This project https://github.com/elucideye/drishti is incorrectly classified as C in the github search. I noticed the language stats bar incorrectly reports a high C=87% to C++=10% ratio despite the more accurate per language file counts of C++=408 and C=5. What is the recommended fix in cases where the language counts are actually correct but not reflected in the statistics?


The language stats are correct. Linguist considers the "amount of code" when determining the statistics, not just the number of files.
In your case, you've got 13 MB of C files (well, it's mostly just src/lib/drishti/hci/gpu/VeraFont_16_2048.h) and 1.51MB of C++ files, which works out at approximate 87.6% to 10.1% of the entire repository size.
Removing this file, marking it as vendored or generated using a manual override will exclude it from the stats and give you more realistic looking stats.
I didn't see that file included when I looked at the 5 reported C files from the language search above.
The C file index here:
https://github.com/elucideye/drishti/search?l=c
Only lists the following 5 files:
C src/app/qt/facefilter/Device.h
C src/app/qt/facefilter/QVideoFrameScopeMap.h
C src/lib/drishti/core/padding.h
C src/lib/drishti/graphics/drishti_graphics.h
C src/app/qt/facefilter/GLVersion.h
I'm guessing that file index uses a different process. I'm currently adding emacs mode hints -*-c++-*- to all C++ files based on an older issue I saw, and will try the .gitattribute overrides if that doesn't work. Thanks for the help.
Your large file won't be included in the search results because it is too large: only files smaller than 384kb are indexed hence you don't see it.
https://help.github.com/articles/searching-code/ has more details on the the various search criteria GitHub uses.
To finish the thread... I left that generated font table as is (C) and added the linguist-generated tag in .gitattributes per your suggestion, and added -*-c++-*- to various files to force C++ labels in cases where it was appropriate. The .gitattributes file looks like this (per docs):
src/lib/drishti/hci/gpu/VeraFont_16_2048.h linguist-generated=true
The language stats updated immediately and are much closer to what I would expect:

Thanks again.
Most helpful comment
To finish the thread... I left that generated font table as is (C) and added the
linguist-generatedtag in.gitattributesper your suggestion, and added-*-c++-*-to various files to force C++ labels in cases where it was appropriate. The.gitattributesfile looks like this (per docs):The language stats updated immediately and are much closer to what I would expect:
Thanks again.