tsx disambiguation

[mking]

Here is an example where a TypeScript .tsx file is incorrectly recognized as tiled map format.

One way to solve this is to add a disambiguation heuristic for tsx (similar to the one used for ts).

    disambiguate ".tsx" do |data|
      if data.include?("<tile")
        Language["XML"]
      else
        Language["TypeScript"]
      end
    end

What do you think?

[pchaigno]

The heuristic for .ts files is a particular case because .ts XML files must always contain a TS tag.
That does not seem to always be the case for .tsx files. As such, I'm not sure there are enough mis-classifications of .tsx files to justify a new heuristic...

If you think this is a recurrent classification mistake, we could evaluate the number of files incorrectly classified for a sample?
Otherwise, you can use Linguist overrides to fix this issue on a per repository basis.

[mking]

Ah ok. So, I did some additional research into the .tsx XML format and into the number of misclassified files:

• Every .tsx file has an outer <tileset> element. So I think if we use <tileset we will catch every valid .tsx XML file and will only mistakenly classify React files if they have a <tileset> element. This should be rare since the React convention for non-builtin elements is to start with an uppercase letter. Also, note that <tileset (with space) should not exclude any valid files because the dtd requires at least one attribute (firstgid).
• GitHub search brings up 322 misclassified .tsx TypeScript files. (Every tsx file will have an import React at the top.)
• In total, I see 1893 tsx TypeScript files, so it seems like the number of misclassified files is not insignificant.

[arfon]

Thanks for documenting this @mking. We'd welcome a Pull Request implementing this heuristic. Alternatively the Linguist overrides (https://github.com/github/linguist#overrides) are probably the best solution for now.

[dsifford]

@arfon I've tried now using every glob combination I can think of to get .tsx to register as TypeScript in .gitattributes, but it simply will not.

Am I missing something? https://github.com/dsifford/academic-bloggers-toolkit/blob/master/.gitattributes

[arfon]

@dsifford - can you tell me what you're not seeing? These are the language stats I see for your repository:

[dsifford]

@arfon Sorry, perhaps I had the wrong idea on how deeply linguist goes in github..

My concern was with this:

Specifically, the .tsx files all are seemingly registered as XML files.

[arfon]

@arfon I've tried now using every glob combination I can think of to get .tsx to register as TypeScript in .gitattributes, but it simply will not.

I'm afraid .gitattributes data only overrides the repository classification. They're currently not supported by the search infrastructure or the syntax-highlighting of files sorry.

[dsifford]

Ah, understood. No problem at all. Thanks for the clarification 👍