted files will appear on GitHub.com with the new grammar: Lightshow link.

[pchaigno]

@NickMRamirez Could you add one or two other sample files? I'm still seeing a lot of misclassifications by the Bayesian classifier between HAProxy and INI files. I'm expecting that it should be able to do a better job with more training data since HAProxy files contain a set of easily identifiable keywords. Files of the same size as the one you created, or a bit larger should be fine; try to avoid huge files, as the Bayesian then keeps all tokens in memory.

[NickMRamirez]

Added more sample files. Let me know if I should add more.

[pchaigno]

I've downloaded 2347 files from your search query and ran Linguist on it. As far as I can see, it detects all HAProxy files perfectly. I'd say we're good to go!

[NickMRamirez]

Great to hear! Thanks for your work on this!

[lildude]

@pchaigno regarding:

I'm still seeing a lot of misclassifications by the Bayesian classifier between HAProxy and INI files. I'm expecting that it should be able to do a better job with more training data since HAProxy files contain a set of easily identifiable keywords.

Shouldn't we be using a heuristic first? This PR doesn't have a heuristic but introduces a duplicate extension.

[pchaigno]

@lildude Adding more samples was enough to improve the Bayesian classifier's accuracy significantly. We went from 1 to 4 samples and now don't have miss-classifications anymore (in the 2347 files corpus I downloaded). In the past, we've kept heuristic rules for cases where the Bayesian classifier is unable to identify recurring keywords. Here, we have a set of recurring keywords (frontend, backend, server, listen, etc.) which the classifier can easily identify as long as we have enough samples.

[lildude]

@pchaigno 👍 Thanks for the explanation and validation work. 🙇

[lildude]

LGTM. Welcome to Linguist and thanks for your contribution.