cloc counts blank lines, comment lines, and physical lines of source code in many programming languages.
Under GNU General Public License v2.0
By AlDanial

programming-language cloc count-lines


Count Lines of Code

cloc counts blank lines, comment lines, and physical lines of source code in many programming languages.

Latest release: v1.90 (May 1, 2021)

cloc moved to GitHub in September 2015 after being hosted
at since August 2006.

Quick Start ▲

Step 1: Download cloc (several methods, see below) or run cloc's
docker image. The Windows executable has no requirements.
The source version of cloc requires a Perl interpreter, and the
Docker version of cloc requires a Docker installation.

Step 2: Open a terminal (cmd.exe on Windows).

Step 3: Invoke cloc to count your source files, directories, archives,
or git commits.
The executable name differs depending on whether you use the
development source version (cloc), source for a
released version ( or a Windows executable
(cloc-1.90.exe). On this page, cloc is the generic term
used to refer to any of these.

a file

prompt> cloc hello.c
1 text file.
1 unique file.
0 files ignored. v 1.65 T=0.04 s (28.3 files/s, 340.0 lines/s)
Language files blank comment code
C 1 0 7 5

a directory

prompt> cloc gcc-5.2.0/gcc/c
16 text files.
15 unique files.
3 files ignored. v 1.65 T=0.23 s (57.1 files/s, 188914.0 lines/s)
Language files blank comment code
C 10 4680 6621 30812
C/C++ Header 3 99 286 496
SUM: 13 4779 6907 31308

an archive

We'll pull cloc's source zip file from GitHub, then count the contents:

prompt> wget

prompt> cloc v 1.65 T=0.07 s (26.8 files/s, 141370.3 lines/s)
Language files blank comment code
Perl 2 725 1103 8713
SUM: 2 725 1103 8713

a git repository, using a specific commit

This example uses code from
PuDB, a fantastic Python debugger.

prompt> git clone

prompt> cd pudb

prompt> cloc 6be804e07a5db
48 text files.
48 unique files.
15 files ignored. v 1.73 T=0.15 s (223.1 files/s, 46159.0 lines/s)
Language files blank comment code
Python 28 1519 728 4659
YAML 2 9 2 75
Bourne Shell 3 6 0 17
make 1 4 6 10
SUM: 34 1538 736 4761

each subdirectory of a particular directory

Say you have a directory with three different git-managed projects,
Project0, Project1, and Project2. You can use your shell's looping
capability to count the code in each. This example uses bash (scroll down for cmd.exe example):

prompt> for d in ./*/ ; do (cd "$d" && echo "$d" && cloc --vcs git); done
7 text files.
7 unique files.
1 file ignored. v 1.71 T=0.02 s (390.2 files/s, 25687.6 lines/s)
Language files blank comment code
D 4 61 32 251
Markdown 1 9 0 38
make 1 0 0 4
SUM: 6 70 32 293
7 text files.
7 unique files.
0 files ignored. v 1.71 T=0.02 s (293.0 files/s, 52107.1 lines/s)
Language files blank comment code
Go 7 165 282 798
SUM: 7 165 282 798
49 text files.
47 unique files.
13 files ignored. v 1.71 T=0.10 s (399.5 files/s, 70409.4 lines/s)
Language files blank comment code
Python 33 1226 1026 3017
C 4 327 337 888
Markdown 1 11 0 28
YAML 1 0 2 12
SUM: 39 1564 1365 3945

each subdirectory of a particular directory (Windows/cmd.exe)

for /D %I in (.\*) do cd %I && cloc --vcs git && cd ..

Overview ▲

cloc counts blank lines, comment lines, and physical lines of source
code in many programming languages. Given two versions of
a code base, cloc can compute differences in blank, comment, and source
lines. It is written entirely in Perl with no dependencies outside the
standard distribution of Perl v5.6 and higher (code from some external
modules is embedded within
cloc) and so is
quite portable. cloc is known to run on many flavors of Linux, FreeBSD,
NetBSD, OpenBSD, macOS, AIX, HP-UX, Solaris, IRIX, z/OS, and Windows.
(To run the Perl source version of cloc on Windows one needs
ActiveState Perl 5.6.1 or
higher, Strawberry Perl,
Windows Subsystem for Linux,
MobaXTerm with the Perl plug-in
a mingw environment and terminal such as provided by
Git for Windows.
Alternatively one can use the Windows binary of cloc
generated with PAR::Packer
to run on Windows computers that have neither Perl nor Cygwin.)

cloc contains code from David Wheeler's
Damian Conway and Abigail's Perl module
Sean M. Burke's Perl module
and Tye McQueen's Perl module
Language scale factors were derived from Mayes Consulting, LLC web site

Run via docker

docker run --rm -v $PWD:/tmp aldanial/cloc

Install via package manager

Depending your operating system, one of these installation methods may
work for you (all but the last two entries for Windows require
a Perl interpreter):

npm install -g cloc              #
sudo apt install cloc # Debian, Ubuntu
sudo yum install cloc # Red Hat, Fedora
sudo dnf install cloc # Fedora 22 or later
sudo pacman -S cloc # Arch
sudo emerge -av dev-util/cloc # Gentoo
sudo apk add cloc # Alpine Linux
doas pkg_add cloc # OpenBSD
sudo pkg install cloc # FreeBSD
sudo port install cloc # macOS with MacPorts
brew install cloc # macOS with Homebrew
choco install cloc # Windows with Chocolatey
scoop install cloc # Windows with Scoop

Note: I don't control any of these packages.
If you encounter a bug in cloc using one of the above
packages, try with cloc pulled from the latest stable release here
on GitHub (link follows below) before submitting a problem report.

Stable release

Development version

License ▲

cloc is licensed under the
GNU General Public License, v 2,
excluding portions which
are copied from other sources. Code
copied from the Regexp::Common, Win32::Autoglob, and Algorithm::Diff
Perl modules is subject to the
Artistic License.

Why Use cloc? ▲

cloc has many features that make it easy to use, thorough, extensible, and portable:

  1. Exists as a single, self-contained file that requires minimal installation effort---just download the file and run it.

  2. Can read language comment definitions from a file and thus potentially work with computer languages that do not yet exist.

  3. Allows results from multiple runs to be summed together by language and by project.

  4. Can produce results in a variety of formats: plain text, SQL, JSON, XML, YAML, comma separated values.

  5. Can count code within compressed archives (tar balls, Zip files, Java .ear files).

  6. Has numerous troubleshooting options.

  7. Handles file and directory names with spaces and other unusual characters.

  8. Has no dependencies outside the standard Perl distribution.

  9. Runs on Linux, FreeBSD, NetBSD, OpenBSD, macOS, AIX, HP-UX, Solaris, IRIX, and z/OS systems that have Perl 5.6 or higher. The source version runs on Windows with either ActiveState Perl, Strawberry Perl, Cygwin, or MobaXTerm+Perl plugin. Alternatively on Windows one can run the Windows binary which has no dependencies.

Other Counters ▲

If cloc does not suit your needs here are other freely available counters to consider:

Other references:

Regexp::Common, Digest::MD5, Win32::Autoglob, Algorithm::Diff

Although cloc does not need Perl modules outside those found in the
standard distribution, cloc does rely on a few external modules. Code
from three of these external modules--Regexp::Common, Win32::Autoglob,
and Algorithm::Diff--is embedded within cloc. A fourth module,
Digest::MD5, is used only if it is available. If cloc finds
Regexp::Common or Algorithm::Diff installed locally it will use those
installation. If it doesn't, cloc will install the parts of
Regexp::Common and/or Algorithm:Diff it needs to temporary directories
that are created at the start of a cloc run then removed when the run is
complete. The necessary code from Regexp::Common v2.120 and
Algorithm::Diff v1.1902 are embedded within the cloc source code (see
subroutines Install_Regexp_Common() and Install_Algorithm_Diff() ).
Only three lines are needed from Win32::Autoglob and these are included
directly in cloc.

Additionally, cloc will use Digest::MD5 to validate uniqueness among
equally-sized input files if Digest::MD5 is installed locally.

A parallel processing option, --processes=N, was introduced with
cloc version 1.76 to enable faster runs on multi-core machines. However,
to use it, one must have the module Parallel::ForkManager installed.
This module does not work reliably on Windows so parallel processing
will only work on Unix-like operating systems.

The Windows binary is built on a computer that has both Regexp::Common
and Digest::MD5 installed locally.

Building a Windows Executable ▲

The Windows downloads
cloc-1.90.exe and
cloc-1.88.exe were built on a 64 bit Windows 10 computer
Strawberry Perl
5.30.2 and
to build the .exe.

Release 1.86 was built on a 64 bit Windows 10 virtual machine
downloaded from;
releases 1.74 through 1.84
were was built on a 32 bit Windows 7 virtual machine
using Strawberry Perl, while
1.70 and 1.72 were built with Strawberry Perl
on an Amazon Web Services t2.micro instance running Microsoft Windows Server 2008
(32 bit for 1.70 and 1.72; 64 bit for 1.74).
Release 1.66 was built on a 32 bit Windows 7 VirtualBox image.
Windows executables of cloc versions 1.60 and earlier were built with
perl2exe on a 32 bit Windows
XP computer. A small modification was made to the cloc source code
before passing it to perl2exe; lines 87 and 88 were uncommented:

85  # Uncomment next two lines when building Windows executable with perl2exe
86 # or if running on a system that already has Regexp::Common.
87 #use Regexp::Common;
88 #$HAVE_Rexexp_Common = 1;

Is the Windows executable safe to run? Does it have malware?

Ideally, no one would need the Windows executable because they
have a Perl interpreter installed on their machines and can
run the cloc source file.
On centrally-managed corporate Windows machines, however, this
this may be difficult or impossible.

The Windows executable distributed with cloc is provided as
a best-effort of a virus and malware-free .exe.
You are encouraged to run your own virus scanners against the
executable and also check sites such .
The entries for recent versions are:










cloc 1.72 exe:

cloc 1.70 exe:

cloc 1.68 exe:

cloc 1.66 exe:

Why is the Windows executable so large?

Windows executables of cloc versions 1.60 and earlier, created with
perl2exe as noted above, are about 1.6 MB, while versions 1.62 and 1.54, created
with PAR::Packer, are 11 MB.
Version 1.66, built with a newer version of PAR::Packer, is about 5.5 MB.
Why are the PAR::Packer, executables so
much larger than those built with perl2exe? My theory is that perl2exe
uses smarter tree pruning logic
than PAR::Packer, but that's pure speculation.

Create your own executable

The most robust option for creating a Windows executable of
cloc is to use ActiveState's Perl Development Kit.
It includes a utility, perlapp, which can build stand-alone
Windows, Mac, and Linux binaries of Perl source code.

will also do the trick. If you do have perl2exe, modify lines
84-87 in the cloc source code for a minor code
modification that is necessary to make a cloc Windows executable.

Otherwise, to build a Windows executable with pp from
PAR::Packer, first install a Windows-based Perl distribution
(for example Strawberry Perl or ActivePerl) following their
instructions. Next, open a command prompt, aka a DOS window and install
the PAR::Packer module. Finally, invoke the newly installed pp
command with the cloc source code to create an .exe file:

C:> cpan -i Digest::MD5
C:> cpan -i Regexp::Common
C:> cpan -i Algorithm::Diff
C:> cpan -i PAR::Packer
C:> pp -M Digest::MD5 -c -x -o cloc-1.90.exe

A variation on the instructions above is if you installed the portable
version of Strawberry Perl, you will need to run portableshell.bat first
to properly set up your environment.

Basic Use ▲

cloc is a command line program that takes file, directory, and/or
archive names as inputs. Here's an example of running cloc against the
Perl v5.22.0 source distribution:

prompt> cloc perl-5.22.0.tar.gz
5605 text files.
5386 unique files.
2176 files ignored. v 1.65 T=25.49 s (134.7 files/s, 51980.3 lines/s)
Language files blank comment code
Perl 2892 136396 184362 536445
C 130 24676 33684 155648
C/C++ Header 148 9766 16569 147858
Bourne Shell 112 4044 6796 42668
Pascal 8 458 1603 8592
XML 33 142 0 2410
YAML 49 20 15 2078
C++ 10 313 277 2033
make 4 426 488 1986
Prolog 12 438 2 1146
JSON 14 1 0 1037
yacc 1 85 76 998
Windows Message File 1 102 11 489
DOS Batch 14 92 41 389
Windows Resource File 3 10 0 85
D 1 5 7 8
Lisp 2 0 3 4
SUM: 3434 176974 243934 903874

To run cloc on Windows computers, open up a command (aka DOS) window
and invoke cloc.exe from the command line there.
Alternatively, try ClocViewer, the GUI wrapper around cloc found at

See also for a
graphical rendering of cloc results.

Options ▲
prompt> cloc --help

Usage: cloc [options] <file(s)/dir(s)/git hash(es)> | <set 1> <set 2> | <report files>

Count, or compute differences of, physical lines of source code in the
given files (may be archives such as compressed tarballs or zip files,
or git commit hashes or branch names) and/or recursively below the
given directories.

Input Options
--extract-with=<cmd> This option is only needed if cloc is unable
to figure out how to extract the contents of
the input file(s) by itself.
Use <cmd> to extract binary archive files (e.g.:
.tar.gz, .zip, .Z). Use the literal '>FILE<' as
a stand-in for the actual file(s) to be
extracted. For example, to count lines of code
in the input files
gcc-4.2.tar.gz perl-5.8.8.tar.gz
on Unix use
--extract-with='gzip -dc >FILE< | tar xf -'
or, if you have GNU tar,
--extract-with='tar zxf >FILE<'
and on Windows use, for example:
--extract-with="\"c:\Program Files\WinZip\WinZip32.exe\" -e -o >FILE< ."
(if WinZip is installed there).
--list-file=<file> Take the list of file and/or directory names to
process from <file>, which has one file/directory
name per line. Only exact matches are counted;
relative path names will be resolved starting from
the directory where cloc is invoked. Set <file>
to - to read file names from a STDIN pipe.
See also --exclude-list-file.
--diff-list-file=<file> Take the pairs of file names to be diff'ed from
<file>, whose format matches the output of
--diff-alignment. (Run with that option to
see a sample.) The language identifier at the
end of each line is ignored. This enables --diff
mode and bypasses file pair alignment logic.
--vcs=<VCS> Invoke a system call to <VCS> to obtain a list of
files to work on. If <VCS> is 'git', then will
invoke 'git ls-files' to get a file list and
'git submodule status' to get a list of submodules
whose contents will be ignored. See also --git
which accepts git commit hashes and branch names.
If <VCS> is 'svn' then will invoke 'svn list -R'.
The primary benefit is that cloc will then skip
files explicitly excluded by the versioning tool
in question, ie, those in .gitignore or have the
svn:ignore property.
Alternatively <VCS> may be any system command
that generates a list of files.
Note: cloc must be in a directory which can read
the files as they are returned by <VCS>. cloc will
not download files from remote repositories.
'svn list -R' may refer to a remote repository
to obtain file names (and therefore may require
authentication to the remote repository), but
the files themselves must be local.
Setting <VCS> to 'auto' selects between 'git'
and 'svn' (or neither) depending on the presence
of a .git or .svn subdirectory below the directory
where cloc is invoked.
--unicode Check binary files to see if they contain Unicode
expanded ASCII text. This causes performance to
drop noticeably.

Processing Options
--autoconf Count .in files (as processed by GNU autoconf) of
recognized languages. See also --no-autogen.
--by-file Report results for every source file encountered.
--by-file-by-lang Report results for every source file encountered
in addition to reporting by language.
--config <file> Read command line switches from <file> instead of
the default location of /home/al/.config/cloc/options.txt.
The file should contain one switch, along with
arguments (if any), per line. Blank lines and lines
beginning with '#' are skipped. Options given on
the command line take priority over entries read from
the file.
--count-and-diff <set1> <set2>
First perform direct code counts of source file(s)
of <set1> and <set2> separately, then perform a diff
of these. Inputs may be pairs of files, directories,
or archives. If --out or --report-file is given,
three output files will be created, one for each
of the two counts and one for the diff. See also
--diff, --diff-alignment, --diff-timeout,
--ignore-case, --ignore-whitespace.
--diff <set1> <set2> Compute differences in code and comments between
source file(s) of <set1> and <set2>. The inputs
may be any mix of files, directories, archives,
or git commit hashes. Use --diff-alignment to
generate a list showing which file pairs where
compared. When comparing git branches, only files
which have changed in either commit are compared.
See also --git, --count-and-diff, --diff-alignment,
--diff-list-file, --diff-timeout, --ignore-case,
--diff-timeout <N> Ignore files which take more than <N> seconds
to process. Default is 10 seconds. Setting <N>
to 0 allows unlimited time. (Large files with many
repeated lines can cause Algorithm::Diff::sdiff()
to take hours.) See also --timeout.
--docstring-as-code cloc considers docstrings to be comments, but this is
not always correct as docstrings represent regular
strings when they appear on the right hand side of an
assignment or as function arguments. This switch
forces docstrings to be counted as code.
--follow-links [Unix only] Follow symbolic links to directories
(sym links to files are always followed).
See also --stat.
Process all files that have a <ext> extension
with the counter for language <lang>. For
example, to count all .f files with the
Fortran 90 counter (which expects files to
end with .f90) instead of the default Fortran 77
counter, use
--force-lang="Fortran 90",f
If <ext> is omitted, every file will be counted
with the <lang> counter. This option can be
specified multiple times (but that is only
useful when <ext> is given each time).
See also --script-lang, --lang-no-ext.
--force-lang-def=<file> Load language processing filters from <file>,
then use these filters instead of the built-in
filters. Note: languages which map to the same
file extension (for example:
Pascal/PHP; Lisp/OpenCL; Lisp/Julia; Perl/Prolog)
will be ignored as these require additional
processing that is not expressed in language
definition files. Use --read-lang-def to define
new language filters without replacing built-in
filters (see also --write-lang-def,
--git Forces the inputs to be interpreted as git targets
(commit hashes, branch names, et cetera) if these
are not first identified as file or directory
names. This option overrides the --vcs=git logic
if this is given; in other words, --git gets its
list of files to work on directly from git using
the hash or branch name rather than from
'git ls-files'. This option can be used with
--diff to perform line count diffs between git
commits, or between a git commit and a file,
directory, or archive. Use -v/--verbose to see
the git system commands cloc issues.
--git-diff-rel Same as --git --diff, or just --diff if the inputs
are recognized as git targets. Only files which
have changed in either commit are compared.
--git-diff-all Git diff strategy #2: compare all files in the
repository between the two commits.
--ignore-whitespace Ignore horizontal white space when comparing files
with --diff. See also --ignore-case.
--ignore-case Ignore changes in case within file contents;
consider upper- and lowercase letters equivalent
when comparing files with --diff. See also
--ignore-case-ext Ignore case of file name extensions. This will
cause problems counting some languages
(specifically, .c and .C are associated with C and
C++; this switch would count .C files as C rather
than C++ on *nix operating systems). File name
case insensitivity is always true on Windows.
--lang-no-ext=<lang> Count files without extensions using the <lang>
counter. This option overrides internal logic
for files without extensions (where such files
are checked against known scripting languages
by examining the first line for #!). See also
--force-lang, --script-lang.
--max-file-size=<MB> Skip files larger than <MB> megabytes when
traversing directories. By default, <MB>=100.
cloc's memory requirement is roughly twenty times
larger than the largest file so running with
files larger than 100 MB on a computer with less
than 2 GB of memory will cause problems.
Note: this check does not apply to files
explicitly passed as command line arguments.
--no-autogen[=list] Ignore files generated by code-production systems
such as GNU autoconf. To see a list of these files
(then exit), run with --no-autogen list
See also --autoconf.
--original-dir [Only effective in combination with
--strip-comments] Write the stripped files
to the same directory as the original files.
--read-binary-files Process binary files in addition to text files.
This is usually a bad idea and should only be
attempted with text files that have embedded
binary data.
--read-lang-def=<file> Load new language processing filters from <file>
and merge them with those already known to cloc.
If <file> defines a language cloc already knows
about, cloc's definition will take precedence.
Use --force-lang-def to over-ride cloc's
definitions (see also --write-lang-def,
--script-lang=<lang>,<s> Process all files that invoke <s> as a #!
scripting language with the counter for language
<lang>. For example, files that begin with
will be counted with the Perl counter by using
The language name is case insensitive but the
name of the script language executable, <s>,
must have the right case. This option can be
specified multiple times. See also --force-lang,
--sdir=<dir> Use <dir> as the scratch directory instead of
letting File::Temp chose the location. Files
written to this location are not removed at
the end of the run (as they are with File::Temp).
--skip-uniqueness Skip the file uniqueness check. This will give
a performance boost at the expense of counting
files with identical contents multiple times
(if such duplicates exist).
--stat Some file systems (AFS, CD-ROM, FAT, HPFS, SMB)
do not have directory 'nlink' counts that match
the number of its subdirectories. Consequently
cloc may undercount or completely skip the
contents of such file systems. This switch forces
File::Find to stat directories to obtain the
correct count. File search speed will decrease.
See also --follow-links.
--stdin-name=<file> Give a file name to use to determine the language
for standard input. (Use - as the input name to
receive source code via STDIN.)
--strip-comments=<ext> For each file processed, write to the current
directory a version of the file which has blank
and commented lines removed (in-line comments
persist). The name of each stripped file is the
original file name with .<ext> appended to it.
It is written to the current directory unless
--original-dir is on.
--strip-str-comments Replace comment markers embedded in strings with
'xx'. This attempts to work around a limitation
in Regexp::Common::Comment where comment markers
embedded in strings are seen as actual comment
markers and not strings, often resulting in a
'Complex regular subexpression recursion limit'
warning and incorrect counts. There are two
disadvantages to using this switch: 1/code count
performance drops, and 2/code generated with
--strip-comments will contain different strings
where ever embedded comments are found.
--sum-reports Input arguments are report files previously
created with the --report-file option in plain
format (eg. not JSON, YAML, XML, or SQL).
Makes a cumulative set of results containing the
sum of data from the individual report files.
--timeout <N> Ignore files which take more than <N> seconds
to process at any of the language's filter stages.
The default maximum number of seconds spent on a
filter stage is the number of lines in the file
divided by one thousand. Setting <N> to 0 allows
unlimited time. See also --diff-timeout.
--processes=NUM [Available only on systems with a recent version
of the Parallel::ForkManager module. Not
available on Windows.] Sets the maximum number of
cores that cloc uses. The default value of 0
disables multiprocessing.
--unix Override the operating system autodetection
logic and run in UNIX mode. See also
--windows, --show-os.
--use-sloccount If SLOCCount is installed, use its compiled
executables c_count, java_count, pascal_count,
php_count, and xml_count instead of cloc's
counters. SLOCCount's compiled counters are
substantially faster than cloc's and may give
a performance improvement when counting projects
with large files. However, these cloc-specific
features will not be available: --diff,
--count-and-diff, --strip-comments, --unicode.
--windows Override the operating system autodetection
logic and run in Microsoft Windows mode.
See also --unix, --show-os.

Filter Options
--exclude-content=<regex> Exclude files containing text that matches the given
regular expression.
--exclude-dir=<D1>[,D2,] Exclude the given comma separated directories
D1, D2, D3, et cetera, from being scanned. For
example --exclude-dir=.cache,test will skip
all files and subdirectories that have /.cache/
or /test/ as their parent directory.
Directories named .bzr, .cvs, .hg, .git, .svn,
and .snapshot are always excluded.
This option only works with individual directory
names so including file path separators is not
allowed. Use --fullpath and --not-match-d=<regex>
to supply a regex matching multiple subdirectories.
Do not count files having the given file name
Exclude the given comma separated languages
L1, L2, L3, et cetera, from being counted.
--exclude-list-file=<file> Ignore files and/or directories whose names
appear in <file>. <file> should have one file
name per line. Only exact matches are ignored;
relative path names will be resolved starting from
the directory where cloc is invoked.
See also --list-file.
--fullpath Modifies the behavior of --match-f, --not-match-f,
and --not-match-d to include the file's path
in the regex, not just the file's basename.
(This does not expand each file to include its
absolute path, instead it uses as much of
the path as is passed in to cloc.)
Note: --match-d always looks at the full
path and therefore is unaffected by --fullpath.
Count only languages having the given comma
separated file extensions. Use --show-ext to
see the recognized extensions.
Count only the given comma separated languages
L1, L2, L3, et cetera. Use --show-lang to see
the list of recognized languages.
--match-d=<regex> Only count files in directories matching the Perl
regex. For example
only counts files in directories containing
/src/ or /include/. Unlike --not-match-d,
--match-f, and --not-match-f, --match-d always
compares the fully qualified path against the
--not-match-d=<regex> Count all files except those in directories
matching the Perl regex. Only the trailing
directory name is compared, for example, when
counting in /usr/local/lib, only 'lib' is
compared to the regex.
Add --fullpath to compare parent directories to
the regex.
Do not include file path separators at the
beginning or end of the regex.
--match-f=<regex> Only count files whose basenames match the Perl
regex. For example
only counts files that start with Widget or widget.
Add --fullpath to include parent directories
in the regex instead of just the basename.
--not-match-f=<regex> Count all files except those whose basenames
match the Perl regex. Add --fullpath to include
parent directories in the regex instead of just
the basename.
--skip-archive=<regex> Ignore files that end with the given Perl regular
expression. For example, if given
the code will skip files that end with .zip,
.tar, .tar.gz, .tar.Z, .tar.bz2, .tar.xz, and
--skip-win-hidden On Windows, ignore hidden files.

Debug Options
--categorized=<file> Save names of categorized files to <file>.
--counted=<file> Save names of processed source files to <file>.
--diff-alignment=<file> Write to <file> a list of files and file pairs
showing which files were added, removed, and/or
compared during a run with --diff. This switch
forces the --diff mode on.
--explain=<lang> Print the filters used to remove comments for
language <lang> and exit. In some cases the
filters refer to Perl subroutines rather than
regular expressions. An examination of the
source code may be needed for further explanation.
--help Print this usage information and exit.
--found=<file> Save names of every file found to <file>.
--ignored=<file> Save names of ignored files and the reason they
were ignored to <file>.
--print-filter-stages Print processed source code before and after
each filter is applied.
--show-ext[=<ext>] Print information about all known (or just the
given) file extensions and exit.
--show-lang[=<lang>] Print information about all known (or just the
given) languages and exit.
--show-os Print the value of the operating system mode
and exit. See also --unix, --windows.
-v[=<n>] Verbose switch (optional numeric value).
-verbose[=<n>] Long form of -v.
--version Print the version of this program and exit.
--write-lang-def=<file> Writes to <file> the language processing filters
then exits. Useful as a first step to creating
custom language definitions. Note: languages which
map to the same file extension will be excluded.
(See also --force-lang-def, --read-lang-def).
Same as --write-lang-def, but includes duplicated
extensions. This generates a problematic language
definition file because cloc will refuse to use
it until duplicates are removed.

Output Options
--3 Print third-generation language output.
(This option can cause report summation to fail
if some reports were produced with this option
while others were produced without it.)
--by-percent X Instead of comment and blank line counts, show
these values as percentages based on the value
of X in the denominator:
X = 'c' -> # lines of code
X = 'cm' -> # lines of code + comments
X = 'cb' -> # lines of code + blanks
X = 'cmb' -> # lines of code + comments + blanks
For example, if using method 'c' and your code
has twice as many lines of comments as lines
of code, the value in the comment column will
be 200%. The code column remains a line count.
--csv Write the results as comma separated values.
--csv-delimiter=<C> Use the character <C> as the delimiter for comma
separated files instead of ,. This switch forces --csv to be on.
--file-encoding=<E> Write output files using the <E> encoding instead of
the default ASCII (<E> = 'UTF-7'). Examples: 'UTF-16',
'euc-kr', 'iso-8859-16'. Known encodings can be
printed with
perl -MEncode -e 'print join("\n", Encode->encodings(":all")), "\n"'
--hide-rate Do not show line and file processing rates in the
output header. This makes output deterministic.
--json Write the results as JavaScript Object Notation
(JSON) formatted output.
--md Write the results as Markdown-formatted text.
--out=<file> Synonym for --report-file=<file>.
--progress-rate=<n> Show progress update after every <n> files are
processed (default <n>=100). Set <n> to 0 to
suppress progress output (useful when redirecting
output to STDOUT).
--quiet Suppress all information messages except for
the final report.
--report-file=<file> Write the results to <file> instead of STDOUT.
--sql=<file> Write results as SQL create and insert statements
which can be read by a database program such as
SQLite. If <file> is -, output is sent to STDOUT.
--sql-append Append SQL insert statements to the file specified
by --sql and do not generate table creation
statements. Only valid with the --sql option.
--sql-project=<name> Use <name> as the project identifier for the
current run. Only valid with the --sql option.
--sql-style=<style> Write SQL statements in the given style instead
of the default SQLite format. Styles include
'Oracle' and 'Named_Columns'.
--sum-one For plain text reports, show the SUM: output line
even if only one input file is processed.
--xml Write the results in XML.
--xsl=<file> Reference <file> as an XSL stylesheet within
the XML output. If <file> is 1 (numeric one),
writes a default stylesheet, cloc.xsl (or
cloc-diff.xsl if --diff is also given).
This switch forces --xml on.
--yaml Write the results in YAML.

Recognized Languages ▲
prompt> cloc --show-lang

ABAP (abap)
ActionScript (as)
Ada (ada, adb, ads, pad)
ADSO/IDSM (adso)
Agda (agda, lagda)
AMPLE (ample, dofile, startup)
Ant (build.xml, build.xml)
ANTLR Grammar (g, g4)
Apex Class (cls)
Apex Trigger (trigger)
APL (apl, apla, aplc, aplf, apli, apln, aplo, dyalog, dyapp, mipage)
Arduino Sketch (ino, pde)
AsciiDoc (adoc, asciidoc)
ASP (asa, ashx, asp, axd)
ASP.NET (asax, ascx, asmx, aspx, master, sitemap, webinfo)
AspectJ (aj)
Assembly (a51, asm, nasm, S, s)
AutoHotkey (ahk, ahkl)
awk (auk, awk, gawk, mawk, nawk)
Bazel (bazel, BUILD)
BizTalk Orchestration (odx)
BizTalk Pipeline (btp)
Blade (blade, blade.php)
Bourne Again Shell (bash)
Bourne Shell (sh)
BrightScript (brs)
builder (xml.builder)
C (c, cats, ec, idc, pgc)
C Shell (csh, tcsh)
C# (cs)
C# Designer (designer.cs)
C++ (C, c++, cc, CPP, cpp, cxx, h++, inl, ipp, pcc, tcc, tpp)
C/C++ Header (H, h, hh, hpp, hxx)
Cake Build Script (cake)
CCS (ccs)
Chapel (chpl)
Clean (dcl, icl)
Clojure (boot, cl2, clj, cljs.hl, cljscm, cljx, hic, riemann.config)
ClojureC (cljc)
ClojureScript (cljs)
CMake (cmake,, CMakeLists.txt)
COBOL (CBL, cbl, ccp, COB, cob, cobol, cpy)
CoffeeScript (_coffee, cakefile, cjsx, coffee, iced)
ColdFusion (cfm, cfml)
ColdFusion CFScript (cfc)
Coq (v)
Crystal (cr)
CSON (cson)
CSS (css)
CSV (csv)
Cucumber (feature)
CUDA (cu, cuh)
Cython (pxd, pxi, pyx)
D (d)
DAL (da)
Dart (dart)
Delphi Form (dfm)
dhall (dhall)
DIET (dt)
diff (diff, patch)
DITA (dita)
Dockerfile (Dockerfile, dockerfile)
DOORS Extension Language (dxl)
DOS Batch (BAT, bat, BTM, btm, CMD, cmd)
Drools (drl)
DTD (dtd)
dtrace (d)
ECPP (ecpp)
EEx (eex)
EJS (ejs)
Elixir (ex, exs)
Elm (elm)
Embedded Crystal (ecr)
ERB (ERB, erb)
Erlang (app.src, emakefile, erl, hrl, rebar.config, rebar.config.lock, rebar.lock, xrl, yrl)
Expect (exp)
F# (fsi, fs, fs)
F# Script (fsx)
Fennel (fnl)
Fish Shell (fish)
Focus (focexec)
Forth (4th, e4, f83, fb, forth, fpm, fr, frt, ft, fth, rx, fs, f, for)
Fortran 77 (F, F77, f77, FOR, FTN, ftn, pfo, f, for)
Fortran 90 (F90, f90)
Fortran 95 (F95, f95)
Freemarker Template (ftl)
FXML (fxml)
GDScript (gd)
Gencat NLS (msg)
Glade (glade, ui)
Gleam (gleam)
GLSL (comp, fp, frag, frg, fsh, fshader, geo, geom, glsl, glslv, gshader, tesc, tese, vert, vrx, vsh, vshader)
Go (go)
Godot Resource (tres)
Godot Scene (tscn)
Gradle (gradle, gradle.kts)
Grails (gsp)
GraphQL (gql, graphql, graphqls)
Groovy (gant, groovy, grt, gtpl, gvy, jenkinsfile)
Haml (haml, haml.deface)
Handlebars (handlebars, hbs)
Harbour (hb)
Haskell (hs, hsc, lhs)
Haxe (hx, hxsl)
HCL (hcl, nomad, tf, tfvars)
HLSL (cg, cginc, fxh, hlsl, hlsli, shader)
Hoon (hoon)
HTML (htm, html, html.hl, xht)
IDL (dlm, idl, pro)
Idris (idr)
Igor Pro (ipf)
Imba (imba)
INI (buildozer.spec, ini, lektorproject, prefs)
InstallShield (ism)
IPL (ipl)
Java (java)
JavaScript (_js, bones, es6, jake, jakefile, js, jsb, jscad, jsfl, jsm, jss, mjs, njs, pac, sjs, ssjs, xsjs, xsjslib)
JavaServer Faces (jsf)
JCL (jcl)
Jinja Template (jinja, jinja2)
JSON (arcconfig, avsc, composer.lock, geojson, gltf, har, htmlhintrc, json, json-tmlanguage, jsonl, mcmeta,, tern-config, tern-project, tfstate, tfstate.backup, topojson, watchmanconfig, webapp, webmanifest, yyp)
JSON5 (json5)
JSP (jsp, jspf)
JSX (jsx)
Julia (jl)
Juniper Junos (junos)
Jupyter Notebook (ipynb)
Kermit (ksc)
Korn Shell (ksh)
Kotlin (kt, ktm, kts)
Lean (hlean, lean)
LESS (less)
lex (l, lex)
LFE (lfe)
liquid (liquid)
Lisp (asd, el, lisp, lsp, cl, jl)
Literate Idris (lidr)
LiveLink OScript (oscript)
LLVM IR (ll)
Logos (x, xm)
Logtalk (lgt, logtalk)
Lua (lua, nse, p8, pd_lua, rbxs, wlua)
m4 (ac, m4)
make (am, Gnumakefile, gnumakefile, Makefile, makefile, mk)
Mako (mako, mao)
Markdown (, markdown, md, mdown, mdwn, mdx, mkd, mkdn, mkdown, ronn, workbook)
Mathematica (cdf, ma, mathematica, mt, nbp, wl, wlt, m)
Maven (pom, pom.xml)
Meson (
Modula3 (i3, ig, m3, mg)
Mojo (mojom)
MSBuild script (btproj, csproj, msbuild, vcproj, wdproj, wixproj)
MUMPS (mps, m)
Mustache (mustache)
MXML (mxml)
NAnt script (build)
Nemerle (n)
Nim (nim, nim.cfg, nimble, nimrod, nims)
Nix (nix)
Objective-C (m)
Objective-C++ (mm)
OCaml (eliom, eliomi, ml, ml4, mli, mll, mly)
Odin (odin)
OpenCL (cl)
Oracle Forms (fmt)
Oracle PL/SQL (bod, fnc, prc, spc, trg)
Oracle Reports (rex)
Pascal (dpr, lpr, p, pas, pascal)
Pascal/Puppet (pp)
Patran Command Language (pcl, ses)
Perl (ack, al, cpanfile,, perl, ph, plh, plx, pm, psgi, rexfile, pl, p6)
PHP (aw, ctp, phakefile, php, php3, php4, php5, php_cs, php_cs.dist, phps, phpt, phtml)
PHP/Pascal (inc)
Pig Latin (pig)
PL/I (pl1)
PL/M (lit, plm)
PO File (po)
PowerBuilder (pbt, sra, srf, srm, srs, sru, srw)
PowerShell (ps1, psd1, psm1)
ProGuard (pro)
Prolog (P, prolog, yap, pl, p6, pro)
Protocol Buffers (proto)
Pug (jade, pug)
PureScript (purs)
Python (buck, build.bazel, gclient, gyp, gypi, lmi, py, py3, pyde, pyi, pyp, pyt, pyw, sconscript, sconstruct, snakefile, tac, workspace, wscript, wsgi, xpy)
QML (qbs, qml)
Qt (ui)
Qt Linguist (ts)
Qt Project (pro)
R (expr-dist, R, r, rd, rprofile, rsx)
Racket (rkt, rktd, rktl, scrbl)
Raku (pm6, raku, rakumod)
Raku/Prolog (P6, p6)
RAML (raml)
RapydScript (pyj)
Razor (cshtml, razor)
ReasonML (re, rei)
ReScript (res, resi)
reStructuredText (rest, rest.txt, rst, rst.txt)
Rexx (pprx, rexx)
Ring (rform, rh, ring)
Rmd (Rmd)
RobotFramework (robot)
Ruby (appraisals, berksfile, brewfile, builder, buildfile, capfile, dangerfile, deliverfile, eye, fastfile, gemfile, gemfile.lock, gemspec, god, guardfile, irbrc, jarfile, jbuilder, mavenfile, mspec, podfile, podspec, pryrc, puppetfile, rabl, rake, rb, rbuild, rbw, rbx, ru, snapfile, thor, thorfile, vagrantfile, watchr)
Ruby HTML (rhtml)
Rust (rs,
SaltStack (sls)
SAS (sas)
Sass (sass, scss)
Scala (kojo, sbt, scala)
Scheme (sc, sch, scm, sld, sps, ss, sls)
sed (sed)
SKILL (il)
SKILL++ (ils)
Slice (ice)
Slim (slim)
Smalltalk (st, cs)
Smarty (smarty, tpl)
Softbridge Basic (SBL, sbl)
Solidity (sol)
SparForte (sp)
Specman e (e)
SQL (cql, mysql, psql, SQL, sql, tab, udf, viw)
SQL Data (data.sql)
SQL Stored Procedure (spc.sql, spoc.sql, sproc.sql, udf.sql)
Squirrel (nut)
Standard ML (fun, sig, sml)
Starlark (bzl)
Stata (ado, DO, do, doh, ihlp, mata, matah, sthlp)
Stylus (styl)
SugarSS (sss)
Svelte (svelte)
SVG (SVG, svg)
Swift (swift)
SWIG (i)
Tcl/Tk (itk, tcl, tk)
Teamcenter met (met)
Teamcenter mth (mth)
TeX (aux, bbx, bib, bst, cbx, dtx, ins, lbx, ltx, mkii, mkiv, mkvi, sty, tex, cls)
Thrift (thrift)
TITAN Project File Information (tpd)
Titanium Style Sheet (tss)
TNSDL (cii, cin, in1, in2, in3, in4, inf, interface, rou, sdl, sdt, spd, ssc, sst)
TOML (toml)
TTCN (ttcn, ttcn2, ttcn3, ttcnpp)
Twig (twig)
TypeScript (tsx, ts)
Unity-Prefab (mat, prefab)
Vala (vala)
Vala Header (vapi)
VB for Applications (VBA, vba)
Velocity Template Language (vm)
Verilog-SystemVerilog (sv, svh, v)
VHDL (VHD, vhd, VHDL, vhdl, vhf, vhi, vho, vhs, vht, vhw)
vim script (vim)
Visual Basic (BAS, bas, ctl, dsr, frm, FRX, frx, VBHTML, vbhtml, vbp, vbw, cls)
Visual Basic .NET (VB, vb, vbproj)
Visual Basic Script (VBS, vbs)
Visual Fox Pro (SCA, sca)
Visual Studio Solution (sln)
Visualforce Component (component)
Visualforce Page (page)
Vuejs Component (vue)
Web Services Description (wsdl)
WebAssembly (wast, wat)
Windows Message File (mc)
Windows Module Definition (def)
Windows Resource File (rc, rc2)
WiX include (wxi)
WiX source (wxs)
WiX string localization (wxl)
WXML (wxml)
WXSS (wxss)
XAML (xaml)
xBase (prg, prw)
xBase Header (ch)
XHTML (xhtml)
XMI (XMI, xmi)
XML (adml, admx, ant, app.config, axml, builds, ccproj, ccxml, classpath, clixml, cproject, cscfg, csdef, csl, ct, depproj, ditamap, ditaval, dll.config, dotsettings, filters, fsproj, gmx, grxml, iml, ivy, jelly, jsproj, kml, launch, mdpolicy, mjml, natvis, ndproj, nproj, nuget.config, nuspec, odd, osm, packages.config, pkgproj, plist, proj, project, props, ps1xml, psc1, pt, rdf, resx, rss, scxml, settings.stylecop, sfproj, shproj, srdf, storyboard, sttheme, sublime-snippet, targets, tmcommand, tml, tmlanguage, tmpreferences, tmsnippet, tmtheme, urdf, ux, vcxproj, vsixmanifest, vssettings, vstemplate, vxml, web.config, web.debug.config, web.release.config, wsf, x3d, xacro, xib, xlf, xliff, XML, xml, xml.dist, xproj, xspec, xul, zcml)
XQuery (xq, xql, xqm, xquery, xqy)
XSD (XSD, xsd)
XSLT (XSL, xsl, XSLT, xslt)
Xtend (xtend)
yacc (y, yacc)
YAML (clang-format, clang-tidy, gemrc, glide.lock, mir, reek, rviz, sublime-syntax, syntax, yaml, yaml-tmlanguage, yml, yml.mysql)
Zig (zig)
zsh (zsh)

The above list can be customized by reading language definitions from a
file with the --read-lang-def or --force-lang-def options.

These file extensions map to multiple languages:

cloc has subroutines that attempt to identify the correct language based
on the file's contents for these special cases. Language identification
accuracy is a function of how much code the file contains; .m files with
just one or two lines for example, seldom have enough information to
correctly distinguish between MATLAB, Mercury, MUMPS, or Objective-C.

Languages with file extension collisions are difficult to customize with
--read-lang-def or --force-lang-def as they have no mechanism to
identify languages with common extensions. In this situation one must
modify the cloc source code.

How It Works ▲

cloc's method of operation resembles SLOCCount's: First, create a list
of files to consider. Next, attempt to determine whether or not found
files contain recognized computer language source code. Finally, for
files identified as source files, invoke language-specific routines to
count the number of source lines.

A more detailed description:

  1. If the input file is an archive (such as a .tar.gz or .zip file),
    create a temporary directory and expand the archive there using a
    system call to an appropriate underlying utility (tar, bzip2, unzip,
    etc) then add this temporary directory as one of the inputs. (This
    works more reliably on Unix than on Windows.)

  2. Use File::Find to recursively descend the input directories and make
    a list of candidate file names. Ignore binary and zero-sized files.

  3. Make sure the files in the candidate list have unique contents
    (first by comparing file sizes, then, for similarly sized files,
    compare MD5 hashes of the file contents with Digest::MD5). For each
    set of identical files, remove all but the first copy, as determined
    by a lexical sort, of identical files from the set. The removed
    files are not included in the report. (The --skip-uniqueness switch
    disables the uniqueness tests and forces all copies of files to be
    included in the report.) See also the --ignored= switch to see which
    files were ignored and why.

  4. Scan the candidate file list for file extensions which cloc
    associates with programming languages (see the --show-lang and
    --show-ext options). Files which match are classified as
    containing source
    code for that language. Each file without an extensions is opened
    and its first line read to see if it is a Unix shell script
    (anything that begins with #!). If it is shell script, the file is
    classified by that scripting language (if the language is
    recognized). If the file does not have a recognized extension or is
    not a recognized scripting language, the file is ignored.

  5. All remaining files in the candidate list should now be source files
    for known programming languages. For each of these files:

    1. Read the entire file into memory.

    2. Count the number of lines (= Loriginal).

    3. Remove blank lines, then count again (= Lnon_blank).

    4. Loop over the comment filters defined for this language. (For
      example, C++ has two filters: (1) remove lines that start with
      optional whitespace followed by // and (2) remove text between
      / and /) Apply each filter to the code to remove comments.
      Count the left over lines (= Lcode).

    5. Save the counts for this language:

      • blank lines = Loriginal - Lnon_blank

      • comment lines = Lnon_blank - Lcode

      • code lines = Lcode

The options modify the algorithm slightly. The --read-lang-def option
for example allows the user to read definitions of comment filters,
known file extensions, and known scripting languages from a file. The
code for this option is processed between Steps 2 and 3.

Advanced Use ▲

Remove Comments from Source Code ▲

How can you tell if cloc correctly identifies comments? One way to
convince yourself cloc is doing the right thing is to use its
--strip-comments option to remove comments and blank lines from files, then
compare the stripped-down files to originals.

Let's try this out with the SQLite amalgamation, a C file containing all
code needed to build the SQLite library along with a header file:

prompt> tar zxf sqlite-amalgamation-3.5.6.tar.gz
prompt> cd sqlite-3.5.6/
prompt> cloc --strip-comments=nc sqlite.c
1 text file.
1 unique file.
0 files ignored. v 1.03 T=1.0 s (1.0 files/s, 82895.0 lines/s)
Language files blank comment code scale 3rd gen. equiv
C 1 5167 26827 50901 x 0.77 = 39193.77

The extension argument given to --strip-comments is arbitrary; here nc was used as an abbreviation for "no comments".

cloc removed over 31,000 lines from the file:

prompt> wc -l sqlite3.c
82895 sqlite3.c
133796 total
prompt> echo "82895 - 50901" | bc

We can now compare the original file, sqlite3.c and the one stripped of
comments, with tools like diff or vimdiff and see what
exactly cloc considered comments and blank lines. A rigorous proof that
the stripped-down file contains the same C code as the original is to
compile these files and compare checksums of the resulting object files.

First, the original source file:

prompt> gcc -c sqlite3.c
prompt> md5sum sqlite3.o
cce5f1a2ea27c7e44b2e1047e2588b49 sqlite3.o

Next, the version without comments:

prompt> mv sqlite3.c
prompt> gcc -c sqlite3.c
prompt> md5sum sqlite3.o
cce5f1a2ea27c7e44b2e1047e2588b49 sqlite3.o

cloc removed over 31,000 lines of comments and blanks but did not modify the source code in any significant way since the resulting object file matches the original.

Work with Compressed Archives ▲

Versions of cloc before v1.07 required an
--extract-with=CMD option to tell cloc how
to expand an archive file. Beginning with v1.07 this is extraction is
attempted automatically. At the moment the automatic extraction method works
reasonably well on Unix-type OS's for the following file types:
Some of these extensions work on Windows if one has WinZip installed
in the default location (C:\Program Files\WinZip\WinZip32.exe).
Additionally, with newer versions of WinZip, the
is needed for correct operation; in this case one would invoke cloc with
something like

 --extract-with="\"c:\Program Files\WinZip\wzunzip\" -e -o >FILE< ."


In situations where the automatic extraction fails, one can try the
option to count lines of code within tar files, Zip files, or
other compressed archives for which one has an extraction tool.
cloc takes the user-provided extraction command and expands the archive
to a temporary directory (created with File::Temp),
counts the lines of code in the temporary directory,
then removes that directory. While not especially helpful when dealing
with a single compressed archive (after all, if you're going to type
the extraction command anyway why not just manually expand the archive?)
this option is handy for working with several archives at once.

For example, say you have the following source tarballs on a Unix machine


and you want to count all the code within them. The command would be

cloc --extract-with='gzip -dc >FILE< | tar xf -' perl-5.8.5.tar.gz Python-2.4.2.tar.gz

If that Unix machine has GNU tar (which can uncompress and extract in
one step) the command can be shortened to

cloc --extract-with='tar zxf >FILE<' perl-5.8.5.tar.gz Python-2.4.2.tar.gz

On a Windows computer with WinZip installed in
c:\Program Files\WinZip the command would look like

cloc.exe --extract-with="\"c:\Program Files\WinZip\WinZip32.exe\" -e -o >FILE< ." perl-5.8.5.tar.gz Python-2.4.2.tar.gz

Java .ear files are Zip files that contain additional Zip
files. cloc can handle nested compressed archives without
difficulty--provided all such files are compressed and archived in the
same way. Examples of counting a
Java .ear file in Unix and Windows:

Unix> cloc --extract-with="unzip -d . >FILE< " Project.ear
DOS> cloc.exe --extract-with="\"c:\Program Files\WinZip\WinZip32.exe\" -e -o >FILE< ." Project.ear

Differences ▲

The --diff switch allows one to measure the relative change in
source code and comments between two versions of a file, directory,
or archive. Differences reveal much more than absolute code
counts of two file versions. For example, say a source file
has 100 lines and its developer delivers a newer version with
102 lines. Did the developer add two comment lines,
or delete seventeen source
lines and add fourteen source lines and five comment lines, or did
the developer
do a complete rewrite, discarding all 100 original lines and
adding 102 lines of all new source? The diff option tells how
many lines of source were added, removed, modified or stayed
the same, and how many lines of comments were added, removed,
modified or stayed the same.

Differences in blank lines are handled much more coarsely
because these are stripped by cloc early on. Unless a
file pair is identical, cloc will report only differences
in absolute counts of blank lines. In other words, one
can expect to see only entries for 'added' if the second
file has more blanks than the first, and 'removed' if the
situation is reversed. The entry for 'same' will be non-zero
only when the two files are identical.

In addition to file pairs, one can give cloc pairs of
directories, or pairs of file archives, or a file archive
and a directory. cloc will try to align
file pairs within the directories or archives and compare diffs
for each pair. For example, to see what changed between
GCC 4.4.0 and 4.5.0 one could do

cloc --diff gcc-4.4.0.tar.bz2  gcc-4.5.0.tar.bz2

Be prepared to wait a while for the results though; the --diff
option runs much more slowly than an absolute code count.

To see how cloc aligns files between the two archives, use the
--diff-alignment option

cloc --diff-alignment=align.txt gcc-4.4.0.tar.bz2  gcc-4.5.0.tar.bz2

to produce the file align.txt which shows the file pairs as well
as files added and deleted. The symbols == and != before each
file pair indicate if the files are identical (==)
or if they have different content (!=).

Here's sample output showing the difference between the Python 2.6.6 and 2.7

prompt> cloc --diff Python-2.7.9.tgz Python-2.7.10.tar.xz
4315 text files.
4313 text files.s
2173 files ignored.

4 errors:
Diff error, exceeded timeout: /tmp/8ToGAnB9Y1/Python-2.7.9/Mac/Modules/qt/_Qtmodule.c
Diff error, exceeded timeout: /tmp/M6ldvsGaoq/Python-2.7.10/Mac/Modules/qt/_Qtmodule.c
Diff error (quoted comments?): /tmp/8ToGAnB9Y1/Python-2.7.9/Mac/Modules/qd/
Diff error (quoted comments?): /tmp/M6ldvsGaoq/Python-2.7.10/Mac/Modules/qd/ v 1.65 T=298.59 s (0.0 files/s, 0.0 lines/s)
Language files blank comment code
Visual Basic
same 2 0 1 12
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
same 11 0 340 2952
modified 1 0 0 1
added 0 0 0 0
removed 0 0 0 0
same 1 0 87 105
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
same 0 0 19 327
modified 1 0 0 1
added 0 0 0 0
removed 0 0 0 0
same 7 0 61 635
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
NAnt script
same 2 0 0 30
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
same 3 0 2 72
modified 1 0 0 1
added 0 0 0 1
removed 0 1 0 0
Windows Resource File
same 3 0 56 206
modified 1 0 0 1
added 0 0 0 0
removed 0 0 0 0
same 6 0 161 565
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
same 14 0 11 2344
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
vim script
same 1 0 7 106
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
same 2 0 18 128
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
Windows Module Definition
same 7 0 187 2080
modified 2 0 0 0
added 0 0 0 1
removed 0 1 0 2
same 1 0 0 24
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
same 3 0 49 229
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
same 51 0 6794 12298
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
Bourne Shell
same 41 0 7698 45024
modified 1 0 0 3
added 0 13 2 64
removed 0 0 0 0
DOS Batch
same 29 0 107 494
modified 1 0 0 9
added 0 1 0 3
removed 0 0 0 0
MSBuild script
same 77 0 3 38910
modified 0 0 0 0
added 0 0 0 0
removed 0 0 0 0
same 1947 0 109012 430335
modified 192 0 94 950
added 2 323 283 2532
removed 2 55 58 646
same 18 0 191 15352
modified 1 0 0 2
added 1 31 0 205
removed 0 0 0 0
same 505 0 37439 347837
modified 45 0 13 218
added 0 90 33 795
removed 0 9 2 148
C/C++ Header
same 255 0 10361 66635
modified 5 0 5 7
added 0 1 3 300
removed 0 0 0 0
same 2986 0 172604 966700
modified 251 0 112 1193
added 3 459 321 3901
removed 2 66 60 796

A pair of errors occurred.
The first pair was caused by timing out when computing diffs of the file
Python-X/Mac/Modules/qt/_Qtmodule.c in each Python version.
This file has > 26,000 lines of C code and takes more than
10 seconds--the default maximum duration for diff'ing a
single file--on my slow computer. (Note: this refers to
performing differences with
the sdiff() function in the Perl Algorithm::Diff module,
not the command line diff utility.) This error can be
overcome by raising the time to, say, 20 seconds
with --diff-timeout 20.

The second error is more problematic. The files
include Python docstring (text between pairs of triple quotes)
containing C comments. cloc treats docstrings as comments and handles them
by first converting them to C comments, then using the C comment removing
regular expression. Nested C comments yield erroneous results however.

Create Custom Language Definitions ▲

cloc can write its language comment definitions to a file or can read
comment definitions from a file, overriding the built-in definitions.
This can be useful when you want to use cloc to count lines of a
language not yet included, to change association of file extensions
to languages, or to modify the way existing languages are counted.

The easiest way to create a custom language definition file is to
make cloc write its definitions to a file, then modify that file:

Unix> cloc --write-lang-def=my_definitions.txt

creates the file my_definitions.txt which can be modified
then read back in with either the --read-lang-def or
--force-lang-def option. The difference between the options is
former merges language definitions from the given file in with
cloc's internal definitions with cloc's taking precedence
if there are overlaps. The --force-lang-def option, on the
other hand, replaces cloc's definitions completely.
This option has a disadvantage in preventing cloc from counting

languages whose extensions map to multiple languages
as these languages require additional logic that is not easily
expressed in a definitions file.

Unix> cloc --read-lang-def=my_definitions.txt  file1 file2 dir1 ...

Each language entry has four parts:
* The language name starting in column 1.
* One or more comment filters starting in column 5.
* One or more filename extensions starting in column 5.
* A 3rd generation scale factor starting in column 5.
This entry must be provided
but its value is not important
unless you want to compare your language to a hypothetical
third generation programming language.

A filter defines a method to remove comment text from the source file.
For example the entry for C++ looks like this

filter call_regexp_common C++
filter remove_inline //.*$
extension C
extension c++
extension cc
extension cpp
extension cxx
extension pcc
3rd_gen_scale 1.51
end_of_line_continuation \\$

C++ has two filters: first, remove lines matching
Regexp::Common's C++ comment regex.
The second filter using remove_inline is currently
unused. Its intent is to identify lines with both
code and comments and it may be implemented in the future.

A more complete discussion of the different filter options may appear
here in the future. The output of cloc's
--write-lang-def option should provide enough examples
for motivated individuals to modify or extend cloc's language definitions.

Combine Reports ▲

If you manage multiple software projects you might be interested in
seeing line counts by project, not just by language.
Say you manage three software projects called MariaDB, PostgreSQL, and SQLite.
The teams responsible for each of these projects run cloc on their
source code and provide you with the output.
For example MariaDB team does

cloc --out mariadb-10.1.txt

and provides you with the file mariadb-10.1.txt.
The contents of the three files you get are

Unix> cat mariadb-10.1.txt v 1.65 T=45.36 s (110.5 files/s, 66411.4 lines/s)
Language files blank comment code
C++ 1613 225338 290077 983026
C 853 62442 73017 715018
C/C++ Header 1327 48300 114577 209394
Bourne Shell 256 10224 10810 61943
Perl 147 10342 8305 35562
Pascal 107 4907 5237 32541
HTML 56 195 6 16489
Javascript 5 3309 3019 15540
m4 30 1599 359 14215
CMake 190 1919 4097 12206
XML 35 648 56 5210
Ruby 59 619 184 4998
Puppet 10 0 1 3848
make 134 724 360 3631
SQL 23 306 377 3405
Python 34 371 122 2545
Bourne Again Shell 27 299 380 1604
Windows Module Definition 37 27 13 1211
lex 4 394 166 991
yacc 2 152 64 810
DOS Batch 19 89 82 700
Prolog 1 9 40 448
RobotFramework 1 0 0 441
CSS 2 33 155 393
JSON 5 0 0 359
dtrace 9 59 179 306
Windows Resource File 10 61 89 250
Assembly 2 70 284 237
WiX source 1 18 10 155
Visual Basic 6 0 0 88
YAML 2 4 4 65
PHP 1 11 2 24
SKILL 1 8 15 16
sed 2 0 0 16
Windows Message File 1 2 8 6
diff 1 1 4 4
D 1 4 11 4
SUM: 5014 372484 512110 2127699

Unix> cat sqlite-3081101.txt v 1.65 T=1.22 s (3.3 files/s, 143783.6 lines/s)
Language files blank comment code
C 2 11059 53924 101454
C/C++ Header 2 211 6630 1546
SUM: 4 11270 60554 103000

Unix> cat postgresql-9.4.4.txt v 1.65 T=22.46 s (172.0 files/s, 96721.6 lines/s)
Language files blank comment code
HTML 1254 3725 0 785991
C 1139 139289 244045 736519
C/C++ Header 667 12277 32488 57014
SQL 410 13400 8745 51926
yacc 8 3163 2669 28491
Bourne Shell 41 2647 2440 17170
Perl 81 1702 1308 9456
lex 9 792 1631 4285
make 205 1525 1554 4114
m4 12 218 25 1642
Windows Module Definition 13 4 17 1152
XSLT 5 76 55 294
DOS Batch 7 29 30 92
CSS 1 20 7 69
Assembly 3 17 38 69
D 1 14 14 66
Windows Resource File 3 4 0 62
Lisp 1 1 1 16
sed 1 1 7 15
Python 1 5 0 13
Bourne Again Shell 1 8 6 10
Windows Message File 1 0 0 5
SUM: 3864 178917 295080 1698471

While these three files are interesting, you also want to see
the combined counts from all projects.
That can be done with cloc's --sum_reports

Unix> cloc --sum-reports --out=databases mariadb-10.1.txt  sqlite-3081101.txt  postgresql-9.4.4.txt
Wrote databases.lang
Wrote databases.file

The report combination produces two output files, one for sums by
programming language (databases.lang) and one by project
Their contents are

Unix> cat databases.lang v 1.65
Language files blank comment code
C 1994 212790 370986 1552991
C++ 1613 225338 290077 983026
HTML 1310 3920 6 802480
C/C++ Header 1996 60788 153695 267954
Bourne Shell 297 12871 13250 79113
SQL 433 13706 9122 55331
Perl 228 12044 9613 45018
Pascal 107 4907 5237 32541
yacc 10 3315 2733 29301
m4 42 1817 384 15857
Javascript 5 3309 3019 15540
CMake 190 1919 4097 12206
make 339 2249 1914 7745
lex 13 1186 1797 5276
XML 35 648 56 5210
Ruby 59 619 184 4998
Puppet 10 0 1 3848
Python 35 376 122 2558
Windows Module Definition 50 31 30 2363
Bourne Again Shell 28 307 386 1614
DOS Batch 26 118 112 792
CSS 3 53 162 462
Prolog 1 9 40 448
RobotFramework 1 0 0 441
JSON 5 0 0 359
Windows Resource File 13 65 89 312
Assembly 5 87 322 306
dtrace 9 59 179 306
XSLT 5 76 55 294
WiX source 1 18 10 155
Visual Basic 6 0 0 88
D 2 18 25 70
YAML 2 4 4 65
sed 3 1 7 31
PHP 1 11 2 24
SKILL 1 8 15 16
Lisp 1 1 1 16
Windows Message File 2 2 8 11
diff 1 1 4 4
SUM: 8882 562671 867744 3929170

Unix> cat databases.file
File files blank comment code
mariadb-10.1.txt 5014 372484 512110 2127699
postgresql-9.4.4.txt 3864 178917 295080 1698471
sqlite-3081101.txt 4 11270 60554 103000
SUM: 8882 562671 867744 3929170

Report files themselves can be summed together. Say you also manage
development of Perl and Python and you want to keep track
of those line counts separately from your database projects. First
create reports for Perl and Python separately:

cloc --out perl-5.22.0.txt   perl-5.22.0.tar.gz
cloc --out python-2.7.10.txt Python-2.7.10.tar.xz

then sum these together with

Unix> cloc --sum-reports --out script_lang perl-5.22.0.txt python-2.7.10.txt
Wrote script_lang.lang
Wrote script_lang.file

Unix> cat script_lang.lang v 1.65
Language files blank comment code
Perl 2892 136396 184362 536445
C 680 75566 71211 531203
Python 2141 89642 109524 434015
C/C++ Header 408 16433 26938 214800
Bourne Shell 154 11088 14496 87759
MSBuild script 77 0 3 38910
m4 20 1604 191 15559
Assembly 51 3775 6794 12298
Pascal 8 458 1603 8592
make 16 897 828 4939
XML 37 198 2 2484
HTML 14 393 11 2344
C++ 12 338 295 2161
Windows Module Definition 9 171 187 2081
YAML 49 20 15 2078
Prolog 12 438 2 1146
JSON 14 1 0 1037
yacc 1 85 76 998
DOS Batch 44 199 148 895
Objective-C 7 98 61 635
Expect 6 104 161 565
Windows Message File 1 102 11 489
CSS 1 98 19 328
Windows Resource File 7 55 56 292
Javascript 3 31 49 229
vim script 1 36 7 106
diff 1 17 87 105
NAnt script 2 1 0 30
IDL 1 0 0 24
Visual Basic 2 1 1 12
D 1 5 7 8
Lisp 2 0 3 4
SUM: 6674 338250 417148 1902571

Unix> cat script_lang.file
File files blank comment code
python-2.7.10.txt 3240 161276 173214 998697
perl-5.22.0.txt 3434 176974 243934 903874
SUM: 6674 338250 417148 1902571

Finally, combine the combination files:

Unix> cloc --sum-reports --report_file=everything databases.lang script_lang.lang
Wrote everything.lang
Wrote everything.file

Unix> cat everything.lang v 1.65
Language files blank comment code
C 2674 288356 442197 2084194
C++ 1625 225676 290372 985187
HTML 1324 4313 17 804824
Perl 3120 148440 193975 581463
C/C++ Header 2404 77221 180633 482754
Python 2176 90018 109646 436573
Bourne Shell 451 23959 27746 166872
SQL 433 13706 9122 55331
Pascal 115 5365 6840 41133
MSBuild script 77 0 3 38910
m4 62 3421 575 31416
yacc 11 3400 2809 30299
Javascript 8 3340 3068 15769
make 355 3146 2742 12684
Assembly 56 3862 7116 12604
CMake 190 1919 4097 12206
XML 72 846 58 7694
lex 13 1186 1797 5276
Ruby 59 619 184 4998
Windows Module Definition 59 202 217 4444
Puppet 10 0 1 3848
YAML 51 24 19 2143
DOS Batch 70 317 260 1687
Bourne Again Shell 28 307 386 1614
Prolog 13 447 42 1594
JSON 19 1 0 1396
CSS 4 151 181 790
Objective-C 7 98 61 635
Windows Resource File 20 120 145 604
Expect 6 104 161 565
Windows Message File 3 104 19 500
RobotFramework 1 0 0 441
dtrace 9 59 179 306
XSLT 5 76 55 294
WiX source 1 18 10 155
diff 2 18 91 109
vim script 1 36 7 106
Visual Basic 8 1 1 100
D 3 23 32 78
sed 3 1 7 31
NAnt script 2 1 0 30
IDL 1 0 0 24
PHP 1 11 2 24
Lisp 3 1 4 20
SKILL 1 8 15 16
SUM: 15556 900921 1284892 5831741

Unix> cat everything.file
File files blank comment code
databases.lang 8882 562671 867744 3929170
script_lang.lang 6674 338250 417148 1902571
SUM: 15556 900921 1284892 5831741

One limitation of the --sum-reports feature is that the individual counts must
be saved in the plain text format. Counts saved as
XML, JSON, YAML, or SQL will produce errors if used in a summation.


Cloc can write results in the form of SQL table create and insert
statements for use
with relational database programs such as SQLite, MySQL,
PostgreSQL, Oracle, or Microsoft SQL.
Once the code count information is in a database,
the information can be interrogated and displayed in interesting ways.

A database created from cloc SQL output has two tables,
metadata and t:

Table metadata:

|Field | Type |
|timestamp | text |
|project | text |
|elapsed_s | text |

Table t:

|Field | Type |
| project |text |
| language |text |
| file |text |
| nBlank |integer |
| nComment |integer |
| nCode |integer |
| nScaled |real |

The metadata table contains information about when the cloc run
was made. The --sql-append switch allows one to combine
many runs in a single database; each run adds a
row to the metadata table.
The code count information resides in table t.

Let's repeat the code count examples of Perl, Python, SQLite, MySQL and
PostgreSQL tarballs shown in the
Combine Reports
example above, this time
using the SQL output options and the
database engine.

The --sql switch tells cloc to generate output in the form
of SQL table create and insert commands. The switch takes
an argument of a file name to write these SQL statements into, or,
if the argument is 1 (numeric one), streams output to STDOUT.
Since the SQLite command line program, sqlite3, can read
commands from STDIN, we can dispense with storing SQL statements to
a file and use --sql 1 to pipe data directly into the
SQLite executable:

cloc --sql 1 --sql-project mariadb | sqlite3 code.db

The --sql-project mariadb part is optional; there's no need
to specify a project name when working with just one code base. However,
since we'll be adding code counts from four other tarballs, we'll only
be able to identify data by input source if we supply a
project name for each run.

Now that we have a database we will need to pass in the --sql-append
switch to tell cloc not to wipe out this database but instead add more data:

cloc --sql 1 --sql-project postgresql --sql-append postgresql-9.4.4.tar.bz2        | sqlite3 code.db
cloc --sql 1 --sql-project sqlite --sql-append | sqlite3 code.db
cloc --sql 1 --sql-project python --sql-append Python-2.7.10.tar.xz | sqlite3 code.db
cloc --sql 1 --sql-project perl --sql-append perl-5.22.0.tar.gz | sqlite3 code.db

Now the fun begins--we have a database, code.db, with lots of
information about the five projects and can query it
for all manner of interesting facts.

Which is the longest file over all projects?

prompt> sqlite3 code.db 'select project,file,nBlank+nComment+nCode as nL from t
where nL = (select max(nBlank+nComment+nCode) from t)'


sqlite3's default output format leaves a bit to be desired.
We can add an option to the program's rc file,
~/.sqliterc, to show column headers:

  .header on

One might be tempted to also include

  .mode column

in ~/.sqliterc but this causes problems when the output has more than
one row since the widths of entries in the first row govern the maximum
width for all subsequent rows. Often this leads to truncated output--not
at all desirable. One option is to write a custom SQLite output
formatter such as sqlite_formatter, included with cloc.

To use it, simply pass sqlite3's STDOUT into sqlite_formatter
via a pipe:

prompt> sqlite3 code.db 'select project,file,nBlank+nComment+nCode as nL from t
where nL = (select max(nBlank+nComment+nCode) from t)' | ./sqlite_formatter

-- Loading resources from ~/.sqliterc
Project File nL
_______ _____________________________________ ______
sqlite sqlite-amalgamation-3081101/sqlite3.c 161623

If the "Project File" line doesn't appear, add .header on to your
~/.sqliterc file as explained above.

What is the longest file over all projects?

prompt> sqlite3 code.db 'select project,file,nBlank+nComment+nCode as nL from t
where nL = (select max(nBlank+nComment+nCode) from t)' | sqlite_formatter

Project File nL
_______ _____________________________________ ______
sqlite sqlite-amalgamation-3081101/sqlite3.c 161623

What is the longest file in each project?

prompt> sqlite3 code.db 'select project,file,max(nBlank+nComment+nCode) as nL from t
group by project order by nL;' | sqlite_formatter

Project File nL
__________ ________________________________________________________________ ______
python Python-2.7.10/Mac/Modules/qt/_Qtmodule.c 28091
postgresql postgresql-9.4.4/src/interfaces/ecpg/preproc/preproc.c 54623
mariadb server-10.1/storage/mroonga/vendor/groonga/lib/nfkc.c 80246
perl perl-5.22.0/cpan/Locale-Codes/lib/Locale/Codes/ 100747
sqlite sqlite-amalgamation-3081101/sqlite3.c 161623

Which files in each project have the most code lines?

prompt> sqlite3 code.db 'select project,file,max(nCode) as nL from t
group by project order by nL desc;' | sqlite_formatter

Project File nL
__________ ________________________________________________________________ ______
perl perl-5.22.0/cpan/Locale-Codes/lib/Locale/Codes/ 100735
sqlite sqlite-amalgamation-3081101/sqlite3.c 97469
mariadb server-10.1/storage/mroonga/vendor/groonga/lib/nfkc.c 80221
postgresql postgresql-9.4.4/src/interfaces/ecpg/preproc/preproc.c 45297
python Python-2.7.10/Mac/Modules/qt/_Qtmodule.c 26705

Which C source files with more than 300 lines have a comment ratio below 1%?

prompt> sqlite3 code.db 'select project, file, nCode, nComment,
(100.0*nComment)/(nComment+nCode) as comment_ratio from t
where language="C" and nCode > 300 and
comment_ratio < 1 order by comment_ratio;' | sqlite_formatter

Project File nCode nComment comment_ratio
__________ _______________________________________________________________________________________________ _____ ________ __________________
mariadb server-10.1/storage/mroonga/vendor/groonga/lib/nfkc.c 80221 14 0.0174487443135789
python Python-2.7.10/Python/graminit.c 2175 1 0.0459558823529412
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_turkish.c 2095 1 0.0477099236641221
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_french.c 1211 1 0.0825082508250825
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_french.c 1201 1 0.0831946755407654
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_hungarian.c 1182 1 0.084530853761623
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_hungarian.c 1178 1 0.0848176420695505
mariadb server-10.1/strings/ctype-eucjpms.c 67466 60 0.0888546633889169
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_english.c 1072 1 0.0931966449207828
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_english.c 1064 1 0.0938967136150235
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_spanish.c 1053 1 0.094876660341556
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_spanish.c 1049 1 0.0952380952380952
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_italian.c 1031 1 0.0968992248062016
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_italian.c 1023 1 0.09765625
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_portuguese.c 981 1 0.10183299389002
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_portuguese.c 975 1 0.102459016393443
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_romanian.c 967 1 0.103305785123967
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_2_romanian.c 961 1 0.103950103950104
mariadb server-10.1/strings/ctype-ujis.c 67177 79 0.117461639110265
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_finnish.c 720 1 0.13869625520111
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_porter.c 717 1 0.139275766016713
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_finnish.c 714 1 0.13986013986014
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_porter.c 711 1 0.140449438202247
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_KOI8_R_russian.c 660 1 0.151285930408472
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_russian.c 654 1 0.152671755725191
python Python-2.7.10/Mac/Modules/qt/_Qtmodule.c 26705 42 0.157026956294164
python Python-2.7.10/Mac/Modules/icn/_Icnmodule.c 1521 3 0.196850393700787
mariadb server-10.1/strings/ctype-extra.c 8282 18 0.216867469879518
postgresql postgresql-9.4.4/src/bin/psql/sql_help.c 3576 8 0.223214285714286
mariadb server-10.1/strings/ctype-sjis.c 34006 86 0.252258594391646
python Python-2.7.10/Python/Python-ast.c 6554 17 0.258712524729874
mariadb server-10.1/strings/ctype-cp932.c 34609 92 0.265122042592432
perl perl-5.22.0/keywords.c 2815 8 0.283386468296139
python Python-2.7.10/Mac/Modules/menu/_Menumodule.c 3263 10 0.305530094714329
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_dutch.c 596 2 0.334448160535117
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_dutch.c 586 2 0.340136054421769
mariadb server-10.1/strings/ctype-gbk.c 10684 38 0.354411490393583
python Python-2.7.10/Mac/Modules/qd/_Qdmodule.c 6694 24 0.357249181303959
python Python-2.7.10/Mac/Modules/win/_Winmodule.c 3056 11 0.358656667753505
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_german.c 476 2 0.418410041841004
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_german.c 470 2 0.423728813559322
mariadb server-10.1/strings/ctype-euc_kr.c 9956 44 0.44
postgresql postgresql-9.4.4/src/backend/utils/fmgrtab.c 4815 23 0.475403059115337
python Python-2.7.10/Mac/Modules/ctl/_Ctlmodule.c 5442 28 0.511882998171846
python Python-2.7.10/Mac/Modules/ae/_AEmodule.c 1347 7 0.51698670605613
python Python-2.7.10/Mac/Modules/app/_Appmodule.c 1712 9 0.52295177222545
mariadb server-10.1/strings/ctype-gb2312.c 6377 35 0.54585152838428
mariadb server-10.1/storage/tokudb/ft-index/third_party/xz-4.999.9beta/src/liblzma/lzma/fastpos_table.c 516 3 0.578034682080925
python Python-2.7.10/Mac/Modules/evt/_Evtmodule.c 504 3 0.591715976331361
python Python-2.7.10/Modules/expat/xmlrole.c 1256 8 0.632911392405063
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_UTF_8_danish.c 312 2 0.636942675159236
postgresql postgresql-9.4.4/src/backend/snowball/libstemmer/stem_ISO_8859_1_danish.c 310 2 0.641025641025641
python Python-2.7.10/Mac/Modules/res/_Resmodule.c 1621 12 0.734843845682792
python Python-2.7.10/Mac/Modules/drag/_Dragmodule.c 1046 8 0.759013282732448
python Python-2.7.10/Mac/Modules/list/_Listmodule.c 1021 8 0.777453838678329
python Python-2.7.10/Mac/Modules/te/_TEmodule.c 1198 10 0.827814569536424
python Python-2.7.10/Mac/Modules/cg/_CGmodule.c 1190 10 0.833333333333333
python Python-2.7.10/Modules/clmodule.c 2379 23 0.957535387177352
python Python-2.7.10/Mac/Modules/folder/_Foldermodule.c 306 3 0.970873786407767

What are the ten longest files (based on code lines) that have no comments at all? Exclude header, .html, and YAML files.

prompt> sqlite3 code.db 'select project, file, nCode from t
where nComment = 0 and
language not in ("C/C++ Header", "YAML", "HTML")
order by nCode desc limit 10;' | sqlite_formatter

Project File nCode
_______ ____________________________________________________________________ _____
perl perl-5.22.0/cpan/Unicode-Collate/Collate/Locale/ 1938
python Python-2.7.10/PCbuild/pythoncore.vcproj 1889
python Python-2.7.10/PC/VS8.0/pythoncore.vcproj 1889
mariadb server-10.1/mysql-test/extra/binlog_tests/ 1862
perl perl-5.22.0/cpan/Unicode-Collate/Collate/Locale/ 1589
perl perl-5.22.0/cpan/Unicode-Collate/Collate/Locale/ 1563
mariadb server-10.1/storage/mroonga/vendor/groonga/ 1526
perl perl-5.22.0/cpan/Unicode-Collate/Collate/Locale/ 1505
mariadb server-10.1/mysql-test/suite/funcs_1/storedproc/ 1465
python Python-2.7.10/PC/VS8.0/_bsddb.vcproj 1463

What are the most popular languages (in terms of lines
of code) in each project?

prompt> sqlite3 code.db 'select project, language, sum(nCode) as SumCode from t
group by project,language
order by project,SumCode desc;' | sqlite_formatter
Project Language SumCode
__________ _________________________ _______
mariadb C++ 983026
mariadb C 715018
mariadb C/C++ Header 209394
mariadb Bourne Shell 61943
mariadb Perl 35562
mariadb Pascal 32541
mariadb HTML 16489
mariadb Javascript 15540
mariadb m4 14215
mariadb CMake 12206
mariadb XML 5210
mariadb Ruby 4998
mariadb Puppet 3848
mariadb make 3631
mariadb SQL 3405
mariadb Python 2545
mariadb Bourne Again Shell 1604
mariadb Windows Module Definition 1211
mariadb lex 991
mariadb yacc 810
mariadb DOS Batch 700
mariadb Prolog 448
mariadb RobotFramework 441
mariadb CSS 393
mariadb JSON 359
mariadb dtrace 306
mariadb Windows Resource File 250
mariadb Assembly 237
mariadb WiX source 155
mariadb Visual Basic 88
mariadb YAML 65
mariadb PHP 24
mariadb SKILL 16
mariadb sed 16
mariadb Windows Message File 6
mariadb D 4
mariadb diff 4
perl Perl 536445
perl C 155648
perl C/C++ Header 147858
perl Bourne Shell 42668
perl Pascal 8592
perl XML 2410
perl YAML 2078
perl C++ 2033
perl make 1986
perl Prolog 1146
perl JSON 1037
perl yacc 998
perl Windows Message File 489
perl DOS Batch 389
perl Windows Resource File 85
perl D 8
perl Lisp 4
postgresql HTML 785991
postgresql C 736519
postgresql C/C++ Header 57014
postgresql SQL 51926
postgresql yacc 28491
postgresql Bourne Shell 17170
postgresql Perl 9456
postgresql lex 4285
postgresql make 4114
postgresql m4 1642
postgresql Windows Module Definition 1152
postgresql XSLT 294
postgresql DOS Batch 92
postgresql Assembly 69
postgresql CSS 69
postgresql D 66
postgresql Windows Resource File 62
postgresql Lisp 16
postgresql sed 15
postgresql Python 13
postgresql Bourne Again Shell 10
postgresql Windows Message File 5
python Python 434015
python C 375555
python C/C++ Header 66942
python Bourne Shell 45091
python MSBuild script 38910
python m4 15559
python Assembly 12298
python make 2953
python HTML 2344
python Windows Module Definition 2081
python Objective-C 635
python Expect 565
python DOS Batch 506
python CSS 328
python Javascript 229
python Windows Resource File 207
python C++ 128
python vim script 106
python diff 105
python XML 74
python NAnt script 30
python Prolog 24
python Visual Basic 12
sqlite C 101454
sqlite C/C++ Header 1546

Custom Column Output ▲

Cloc's default output is a text table with five columns:
language, file count, number of blank lines, number of comment
lines and number of code lines. The switches --by-file,
--3, and --by-percent generate additional information but
sometimes even those are insufficient.

The --sql option described in the previous section offers the
ability to create custom output. This section has a pair of examples
that show how to create custom columns.
The first example includes an extra column, Total, which is the
sum of the numbers of blank, comment, and code lines.
The second shows how to include the language name when running
with --by-file.

Example 1: Add a "Totals" column.

The first step is to run cloc and save the output to a relational database,
SQLite in this case:

cloc --sql 1 --sql-project x yaml-cpp-yaml-cpp-0.5.3.tar.gz | sqlite3 counts.db

(the tar file comes from the
YAML-C++ project).

Second, we craft an SQL query that returns the regular cloc output
plus an extra column for totals, then save the SQL statement to
a file, query_with_totals.sql:

-- file query_with_totals.sql
select Language, count(File) as files ,
sum(nBlank) as blank ,
sum(nComment) as comment ,
sum(nCode) as code ,
sum(nBlank)+sum(nComment)+sum(nCode) as Total
from t group by Language order by code desc;

Third, we run this query through SQLite using the counts.db database.
We'll include the -header switch so that SQLite prints the
column names:

> cat query_with_totals.sql | sqlite3 -header counts.db
C/C++ Header|110|8566|17420|51502|77488
Bourne Shell|10|6351|6779|38264|51394
MSBuild script|11|0|0|1711|1711

The extra column for Total is there but the format is unappealing.
Running the output through sqlite_formatter yields the desired result:

> cat query_with_totals.sql | sqlite3 -header counts.db | sqlite_formatter
Language files blank comment code Total
______________ _____ _____ _______ _____ _____
C++ 141 12786 17359 60378 90523
C/C++ Header 110 8566 17420 51502 77488
Bourne Shell 10 6351 6779 38264 51394
m4 11 2037 260 17980 20277
Python 30 1613 2486 4602 8701
MSBuild script 11 0 0 1711 1711
CMake 7 155 285 606 1046
make 5 127 173 464 764
Markdown 2 30 0 39 69

The next section,
Wrapping cloc in other scripts,
shows one way these commands can be combined
into a new utility program.

Example 2: Include a column for "Language" when running with --by-file.

Output from --by-file omits each file's language to save screen real estate;
file paths for large projects can be long and including an extra 20 or so
characters for a Language column can be excessive.

As an example, here are the first few lines of output using the same
code base as in Example 1:

> cloc --by-file yaml-cpp-yaml-cpp-0.5.3.tar.gz v 1.81 T=1.14 s (287.9 files/s, 221854.9 lines/s)
File blank comment code
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/configure 2580 2264 13691
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/configure 2541 2235 13446
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/fused-src/gtest/gtest.h 1972 4681 13408
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/fused-src/gmock/gmock.h 1585 3397 9216
yaml-cpp-yaml-cpp-0.5.3/test/integration/gen_emitter_test.cpp 999 0 8760
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/aclocal.m4 987 100 8712
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/m4/libtool.m4 760 65 7176
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/build-aux/ 959 1533 7169
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/fused-src/ 1514 3539 6390
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/fused-src/gtest/ 1312 2896 5384
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/test/ 1226 1091 5098
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/include/gtest/internal/gtest-param-util-generated.h 349 235 4559

The absence of language identification for each file
is a bit disappointing, but
this can be remedied with a custom column solution.

The first step, creating a database, matches that from Example 1 so
we'll go straight to the second step of creating the desired
SQL query. We'll store this one in the file by_file_with_language.sql:

-- file by_file_with_language.sql
select File, Language, nBlank as blank ,
nComment as comment,
nCode as code
from t order by code desc;

Our desired extra column appears when we pass this custom SQL query
through our database:

> cat by_file_with_language.sql | sqlite3 -header counts.db | sqlite_formatter
File Language blank comment code
__________________________________________________________________________________________________ ______________ _____ _______ _____
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/configure Bourne Shell 2580 2264 13691
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/configure Bourne Shell 2541 2235 13446
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/fused-src/gtest/gtest.h C/C++ Header 1972 4681 13408
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/fused-src/gmock/gmock.h C/C++ Header 1585 3397 9216
yaml-cpp-yaml-cpp-0.5.3/test/integration/gen_emitter_test.cpp C++ 999 0 8760
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/aclocal.m4 m4 987 100 8712
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/m4/libtool.m4 m4 760 65 7176
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/build-aux/ Bourne Shell 959 1533 7169
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/fused-src/ C++ 1514 3539 6390
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/fused-src/gtest/ C++ 1312 2896 5384
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/test/ C++ 1226 1091 5098
yaml-cpp-yaml-cpp-0.5.3/test/gmock-1.7.0/gtest/include/gtest/internal/gtest-param-util-generated.h C/C++ Header 349 235 4559


Wrapping cloc in other scripts ▲

More complex code counting solutions are possible by wrapping
cloc in scripts or programs. The "total lines" column from
example 1 of Custom Column Output
could be simplified to a single command with this shell script (on Linux):

# These commands must be in the user's $PATH:
# cloc
# sqlite3
# sqlite_formatter
if test $# -eq 0 ; then
echo "Usage: $0 [cloc arguments]"
echo " Run cloc to count lines of code with an additional"
echo " output column for total lines (code+comment+blank)."
cloc --sql 1 --sql-project x $@ | sqlite3 ${DBFILE}
SQL="select Language, count(File) as files ,
sum(nBlank) as blank ,
sum(nComment) as comment ,
sum(nCode) as code ,
sum(nBlank)+sum(nComment)+sum(nCode) as Total
from t group by Language order by code desc;
echo ${SQL} | sqlite3 -header ${DBFILE} | sqlite_formatter
rm ${DBFILE}

Saving the lines above to and making it
executable (chmod +x would let us do

./ yaml-cpp-yaml-cpp-0.5.3.tar.gz

to directly get

Language       files blank comment code  Total
______________ _____ _____ _______ _____ _____
C++ 141 12786 17359 60378 90523
C/C++ Header 110 8566 17420 51502 77488
Bourne Shell 10 6351 6779 38264 51394
m4 11 2037 260 17980 20277
Python 30 1613 2486 4602 8701
MSBuild script 11 0 0 1711 1711
CMake 7 155 285 606 1046
make 5 127 173 464 764
Markdown 2 30 0 39 69

Other examples:
* Count code from a specific branch of a web-hosted
git repository and send the results as a .csv email attachment:

git and UTF8 pathnames ▲

cloc's --git option may fail if you work with directory or
file names with UTF-8 characters (for example, see
issue 457).
The solution,,
is to apply this git configuration command:

git config --global core.quotepath off

Your console's font will need to be capable of displaying
Unicode characters.

Third Generation Language Scale Factors ▲

cloc versions before 1.50 by default computed, for the provided inputs, a
rough estimate of how many lines of code would be needed to write the
same code in a hypothetical third-generation computer language.
To produce this output one must now use the --3 switch.

Scale factors were derived from the 2006 version of language gearing ratios
listed at Mayes Consulting web site,, using this equation:

cloc scale factor for language X = 3rd generation default gearing ratio / language X gearing ratio

For example, cloc 3rd generation scale factor for DOS Batch = 80 / 128 = 0.625.

The biggest flaw with this approach is that gearing ratios are defined
for logical lines of source code not physical lines (which cloc counts).
The values in cloc's 'scale' and '3rd gen. equiv.' columns should be
taken with a large grain of salt.

options.txt configuration file ▲

If you find yourself using the same command line switches every
time you invoke cloc, you can save some typing by adding those
switches to the options.txt runtime configuration file.
cloc will look for this file in the following default locations:

# Linux, NetBSD, FreeBSD, macOS:

# Windows

Place each switch and arguments, if any, on a line by itself.
Lines prefixed with # symbol are ignored as comments and
blank lines are skipped. Leading hyphens on the switches are
optional. Here's a sample file:

# options.txt
--vcs git
v # verbose level 1
exclude-ext svg,html

The path to the options.txt file can also be specified
with the --config FILE switch.

Finally, if cloc finds an options.txt file in the same
directory as files given by any of these switches (in the
listed priority), it will use that configuration file
from that location:

  1. --list-file

  2. --exclude-list-file

  3. --read-lang-def

  4. --force-lang-def

  5. --diff-list-file

Run with --verbose to have cloc tell you which, if
any, options.txt file it uses.

Complex regular subexpression recursion limit ▲

cloc relies on the Regexp::Common module's regular expressions to remove
comments from source code. If comments are malformed, for example the
/* start comment marker appears in a C program without a corresponding */
marker, the regular expression engine could enter a recursive
loop, eventually triggering the warning
Complex regular subexpression recursion limit.

The most common cause for this warning is the existence of comment markers
in string literals. While language compilers and interpreters are smart
enough to recognize that "/*" (for example) is a string and not a comment,
cloc is fooled. File path globs, as in this line of JavaScript

var paths = globArray("**/*.js", {cwd: srcPath});

are frequent culprits.

In an attempt to overcome this problem, a different
algorithm which removes comment markers in strings can be enabled
with the --strip-str-comments switch. Doing so, however,
has drawbacks: cloc
will run more slowly and the output of --strip-comments
will contain strings that no longer match the input source.

Limitations ▲

Identifying comments within source code is trickier than one might expect.
Many languages would need a complete parser to be counted correctly.
cloc does not attempt to parse any of
the languages it aims to count and therefore is an imperfect tool.
The following are known problems:

  1. Lines containing both source code and comments are counted as lines of code.

  2. Comment markers within strings or
    are treated as actual comment markers and not string literals.
    For example the following lines of C code
    printf(" /* ");
    for (i = 0; i < 100; i++) {
    a += i;
    printf(" */ ");

    look to cloc like this:
    printf(" xxxxxxx
    xxxxxxx ");

    where xxxxxxx represents cloc's view of commented text.
    Therefore cloc counts the five lines as two lines of C code and three
    lines of comments (lines with both code and comment are counted as code).

    If you suspect your code has such strings, use the switch
    ``--strip-str-comments`` to switch to the algorithm which removes
    embedded comment markers. Its use will render the five lines above as

    printf("  ");
    for (i = 0; i < 100; i++) {
    a += i;
    printf(" ");

    and therefore return a count of five lines of code.
    See the
    [previous section](#complex-regular-subexpression-recursion-limit-)
    on drawbacks to using ``--strip-str-comments``.

  3. Embedded languages are not recognized. For example, an HTML file containing
    JavaScript will be counted entirely as HTML.

  4. Python docstrings can serve several purposes. They may
    contain documentation,
    comment out blocks of code, or they can be regular strings (when
    they appear on the right hand side of an assignment or as a function argument).
    cloc is unable to infer the meaning of docstrings by context; by default
    cloc treats all docstrings as comments. The switch
    treats all docstrings as code.

  5. Language definition files read with --read-lang-def or
    --force-lang-def must be plain ASCII text files.

  6. cloc treats compiler pragma's, for example #if / #endif, as code
    even if these are used to block lines of source from being compiled;
    the blocked lines still contribute to the code count.

  7. On Windows, cloc will fail with Can't cd to ... No such file or
    directory at /File/ if the code being scanned has
    file paths longer than 255 characters. A work-around is to run cloc
    from the Windows Subsystem for Linux (WSL).

Requesting Support for Additional Languages ▲

If cloc does not recognize a language you are interested in counting,
create a GitHub issue
requesting support for your language. Include this information:

  1. File extensions associated with the language. If the language does
    not rely on file extensions and instead works with fixed file names or
    with `#!` style program invocations, explain what those are.

  2. A description of how comments are defined.

  3. Links to sample code.

Reporting Problems ▲

If you encounter a problem with cloc, first check to see if
you're running with the latest version of the tool:

  cloc --version

If the version is older than the most recent release
at, download the
latest version and see if it solves your problem.

If the problem happens with the latest release, submit
a new issue at only
if you can supply enough information for anyone reading the
issue report to reproduce the problem.
That means providing

  1. the operating system you're running on

  2. the cloc command with all options

  3. the code you are counting (URL to a public git repo or zip file or
    tar file, et cetera)

The last item is generally problematic. If the code base is
proprietary or amounts to more than a few dozen kilobytes,
you'll need to try to reconstruct similar inputs or demonstrate
the problem with an existing public code base.

Problem reports that cannot be reproduced will be ignored and
eventually closed.

Acknowledgments ▲

Wolfram Rösler provided most of the code examples in the test suite.
These examples come from his Hello World collection.

Ismet Kursunoglu found errors with the MUMPS counter and provided
access to a computer with a large body of MUMPS code to test cloc.

Tod Huggins gave helpful suggestions for the Visual Basic filters.

Anton Demichev found a flaw with the JSP counter in cloc v0.76
and wrote the XML output generator for the --xml option.

Reuben Thomas pointed out that ISO C99 allows // as a comment
marker, provided code for the --no3 and --stdin-name
options, counting the m4 language,
and suggested several user-interface enhancements.

Michael Bello provided code for the --opt-match-f,
--opt-match-d, and --opt-not-match-d

Mahboob Hussain inspired the --original-dir and
--skip-uniqueness options, found a
bug in the duplicate file detection logic and improved the JSP filter.

Randy Sharo found and fixed an uninitialized variable bug for shell
scripts having only one line.

Steven Baker found and fixed a problem with the YAML output generator.

Greg Toth provided code to improve blank line detection in COBOL.

Joel Oliveira provided code to let --exclude-list-file handle
directory name exclusion.

Blazej Kroll provided code to produce an XSLT file, cloc-diff.xsl,
when producing XML output for the --diff option.

Denis Silakov enhanced the code which generates cloc.xsl when
using --by-file and --by-file-by-lang options, and
provided an XSL file that works with --diff output.

Andy ([email protected]) provided code to fix several bugs:
correct output of --counted
so that only files that are used in the code count appear and
that results are shown by language rather than file name;
allow --diff output from multiple runs to be summed
together with --sum-reports.

Jari Aalto created the initial version of cloc.1.pod and
maintains the Debian package for cloc.

Mikkel Christiansen ([email protected]) provided counter definitions
for Clojure and ClojureScript.

Vera Djuraskovic from
provided the

Gill Ajoft of Ajoft Softwares
provided the

Knowledge Team
provided the
Slovakian translation.

Erik Gooven Arellano Casillas provided an update to the MXML counter to
recognize ActionScript comments.

Gianluca Casati created the
cloc CPAN package.

Ryan Lindeman implemented the --by-percent feature.

Kent C. Dodds, @kentcdodds,
created and maintains the npm package of cloc.

Viktoria Parnak
provided the

Natalie Harmann provided the

Nithyal at Healthcare Administration Portal
provided the

Patricia Motosan
provided the

Gajk Melikyan provided the
provided the
Armenian translation

Hungarian translation
courtesy of Zsolt Boros.

Sietse Snel implemented the parallel
processing capability available with the --processes=N

The development of cloc was partially funded by the Northrop Grumman

Copyright ▲

Copyright (c) 2006-2018, Al Danial