loc counts lines of code in files, reporting blank, comment, and code line counts broken down by language.
There are many other tools like it. What makes loc different is that it does not decide which files to read; instead you pass it a list of files through stdin, and it returns statistics about those files. You can use tools like find or git ls-files to produce a list of files to pass to loc, depending on what you want to count.
loc is written in Rust. Clone the repository and then run cargo install --path . to build and install it.
Example invocations:
git ls-files | loc
find src/ -name '*.rs' | loc
fd --extension py | loc
Example output:
Language Files Bytes Lines Blanks Comments Code
TypeScript 13 76.7 KB 2,819 325 62 2,432
Rust 3 32.5 KB 1,019 146 39 834
CSS 1 10.0 KB 562 86 10 466
TOML 1 801 28 2 0 26
Total 18 119.9 KB 4,428 559 111 3,758
loc reads supported files line by line, and classifies each line as exactly one of code, blanks, or comments.
- Blank lines are those that are empty or contain only whitespace
- Comment lines are those that only contain comments (and whitespace)
- Code lines are those that contain any code (even if they also contain comments)
As it reads the file, loc keeps track of whether it is currently in a block comment or a string literal. While inside a block comment, non-blank lines are counted as comment lines. While in a multiline string, non-blank lines are counted as code. Block comment start markers are also ignored while inside strings, to avoid things like glob("src/*.js") being misidentified as the start of a block comment (which would then likely cause the remainder of the file to be miscounted as comment lines).
loc works by scanning for simple substrings (like // or /*); it does not parse code into a proper AST. This means there are edge cases where loc will misclassify lines.
See languages.toml (or run loc --languages) for a list of supported languages.
Language detection is based on file extension or filename (for things like Makefile). If that fails, loc checks if it has a #! line and tries to detect the language based on that. Files whose language can't be identified are silently skipped.
It's pretty fast. loc uses memchr to scan through files, avoids most allocations, and processes files in parallel.
On my computer, loc counts all of VSCode's source code (~2.1M lines, mainly TypeScript) in 60 milliseconds.
This code is available under the ISC license; see LICENSE file for details.
