Misclassified C as C++ in data.table project

[jangorecki]

This repository does not use any C++ code but some header files are classified as such:
https://github.com/Rdatatable/data.table
It is strange that h files are not considered as C unless C++ syntax is detected inside, I think that should be the default option.
I would appreciate fixing it.

[smola]

@jangorecki This is a known problem: #1626

You can workaround it by overriding the language of *.h files in the repository with .gitattributes: https://github.com/github/linguist#overrides

[jangorecki]

I think it should work other way around. *.h files classified as C but actually being C++, those should require workarounds.

[smola]

@jangorecki I agree. But there's some technical limitations that need to be solved before that's possible. There's a lengthy discussion about it at #1626. Short-term, overriding with .gitattributes is the only solution.

[stale]

This issue has been automatically marked as stale because it has not had activity in a long time. If this issue is still relevant and should remain open, please reply with a short explanation (e.g. "I have checked the code and this issue is still relevant because ___."). Thank you for your contributions.

[jangorecki]

Issue is still relevant and mentioned technical limitations doesn't seem to be addressed.

[Starwort]

Also occurring in my repository:
https://github.com/Starwort/NEA/search?l=c%2B%2B
This seems very common...

[jangorecki]

Yes, h files should be recognized as C till the proper fix will be implemented.

[smola]

(EDITED): I was writing at the wrong issue.

[0hana]

I also have this issue
https://github.com/0hana/C/test.graph.h

[lildude]

I'm closing this issue as it's essentially a duplicate of #1626 which I have now pinned to try and reduce the duplicates.

[jangorecki]

@lildude so till the proper solution is in place maybe we can just change classification of *.h files to be C rather than C++? that would resolve this particular issue which is not exactly the same as #1626.

[lildude]

It's not that simple. .h is already classified as C:

https://github.com/github/linguist/blob/8011321f1394b5c3fc1f8ef242775af640df8f7d/lib/linguist/languages.yml#L614-L620

However, it is also classified as C++:

https://github.com/github/linguist/blob/8011321f1394b5c3fc1f8ef242775af640df8f7d/lib/linguist/languages.yml#L646-L661

... and Objective-C:

https://github.com/github/linguist/blob/8011321f1394b5c3fc1f8ef242775af640df8f7d/lib/linguist/languages.yml#L3847-L3857

... because it's valid in all three. Simply stripping from all but C will cause even more misclassification than currently seen with the current heuristics and classifier.

Where things go wrong is when a file doesn't match any of the heuristics:

https://github.com/github/linguist/blob/aad49acc0624c70d654a8dce447887dbbc713c7a/lib/linguist/heuristics.yml#L226-L231

So it falls through to the classifier which then has to make a guess based on the content of the samples it has. This is where things go wrong most of the time, especially with very small files.

What we could do, which may be the best solution, is to completely remove the classifier step for .h files, which is effectively what #1626 (comment) is suggesting.

I've started a PR at #5357 to implement this.