Introduction

SLOCCount (pronounced "sloc-count") is a suite of programs for counting physical source lines of code (SLOC) in potentially large software systems. Thus, SLOCCount is a "software metrics tool" or "software measurement tool". SLOCCount was developed by David A. Wheeler, originally to count SLOC in a GNU/Linux distribution, but it can be used for counting the SLOC of arbitrary software systems.

SLOCCount is known to work on Linux systems, and has been tested on Red Hat Linux versions 6.2, 7, and 7.1. SLOCCount should run on many other Unix-like systems (if Perl is installed), in particular, I would expect a *BSD system to work well. Windows users can run sloccount by first installing Cygwin. SLOCCount is much slower on Windows/Cygwin, and it's not as easy to install or use on Windows, but it works. Of course, feel free to upgrade to an open source Unix-like system (such as Linux or *BSD) instead :-).

SLOCCount can count physical SLOC for a wide number of languages. Listed alphabetically, they are Ada, Assembly (for many machines and assemblers), awk (including gawk and nawk), Bourne shell (and relatives such as bash, ksh, zsh, and pdksh), C, C++, C# (also called C-sharp or cs), C shell (including tcsh), COBOL, Expect, Fortran (including Fortran 90), Haskell, Java, lex (including flex), LISP (including Scheme), makefiles (though they aren't usually shown in final reports), Modula3, Objective-C, Pascal, Perl, PHP, Python, Ruby, sed, SQL (normally not shown), TCL, and Yacc. It can gracefully handle awkward situations in many languages, for example, it can determine the syntax used in different assembly language files and adjust appropriately, it knows about Python's use of string constants as comments, and it can handle various Perl oddities (e.g., perlpods, here documents, and Perl's _ _END_ _ marker). It even has a "generic" SLOC counter that you may be able to use count the SLOC of other languages (depending on the language's syntax).

SLOCCount can also take a large list of files and automatically categorize them using a number of different heuristics. The heuristics automatically determine if a file is a source code file or not, and if so, which language it's written in. For example, it knows that ".pc" is usually a C source file for an Oracle preprocessor, but it can detect many circumstances where it's actually a file about a "PC" (personal computer). For another example, it knows that ".m" is the standard extension for Objective-C, but it will check the file contents to see if really is Objective-C. It will even examine file headers to attempt to accurately determine the file's true type. As a result, you can analyze large systems completely automatically.

Finally, SLOCCount has some report-generating tools to collect the data generated, and then present it in several different formats and sorted different ways. The report-generating tool can also generate simple tab-separated files so data can be passed on to other analysis tools (such as spreadsheets and database systems).

SLOCCount will try to quickly estimate development time and effort given only the lines of code it computes, using the original Basic COCOMO model. This estimate can be improved if you can give more information about the project. See the discussion below about COCOMO, including intermediate COCOMO, if you want to improve the estimates by giving additional information about the project.

SLOCCount is open source software/free software (OSS/FS), released under the GNU General Public License (GPL), version 2; see the license below. The master web site for SLOCCount is http://www.dwheeler.com/sloccount. You can learn a lot about SLOCCount by reading the paper that caused its creation, available at http://www.dwheeler.com/sloc. Feel free to see my master web site at http://www.dwheeler.com, which has other material such as the Secure Programming for Linux and Unix HOWTO, my list of OSS/FS references, and my paper Why OSS/FS? Look at the Numbers! Please send improvements by email to dwheeler, at, dwheeler.com (DO NOT SEND SPAM - please remove the commas, remove the spaces, and change the word "at" into the at symbol).

The following sections first give a "quick start" (discussing how to use SLOCCount once it's installed), discuss basic SLOCCount concepts, how to install it, how to set your PATH, how to install source code on RPM-based systems if you wish, and more information on how to use the "sloccount" front-end. This is followed by material for advanced users: how to use SLOCCount tools individually (for when you want more control than the "sloccount" tool gives you), designer's notes, the definition of SLOC, and miscellaneous notes. The last sections states the license used (GPL) and gives hints on how to submit changes to SLOCCount (if you decide to make changes to the program).

Quick Start

Once you've installed SLOCCount (discussed below), you can measure an arbitrary program by typing everything after the dollar sign into a terminal session:

  $  sloccount topmost-source-code-directory

The directory listed and all its descendants will be examined. You'll see output while it calculates, culminating with physical SLOC totals and estimates of development time, schedule, and cost. If the directory contains a set of directories, each of which is a different project developed independently, use the "--multiproject" option so the effort estimations can correctly take this into account.

You can redisplay the data different ways by using the "--cached" option, which skips the calculation stage and re-prints previously computed information. You can use other options to control what's displayed: "--filecount" shows counts of files instead of SLOC, and "--details" shows the detailed information about every source code file. So, to display all the details of every file once you've previously calculated the results, just type:

  sloccount --cached --details

You'll notice that the default output ends with a request. If you use this data (e.g., in a report), please credit that data as being "generated using 'SLOCCount' by David A. Wheeler." I make no money from this program, so at least please give me some credit.

SLOCCount tries to ignore all automatically generated files, but its heuristics to detect this are necessarily imperfect (after all, even humans sometimes have trouble determining if a file was automatically genenerated). If possible, try to clean out automatically generated files from the source directories -- in many situations "make clean" does this.

There's more to SLOCCount than this, but first we'll need to explain some basic concepts, then we'll discuss other options and advanced uses of SLOCCount.

Basic Concepts

SLOCCount counts physical SLOC, also called "non-blank, non-comment lines". More formally, physical SLOC is defined as follows: ``a physical source line of code (SLOC) is a line ending in a newline or end-of-file marker, and which contains at least one non-whitespace non-comment character.'' Comment delimiters (characters other than newlines starting and ending a comment) are considered comment characters. Data lines only including whitespace (e.g., lines with only tabs and spaces in multiline strings) are not included.

In SLOCCount, there are 3 different directories:

1. The "source code directory", a directory containing the source code being measured (possibly in recursive subdirectories). The directories immediately contained in the source code directory will normally be counted separately, so it helps if your system is designed so that this top set of directories roughly represents the system's major components. If it doesn't, there are various tricks you can use to group source code into components, but it's more work. You don't need write access to the source code directory, but you do need read access to all files, and read and search (execute) access to all subdirectories.
2. The "bin directory", the directory containing the SLOCCount executables. By default, installing the program creates a subdirectory named "sloccount-VERSION" which is the bin directory. The bin directory must be part of your PATH.
3. The "data directory", which stores the analysis results. When measuring programs using "sloccount", by default this is the directory ".slocdata" inside your home directory. When you use the advanced SLOCCount tools directly, in many cases this must be your "current" directory. Inside the data directory are "data directory children" - these are subdirectories that contain a file named "filelist", and each child is used to represent a different project or a different major component of a project.

SLOCCount can handle many different programming languages, and separate them by type (so you can compare the use of each). Here is the set of languages, sorted alphabetically; common filename extensions are in parentheses, with SLOCCount's ``standard name'' for the language listed in brackets:

1. Ada (.ada, .ads, .adb, .pad) [ada]
2. Assembly for many machines and assemblers (.s, .S, .asm) [asm]
3. awk (.awk) [awk]
4. Bourne shell and relatives such as bash, ksh, zsh, and pdksh (.sh) [sh]
5. C (.c, .pc, .ec, .ecp) [ansic]
6. C++ (.C, .cpp, .cxx, .cc, .pcc) [cpp]
7. C# (.cs) [cs]
8. C shell including tcsh (.csh) [csh]
9. COBOL (.cob, .cbl, .COB, .CBL) [cobol]
10. Expect (.exp) [exp]
11. Fortran 77 (.f, .f77, .F, .F77) [fortran]
12. Fortran 90 (.f90, .F90) [f90]
13. Haskell (.hs, .lhs) [haskell]; deals with both types of literate files.
14. Java (.java) [java]
15. lex (.l) [lex]
16. LISP including Scheme (.cl, .el, .scm, .lsp, .jl) [lisp]
17. makefiles (makefile) [makefile]
18. ML (.ml, .ml3) [ml]
19. Modula3 (.m3, .mg, .i3, .ig) [modula3]
20. Objective-C (.m) [objc]
21. Pascal (.p, .pas) [pascal]
22. Perl (.pl, .pm, .perl) [perl]
23. PHP (.php, .php[3456], .inc) [php]
24. Python (.py) [python]
25. Ruby (.rb) [ruby]
26. sed (.sed) [sed]
27. sql (.sql) [sql]
28. TCL (.tcl, .tk, .itk) [tcl]
29. Yacc (.y) [yacc]

Installing SLOCCount

Obviously, before using SLOCCount you'll need to install it. SLOCCount depends on other programs, in particular perl, bash, a C compiler (gcc will do), and md5sum (you can get a useful md5sum program in the ``textutils'' package on many Unix-like systems), so you'll need to get them installed if they aren't already.

If your system uses RPM version 4 or greater to install software (e.g., Red Hat Linux 7 or later), just download the SLOCCount RPM and install it using a normal installation command; from the text line you can use:

  rpm -Uvh sloccount*.rpm

Everyone else will need to install from a tar file, and Windows users will have to install Cygwin before installing sloccount.

If you're using Windows, you'll need to first install Cygwin. By installing Cygwin, you'll install an environment and a set of open source Unix-like tools. Cygwin essentially creates a Unix-like environment in which sloccount can run. You may be able to run parts of sloccount without Cygwin, in particular, the perl programs should run in the Windows port of Perl, but you're on your own - many of the sloccount components expect a Unix-like environment. If you want to install Cygwin, go to the Cygwin main page and install it. If you're using Cygwin, install it to use Unix newlines, not DOS newlines - DOS newlines will cause odd errors in SLOCCount (and probably other programs, too). I have only tested a "full" Cygwin installation, so I suggest installing everything. If you're short on disk space, at least install binutils, bash, fileutils, findutils, gcc, grep, gzip, make, man, perl, readline, sed, sh-utils, tar, textutils, unzip, and zlib; you should probably install vim as well, and there may be other dependencies as well. By default Cygwin will create a directory C:\cygwin\home\NAME, and will set up the ability to run Unix programs (which will think that the same directory is called /home/NAME). Now double-click on the Cygwin icon, or select from the Start menu the selection Programs / Cygnus Solutions / Cygwin Bash shell; you'll see a terminal screen with a Unix-like interface. Now follow the instructions (next) for tar file users.

If you're installing from the tar file, download the file (into your home directory is fine). Unpacking the file will create a subdirectory, so if you want the unpacked subdirectory to go somewhere special, "cd" to where you want it to go. Most likely, your home directory is just fine. Now gunzip and untar SLOCCount (the * replaces the version #) by typing this at a terminal session:

  gunzip -c sloccount*.tar.gz | tar xvf -

Now you need to compile the few compiled programs in the "bin directory" so SLOCCount will be ready to go. First, cd into the newly-created bin directory, by typing:

  cd sloccount*

You may then need to override some installation settings. You can can do this by editing the supplied makefile, or alternatively, by providing options to "make" whenever you run make. The supplied makefile assumes your C compiler is named "gcc", which is true for most Linux systems, *BSD systems, and Windows systems using Cygwin. If this isn't true, you'll need to set the "CC" variable to the correct value (e.g., "cc"). You can also modify where the files are stored; this variable is called PREFIX and its default is /usr/local (older versions of sloccount defaulted to /usr).

If you're using Windows and Cygwin, you must override one of the installation settings, EXE_SUFFIX, for installation to work correctly. One way to set this value is to edit the "makefile" file so that the line beginning with "EXE_SUFFIX" reads as follows:

  EXE_SUFFIX=.exe

Finally, compile the few compiled programs in it by typing "make":

  make

  make EXE_SUFFIX=.exe

If you want, you can install sloccount for system-wide use without using the RPM version. Windows users using Cygwin should probably do this, particularly if they chose a "local" installation. To do this, first log in as root (Cygwin users don't need to do this for local installation). Edit the makefile to match your system's conventions, if necessary, and then type "make install":

  make install

A note for Cygwin users (and some others): some systems, including Cygwin, don't set up the environment quite right and thus can't display the manual pages as installed. The problem is that they forget to search /usr/local/share/man for manual pages. If you want to read the installed manual pages, type this into a Bourne-like shell:

  MANPATH=/usr/local/share/man:/usr/share/man:/usr/man
  export MANPATH

  setenv MANPATH "/usr/local/share/man:/usr/share/man:/usr/man"

Installing The Source Code To Measure

Obviously, you must install the software source code you're counting, so somehow you must create the "source directory" with the source code to measure. You must also make sure that permissions are set so the software can read these directories and files.

For example, if you're trying to count the SLOC for an RPM-based Linux system, install the software source code by doing the following as root (which will place all source code into the source directory /usr/src/redhat/BUILD):

1. Install all source rpm's:

       mount /mnt/cdrom
       cd /mnt/cdrom/SRPMS
       rpm -ivh *.src.rpm
   

2. Remove RPM spec files you don't want to count:

       cd ../SPECS
       (look in contents of spec files, removing what you don't want)
   

3. build/prep all spec files:

       rpm -bp *.spec
   

4. Set permissions so the source files can be read by all:

       chmod -R a+rX /usr/src/redhat/BUILD
   

Here's an example of how to download source code from an anonymous CVS server. Let's say you want to examine the source code in GNOME's "gnome-core" directory, as stored at the CVS server "anoncvs.gnome.org". Here's how you'd do that:

1. Set up site and login parameters:

     export CVSROOT=':pserver:anonymous@anoncvs.gnome.org:/cvs/gnome'
   

2. Log in:

     cvs login
   

3. Check out the software (copy it to your local directory), using mild compression to save on bandwidth:

     cvs -z3 checkout gnome-core
   

Of course, if you have a non-anonymous account, you'd set CVSROOT to reflect this. For example, to log in using the "pserver" protocol as ACCOUNT_NAME, do:

  export CVSROOT=':pserver:ACCOUNT_NAME@cvs.gnome.org:/cvs/gnome'

You may need root privileges to install the source code and to give another user permission to read it, but please avoid running the sloccount program as root. Although I know of no specific reason this would be a problem, running any program as root turns off helpful safeguards.

Although SLOCCount tries to detect (and ignore) many cases where programs are automatically generated, these heuristics are necessarily imperfect. So, please don't run any programs that generate other programs - just do enough to get the source code prepared for counting. In general you shouldn't run "make" on the source code, and if you have, consider running "make clean" or "make really_clean" on the source code first. It often doesn't make any difference, but identifying those circumstances is difficult.

SLOCCount will not automatically uncompress files that are compressed/archive files (such as .zip, .tar, or .tgz files). Often such files are just "left over" old versions or files that you're already counting. If you want to count the contents of compressed files, uncompress them first.

SLOCCount also doesn't delve into files using "literate programming" techniques, in part because there are too many incompatible formats that implement it. Thus, run the tools to extract the code from the literate programming files before running SLOCCount. Currently, the only exception to this rule is Haskell.

Setting your PATH

Otherwise, in Bourne-shell variants, type:

    PATH="$PATH:the directory with SLOCCount's executable files"
    export PATH

    setenv PATH "$PATH:the directory with SLOCCount's executable files"

Using SLOCCount: The Basics

1. if directory has at least two subdirectories, then those subdirectories will be used as the breakdown (see the example below).
2. If the single directory contains files as well as directories (or if you give sloccount some files as parameters), those files will be assigned to the directory "top_dir" so you can tell them apart from other directories.
3. If there's a subdirectory named "src", then that subdirectory is again broken down, with all the further subdirectories prefixed with "src_". So if directory "X" has a subdirectory "src", which contains subdirectory "modules", the program will report a separate count from "src_modules".

You can also give "sloccount" a list of directories, in which case the report will be broken down by these directories (make sure that the basenames of these directories differ). SLOCCount normally considers all descendants of these directories, though unless told otherwise it ignores symbolic links.

This is all easier to explain by example. Let's say that we want to measure Apache 1.3.12 as installed using an RPM. Once it's installed, we just type:

 sloccount /usr/src/redhat/BUILD/apache_1.3.12

SLOC	Directory	SLOC-by-Language (Sorted)
24728   src_modules     ansic=24728
19067   src_main        ansic=19067
8011    src_lib         ansic=8011
5501    src_os          ansic=5340,sh=106,cpp=55
3886    src_support     ansic=2046,perl=1712,sh=128
3823    src_top_dir     sh=3812,ansic=11
3788    src_include     ansic=3788
3469    src_regex       ansic=3407,sh=62
2783    src_ap          ansic=2783
1378    src_helpers     sh=1345,perl=23,ansic=10
1304    top_dir         sh=1304
104     htdocs          perl=104
31      cgi-bin         sh=24,perl=7
0       icons           (none)
0       conf            (none)
0       logs            (none)


ansic:       69191 (88.85%)
sh:           6781 (8.71%)
perl:         1846 (2.37%)
cpp:            55 (0.07%)


Total Physical Source Lines of Code (SLOC)                   = 77873
Estimated Development Effort in Person-Years (Person-Months) = 19.36 (232.36)
 (Basic COCOMO model, Person-Months = 2.4 * (KSLOC**1.05))
Estimated Schedule in Years (Months)                         = 1.65 (19.82)
 (Basic COCOMO model, Months = 2.5 * (person-months**0.38))
Estimated Average Number of Developers  (Effort/Schedule)    = 11.72
Total Estimated Cost to Develop                              = $ 2615760
 (average salary = $56286/year, overhead = 2.4).

Please credit this data as "generated using 'SLOCCount' by David A. Wheeler."

Interpreting this should be straightforward. The Apache directory has several subdirectories, including "htdocs", "cgi-bin", and "src". The "src" directory has many subdirectories in it ("modules", "main", and so on). Code files directly contained in the main directory /usr/src/redhat/BUILD/apache_1.3.12 is labelled "top_dir", while code directly contained in the src subdirectory is labelled "src_top_dir". Code in the "src/modules" directory is labelled "src_modules" here. The output shows each major directory broken out, sorted from largest to smallest. Thus, the "src/modules" directory had the most code of the directories, 24728 physical SLOC, all of it in C. The "src/helpers" directory had a mix of shell, perl, and C; note that when multiple languages are shown, the list of languages in that child is also sorted from largest to smallest.

Below the per-component set is a list of all languages used, with their total SLOC shown, sorted from most to least. After this is the total physical SLOC (77,873 physical SLOC in this case).

Next is an estimation of the effort and schedule (calendar time) it would take to develop this code. For effort, the units shown are person-years (with person-months shown in parentheses); for schedule, total years are shown first (with months in parentheses). When invoked through "sloccount", the default assumption is that all code is part of a single program; the "--multiproject" option changes this to assume that all top-level components are independently developed programs. When "--multiproject" is invoked, each project's efforts are estimated separately (and then summed), and the schedule estimate presented is the largest estimated schedule of any single component.

By default the "Basic COCOMO" model is used for estimating effort and schedule; this model includes design, code, test, and documentation time (both user/admin documentation and development documentation). See below for more information on COCOMO as it's used in this program.

Next are several numbers that attempt to estimate what it would have cost to develop this program. This is simply the amount of effort, multiplied by the average annual salary and by the "overhead multiplier". The default annual salary is $56,286 per year; this value was from the ComputerWorld, September 4, 2000's Salary Survey of an average U.S. programmer/analyst salary in the year 2000. You might consider using other numbers (ComputerWorld's September 3, 2001 Salary Survey found an average U.S. programmer/analyst salary making $55,100, senior systems programmers averaging $68,900, and senior systems analysts averaging $72,300).

Overhead is much harder to estimate; I did not find a definitive source for information on overheads. After informal discussions with several cost analysts, I determined that an overhead of 2.4 would be representative of the overhead sustained by a typical software development company. As discussed in the next section, you can change these numbers too.

You may be surprised by the high cost estimates, but remember, these include design, coding, testing, documentation (both for users and for programmers), and a wrap rate for corporate overhead (to cover facilities, equipment, accounting, and so on). Many programmers forget these other costs and are shocked by the high figures. If you only wanted to know the costs of the coding, you'd need to get those figures.

Note that if any top-level directory has a file named PROGRAM_LICENSE, that file is assumed to contain the name of the license (e.g., "GPL", "LGPL", "MIT", "BSD", "MPL", and so on). If there is at least one such file, sloccount will also report statistics on licenses.

Note: sloccount internally uses MD5 hashes to detect duplicate files, and thus needs some program that can compute MD5 hashes. Normally it will use "md5sum" (available, for example, as a GNU utility). If that doesn't work, it will try to use "md5" and "openssl", and you may see error messages in this format:

 Can't exec "md5sum": No such file or directory at
     /usr/local/bin/break_filelist line 678, <CODE_FILE> line 15.
 Can't exec "md5": No such file or directory at
     /usr/local/bin/break_filelist line 678, <CODE_FILE> line 15.

Options

There are several options that control which files are selected for counting:

 --duplicates   Count all duplicate files as normal files
 --crossdups    Count duplicate files if they're in different data directory
                children.
 --autogen      Count automatically generated files
 --follow       Follow symbolic links (normally they're ignored)
 --addlang      Add languages to be counted that normally aren't shown.
 --append       Add more files to the data directory

Possibly the most important option is "--cached". Normally, when sloccount runs, it computes a lot of information and stores this data in a "data directory" (by default, "~/.slocdata"). The "--cached" option tells sloccount to use data previously computed, greatly speeding up use once you've done the computation once. The "--cached" option can't be used along with the options used to select what files should be counted. You can also select a different data directory by using the "--datadir" option.

There are many options for controlling the output:

 --filecount     Show counts of files instead of SLOC.
 --details       Present details: present one line per source code file.
 --wide          Show "wide" format.  Ignored if "--details" selected
 --multiproject  Assume each directory is for a different project
                 (this modifies the effort estimation calculations)
 --effort F E    Change the effort estimation model, so that it uses
                 F as the factor and E as the exponent.
 --schedule F E  Change the schedule estimation model, so that it uses
                 F as the factor and E as the exponent.
 --personcost P  Change the average annual salary to P.
 --overhead O    Change the annual overhead to O.
 --              End of options

Basically, the first time you use sloccount, if you're measuring a set of projects (not a single project) you might consider using "--crossdups" instead of the defaults. Then, you can redisplay data quickly by using "--cached", combining it with options such as "--filecount". If you want to send the data to another tool, use "--details".

If you're measuring a set of projects, you probably ought to pass the option "--multiproject". When "--multiproject" is used, efforts are computed for each component separately and summed, and the time estimate used is the maximum single estimated time.

The "--details" option dumps the available data in 4 columns, tab-separated, where each line represents a source code file in the data directory children identified. The first column is the SLOC, the second column is the language type, the third column is the name of the data directory child (as it was given to get_sloc_details), and the last column is the absolute pathname of the source code file. You can then pipe this output to "sort" or some other tool for further analysis (such as a spreadsheet or RDBMS).

You can change the parameters used to estimate effort using "--effort". For example, if you believe that in the environment being used you can produce 2 KSLOC/month scaling linearly, then that means that the factor for effort you should use is 1/2 = 0.5 month/KSLOC, and the exponent for effort is 1 (linear). Thus, you can use "--effort 0.5 1".

You can also set the annual salary and overheads used to compute estimated development cost. While "$" is shown, there's no reason you have to use dollars; the unit of development cost is the same unit as the unit used for "--personcost".

More about COCOMO

By default SLOCCount uses a very simple estimating model for effort and schedule: the basic COCOMO model in the "organic" mode (modes are more fully discussed below). This model estimates effort and schedule, including design, code, test, and documentation time (both user/admin documentation and development documentation). Basic COCOMO is a nice simple model, and it's used as the default because it doesn't require any information about the code other than the SLOC count already computed.

However, basic COCOMO's accuracy is limited for the same reason - basic COCOMO doesn't take a number of important factors into account. If you have the necessary information, you can improve the model's accuracy by taking these factors into account. You can at least quickly determine if the right "mode" is being used to improve accuracy. You can also use the "Intermediate COCOMO" and "Detailed COCOMO" models that take more factors into account, and are likely to produce more accurate estimates as a result. Take these estimates as just that - estimates - they're not grand truths. If you have the necessary information, you can improve the model's accuracy by taking these factors into account, and pass this additional information to sloccount using its "--effort" and "--schedule" options (as discussed in options).

To use the COCOMO model, you first need to determine if your application's mode, which can be "Organic", "embedded", or "semidetached". Most software is "organic" (which is why it's the default). Here are simple definitions of these modes:

• Organic: Relatively small software teams develop software in a highly familiar, in-house environment.  It has a generally stable development environment, minimal need for innovative algorithms, and requirements can be relaxed to avoid extensive rework.
• Semidetached: This is an intermediate step between organic and embedded. This is generally characterized by reduced flexibility in the requirements.
• Embedded: The project must operate within tight (hard-to-meet) constraints, and requirements and interface specifications are often non-negotiable. The software will be embedded in a complex environment that the software must deal with as-is.

• Organic: effort factor = 2.4, exponent = 1.05; schedule factor = 2.5, exponent = 0.38
• Semidetached: effort factor = 3.0, exponent = 1.12; schedule factor = 2.5, exponent = 0.35
• Embedded: effort factor = 3.6, exponent = 1.20; schedule factor = 2.5, exponent = 0.32

• Organic: effort base factor = 2.3, exponent = 1.05; schedule factor = 2.5, exponent = 0.38
• Semidetached: effort base factor = 3.0, exponent = 1.12; schedule factor = 2.5, exponent = 0.35
• Embedded: effort base factor = 2.8, exponent = 1.20; schedule factor = 2.5, exponent = 0.32

For example, imagine that you're examining a fairly simple application that meets the "organic" requirements. Organic projects have a base factor of 2.3 and exponents of 1.05, as noted above. We then examine all the factors to determine a corrected base factor. For this example, imagine that we determine the values of these cost drivers are as follows:

Cost Drivers		Ratings
ID	Driver Name	Rating	Multiplier
RELY	Required software reliability	Low - easily recovered losses	0.88
DATA	Database size	Low	0.94
CPLX	Product complexity	Nominal	1.00
TIME	Execution time constraint	Nominal	1.00
STOR	Main storage constraint	Nominal	1.00
VIRT	Virtual machine (HW and OS) volatility	Low (major change every 12 months, minor every month)	0.87
TURN	Computer turnaround time	Nominal	1.00
ACAP	Analyst capability	Nominal (55th percentile)	1.00
AEXP	Applications experience	Nominal (3 years)	1.00
PCAP	Programmer capability	Nominal (55th percentile)	1.00
VEXP	Virtual machine experience	High (3 years)	0.90
LEXP	Programming language experience	High (3 years)	0.95
MODP	Use of "modern" programming practices (e.g. structured programming)	High (Routine use)	0.82
TOOL	Use of software tools	Nominal (basic tools)	1.00
SCED	Required development schedule	Nominal	1.00

So, starting with the base factor (2.3 in this case), and then multiplying the driver values, we'll compute a final factor of: By multiplying these driver values together in this example, we compute:

2.3*0.88*0.94*1*1*1*0.87*1.00*1*1*1*0.90*0.95*0.82*1*1

The detailed COCOMO model requires breaking information down further.

For more information about the original COCOMO model, including the detailed COCOMO model, see the book Software Engineering Economics by Barry Boehm.

You may be surprised by the high cost estimates, but remember, these include design, coding, testing (including integration and testing), documentation (both for users and for programmers), and a wrap rate for corporate overhead (to cover facilities, equipment, accounting, and so on). Many programmers forget these other costs and are shocked by the high cost estimates.

If you want to know a subset of this cost, you'll need to isolate just those figures that you're trying to measure. For example, let's say you want to find the money a programmer would receive to do just the coding of the units of the program (ignoring wrap rate, design, testing, integration, and so on). According to Boehm's book (page 65, table 5-2), the percentage varies by product size. For effort, code and unit test takes 42% for small (2 KSLOC), 40% for intermediate (8 KSLOC), 38% for medium (32 KSLOC), and 36% for large (128 KSLOC). Sadly, Boehm doesn't separate coding from unit test; perhaps 50% of the time is spent in unit test in traditional proprietary development (including fixing bugs found from unit test). If you want to know the income to the programmer (instead of cost to the company), you'll also want to remove the wrap rate. Thus, a programmer's income to only write the code for a small program (circa 2 KSLOC) would be 8.75% (42% x 50% x (1/2.4)) of the default figure computed by SLOCCount.

In other words, less than one-tenth of the cost as computed by SLOCCount is what actually would be made by a programmer for a small program for just the coding task. Note that a proprietary commercial company that bid using this lower figure would rapidly go out of business, since this figure ignores the many other costs they have to incur to actually develop working products. Programs don't arrive out of thin air; someone needs to determine what the requirements are, how to design it, and perform at least some testing of it.

There's another later estimation model for effort and schedule called "COCOMO II", but COCOMO II requires logical SLOC instead of physical SLOC. SLOCCount doesn't currently measure logical SLOC, so SLOCCount doesn't currently use COCOMO II. Contributions of code to compute logical SLOC and then optionally use COCOMO II will be gratefully accepted.

Counting Specific Files

If you want to count a specific subset, you can use the "--details" option to list individual files, pipe this into "grep" to select the files you're interested in, and pipe the result to my tool "print_sum" (which reads lines beginning with numbers, and returns the total of those numbers). If you've already done the analysis, an example would be:

  sloccount --cached --details | grep "/some/subdirectory/" | print_sum

If you just want to count specific files, and you know what language they're in, you can just invoke the basic SLOC counters directly. By convention the simple counters are named "LANGUAGE_count", and they take on the command line a list of the source files to count. Here are some examples:

  c_count *.c *.cpp *.h  # Count C and C++ in current directory.
  asm_count *.S          # Count assembly.

There is also "generic_count", which takes as its first parameter the ``comment string'', followed by a list of files. The comment string begins a comment that ends at the end of the line. Sometimes, if you have source for a language not listed, generic_count will be sufficient.

The basic SLOC counters will send output to standard out, one line per file (showing the SLOC count and filename). The assembly counter shows some additional information about each file. The basic SLOC counters always complete their output with a line saying "Total:", followe by a line with the total SLOC count.

Countering Problems and Handling Errors

 count_unknown_ext

One error report that you may see is:

  c_count ERROR - terminated in string in (filename)

Note that c_count really does have to keep track of whether or not it's a string. For example, this is three lines of code, not two, because the ``comment'' is actually in string data:

 a = "hello
 /* this is not a comment */
 bye";

Usually this error means you have code that won't compile given certain #define settings. E.G., XFree86 has a line of code that's actually wrong (it has a string that's not terminated), but people don't notice because the #define to enable it is not usually set. Legitimate code can trigger this message, but code that triggers this message is horrendously formatted and is begging for problems.

In either case, the best way to handle the situation is to modify the source code (slightly) so that the code's intent is clear (by making sure that double-quotes balance). If it's your own code, you definitely should fix this anyway. You need to look at the double-quote (") characters. One approach is to just grep for double-quote, and look at every line for text that isn't terminated, e.g., printf("hello %s, myname);

SLOCcount reports warnings when an unusually large number of duplicate files are reported. A large number of duplicates may suggest that you're counting two different versions of the same program as though they were independently developed. You may want to cd into the data directory (usually ~/.slocdata), cd into the child directories corresponding to each component, and then look at their dup_list.dat files, which list the filenames that appeared to be duplicated (and what they duplicate with).

Adding Support for New Languages

1. To detect a new language, you'll need to modify the program break_filelist. If the filename extension is reliable, you can modify the array %file_extensions, which maps various filename extensions into languages. If your needs are more complex, you'll need to modify the code (typically in functions get_file_type or file_type_from_contents) so that the correct file type is determined. For example, if a file with a given filename extension is only sometimes that type, you'll need to write code to examine the file contents.
2. You'll need to create a SLOC counter for that language type. It must have the name XYZ_count, where XYZ is the standard name for the language.

   For some languages, you may be able to use the ``generic_count'' program to implement your counter - generic_count takes as its first argument the pattern which identifies comment begins (which continue until the end of the line); the other arguments are the files to count. Thus, the LISP counter looks like this:

    #!/bin/sh
    generic_count ';' $@
   

   The generic_count program won't work correctly if there are multiline comments (e.g., C) or multiline string constants. If your language is identical to C/C++'s syntax in terms of string constant definitions and commenting syntax (using // or /* .. */), then you can use the c_count program - in this case, modify compute_sloc_lang so that the c_count program is used.

   Otherwise, you'll have to devise your own counting program. The program must generate files with the same format, e.g., for every filename passed as an argument, it needs to return separate lines, where each line presents the SLOC for that file, a space, and the filename. (Note: the assembly language counter produces a slightly different format.) After that, print "Total:" on its own line, and the actual SLOC total on the following (last) line.

Advanced SLOCCount Use

Here's how to manually create a "data directory" to hold intermediate results, and how to invoke each tool in sequence (with discussion of options):

1. Set your PATH to include the SLOCCount "bin directory", as discussed above.
2. Make an empty "data directory" (where all intermediate results will be stored); you can pick any name and location you like for this directory. Here, I'll use the name "data":

       mkdir ~/data
   

3. Change your current directory to this "data directory":

       cd ~/data
   

   The rest of these instructions assume that your current directory is the data directory. You can set up many different data directories if you wish, to analyze different source programs or analyze the programs in different ways; just "cd" to the one you want to work with.
4. (Optional) Some of the later steps will produce a lot of output while they're running. If you want to capture this information into a file, use the standard "script" command do to so. For example, "script run1" will save the output of everything you do into file "run1" (until you type control-D to stop saving the information). Don't forget that you're creating such a file, or it will become VERY large, and in particular don't type any passwords into such a session. You can store the script in the data directory, or create a subdirectory for such results - any data directory subdirectory that doesn't have the special file "filelist" is not a "data directory child" and is thus ignored by the later SLOCCount analysis routines.
5. Now initialize the "data directory". In particular, initialization will create the "data directory children", a set of subdirectories equivalent to the source code directory's top directories. Each of these data directory children (subdirectories) will contain a file named "filelist", which lists all filenames in the corresponding source code directory. These data directory children will also eventually contain intermediate results of analysis, which you can check for validity (also, having a cache of these values speeds later analysis steps).

   You use the "make_filelists" command to initialize a data directory. For example, if your source code is in /usr/src/redhat/BUILD, run:

      make_filelists /usr/src/redhat/BUILD/*
   

   Internally, make_filelists uses "find" to create the list of files, and by default it ignores all symbolic links. However, you may need to follow symbolic links; if you do, give make_filelists the "--follow" option (which will use find's "-follow" option). Here are make_filelists' options:

    --follow         Follow symbolic links
    --datadir D      Use this data directory
    --skip S         Skip basenames named S
    --prefix P       When creating children, prepend P to their name.
    --               No more options
   

   Although you don't normally need to do so, if you want certain files to not be counted at all in your analysis, you can remove data directory children or edit the "filelist" files to do so. There's no need to remove files which aren't source code files normally; this is handled automatically by the next step.

   If you don't have a single source code directory where the subdirectories represent the major components you want to count separately, you can still use the tool but it's more work. One solution is to create a "shadow" directory with the structure you wish the program had, using symbolic links (you must use "--follow" for this to work). You can also just invoke make_filelists multiple times, with parameters listing the various top-level directories you wish to include. Note that the basenames of the directories must be unique.

   If there are so many directories (e.g., a massive number of projects) that the command line is too long, you can run make_filelists multiple times in the same directory with different arguments to create them. You may find "find" and/or "xargs" helpful in doing this automatically. For example, here's how to do the same thing using "find":

    find /usr/src/redhat/BUILD -maxdepth 1 -mindepth 1 -type d \
           -exec make_filelists {} \;
   

6. Categorize each file. This means that we must determine which files contain source code (eliminating auto-generated and duplicate files), and of those files which language each file contains. The result will be a set of files in each subdirectory of the data directory, where each file represents a category (e.g., a language).

      break_filelist *
   

   At this point you might want to examine the data directory subdirectories to ensure that "break_filelist" has correctly determined the types of the various files. In particular, the "unknown" category may have source files in a language SLOCCount doesn't know about. If the heuristics got some categorization wrong, you can modify the break_filelist program and re-run break_filelist.

   By default break_filelist removes duplicates, doesn't count automatically generated files as normal source code files, and only gives some feedback. You can change these defaults with the following options:

    --duplicates   Count all duplicate files as normal files
    --crossdups    Count duplicate files if they're in different data directory
                   children (i.e., in different "filelists")
    --autogen      Count automatically generated files
    --verbose      Present more verbose status information while processing.
   

   Duplicate control in particular is an issue; you probably don't want duplicates counted, so that's the default. Duplicate files are detected by determining if their MD5 checksums are identical; the "first" duplicate encountered is the only one kept. Normally, since shells sort directory names, this means that the file in the alphabetically first child directory is the one counted. You can change this around by listing directories in the sort order you wish followed by "*"; if the same data directory child is requested for analysis more than once in a given execution, it's skipped after the first time. So, if you want any duplicate files with child directory "glibc" to count as part of "glibc", then you should provide the data directory children list as "glibc *".

   Beware of choosing something other than "*" as the parameter here, unless you use the "--duplicates" or "--crossdups" options. The "*" represents the list of data directory children to examine. Since break_filelist skips duplicate files identified in a particular run, if you run break_filelist on only certain children, some duplicate files won't be detected. If you're allowing duplicates (via "--duplicates" or "--crossdups"), then this isn't a problem. Or, you can use the ``--duplistfile'' option to store and retrieve hashes of files, so that additional files can be handled.

   If there are so many directories that the command line is too long, you can run break_filelist multiple times and give it a subset of the directories each time. You'll need to use one of the duplicate control options to do this. I would suggest using "--crossdups", which means that duplicates inside a child will only be counted once, eliminating at least some of the problems of duplicates. Here's the equivalent of "break_filelist *" when there are a large number of subdirectories:

    find . -maxdepth 1 -mindepth 1 -type d -exec break_filelist --crossdups {} \;
   

   Indeed, for all of the later commands where "*" is listed as the parameter in these instructions (for the list of data directory children), just run the above "find" command and replace "break_filelist --crossdups" with the command shown.
7. (Optional) If you're not very familiar with the program you're analyzing, you might not be sure that "break_filelist" has correctly identified all of the files. In particular, the system might be using an unexpected programming language or extension not handled by SLOCCount. If this is your circumstance, you can just run the command:

    count_unknown_ext
   

   (note that this command is unusual - it doesn't take any arguments, since it's hard to imagine a case where you wouldn't want every directory examined). Unlike the other commands discussed, this one specifically looks at ${HOME}/.slocdata. This command presents a list of extensions which are unknown to break_filelist, with the most common ones listed first. The output format is a name, followed by the number of instances; the name begins with a "." if it's an extension, or, if there's no extension, it begins with "/" followed by the base name of the file. break_filelist already knows about common extensions such as ".gif" and ".png", as well as common filenames like "README". You can also view the contents of each of the data directory children's files to see if break_filelist has correctly categorized the files.
8. Now compute SLOC and filecounts for each language; you can compute for all languages at once by calling:

      compute_all *
   

   If you only want to compute SLOC for a specific language, you can invoke compute_sloc_lang, which takes as its first parameter the SLOCCount name of the language ("ansic" for C, "cpp" for C++, "ada" for Ada, "asm" for assembly), followed by the list of data directory children. Note that these names are a change from version 1.0, which called the master program "compute_all", and had "compute_*" programs for each language.

   Notice the "*"; you can replace the "*" with just the list of data directory children (subdirectories) to compute, if you wish. Indeed, you'll notice that nearly all of the following commands take a list of data directory children as arguments; when you want all of them, use "*" (as shown in these instructions), otherwise, list the ones you want.

   When you run compute_all or compute_sloc_lang, each data directory child (subdirectory) is consulted in turn for a list of the relevant files, and the SLOC results are placed in that data directory child. In each child, the file "LANGUAGE-outfile.dat" lists the information from the basic SLOC counters. That is, the oufile lists the SLOC and filename (the assembly outfile has additional information), and ends with a line saying "Total:" followed by a line showing the total SLOC of that language in that data directory child. The file "all-physical.sloc" has the final total SLOC for every language in that child directory (i.e., it's the last line of the outfile).

9. (Optional) If you want, you can also use USC's CodeCount. I've had trouble with these programs, so I don't do this normally. However, you're welcome to try - they support logical SLOC measures as well as physical ones (though not for most of the languages supported by SLOCCount). Sadly, they don't seem to compile in gcc without a lot of help, they used fixed-width buffers that make me nervous, and I found a number of bugs (e.g., it couldn't handle "/* text1 *//* text2 */" in C code, a format that's legal and used often in the Linux kernel). If you want to do this, modify the files compute_c_usc and compute_java_usc so they point to the right directories, and type:

    compute_c_usc *
   

10. Now you can analyze the results. The main tool for presenting SLOCCount results is "get_sloc", e.g,:

      get_sloc * | less
    

    The get_sloc program takes many options, including:

     --filecount    Display number of files instead of SLOC (SLOC is the default)
     --wide         Use "wide" format instead (tab-separated columns)
     --nobreak      Don't insert breaks in long lines
     --sort  X      Sort by "X", where "X" is the name of a language
                    ("ansic", "cpp", "fortran", etc.), or "total".
                    By default, get_sloc sorts by "total".
     --nosort       Don't sort - just present results in order of directory
                    listing given.
     --showother    Show non-language totals (e.g., # duplicate files).
     --oneprogram   When computing effort, assume that all files are part of
                    a single program.  By default, each subdirectory specified
                    is assumed to be a separate, independently-developed program.
     --noheader     Don't show the header
     --nofooter     Don't show the footer (the per-language values and totals)
    

    Note that unlike the "sloccount" tool, get_sloc requires the current directory to be the data directory.

    If you're displaying SLOC, get_sloc will also estimate the time it would take to develop the software using COCOMO (using its "basic" model). By default, this figure assumes that each of the major subdirectories was developed independently of the others; you can use "--oneprogram" to make the assumption that all files are part of the same program. The COCOMO model makes many other assumptions; see the paper at http://www.dwheeler.com/sloc for more information.

    If you need to do more analysis, you might want to use the "--wide" option and send the data to another tool such as a spreadsheet (e.g., gnumeric) or RDBMS (e.g., PostgreSQL). Using the "--wide" option creates tab-separated data, which is easier to import. You may also want to use the "--noheader" and/or "--nofooter" options to simplify porting the data to another tool.

    Note that in version 1.0, "get_sloc" was called "get_data".

    If you have so many data directory children that you can't use "*" on the command line, get_sloc won't be as helpful. Feel free to patch get_sloc to add this capability (as another option), or use get_sloc_detail (discussed next) to feed the data into another tool.

11. (Optional) If you just can't get the information you need from get_sloc, then you can get the raw results of everything and process the data yourself. I have a little tool to do this, called get_sloc_details. You invoke it in a similar manner:

    get_sloc_details *
    

Designer's Notes

Here are some ``designer's notes'' on how SLOCCount works, including what it can handle.

The program break_filelist has categories for each programming language it knows about, plus the special categories ``not'' (not a source code file), ``auto'' (an automatically-generated file and thus not to be counted), ``zero'' (a zero-length file), ``dup'' (a duplicate of another file as determined by an md5 checksum), and ``unknown'' (a file which doesn't seem to be a source code file nor any of these other categories). It's a good idea to examine the ``unknown'' items later, checking the common extensions to ensure you have not missed any common types of code.

The program break_filelist uses lots of heuristics to correctly categorize files. Here are few notes about its heuristics:

1. break_filelist first checks for well-known extensions (such as .gif) that cannot be program files, and for a number of common generated filenames.
2. It then peeks at the first few lines for "#!" followed by a legal script name. Sometimes it looks further, for example, many Python programs invoke "env" and then use it to invoke python.
3. If that doesn't work, it uses the extension to try to determine the category. For a number of languages, the extension is not reliable, so for those languages it examines the file contents and uses a set of heuristics to determine if the file actually belongs to that category.
4. Detecting automatically generated files is not easy, and it's quite conceivable that it won't detect some automatically generated files. The first 15 lines are examined, to determine if any of them include at the beginning of the line (after spaces and possible comment markers) one of the following phrases (ignoring upper and lower case distinctions): ``generated automatically'', ``automatically generated'', ``this is a generated file'', ``generated with the (something) utility'', or ``do not edit''.
5. A number of filename conventions are used, too. For example, any ``configure'' file is presumed to be automatically generated if there's a ``configure.in'' file in the same directory.
6. To eliminate duplicates, the program keeps md5 checksums of each program file. Any given md5 checksum is only counted once. Build directories are processed alphabetically, so if the same file content is in both directories ``a'' and ``b'', it will be counted only once as being part of ``a'' unless you make other arrangements. Thus, some data directory children with names later in the alphabet may appear smaller than would make sense at first glance. It is very difficult to eliminate ``almost identical'' files (e.g., an older and newer version of the same code, included in two separate packages), because it is difficult to determine when two ``similar'' files are essentially the same file. Changes such as the use of pretty-printers and massive renaming of variables could make small changes seem large, while the small files might easily appear to be the ``same''. Thus, files with different contents are simply considered different.
7. If all else fails, the file is placed in the ``unknown'' category for later analysis.

One complicating factor is that I wished to separate C, C++, and Objective-C code, but a header file ending with ``.h'' or ``.hpp'' file could be any of these languages. In theory, ``.hpp'' is only C++, but I found that in practice this isn't true. I developed a number of heuristics to determine, for each file, what language a given header belonged to. For example, if a given directory has exactly one of these languages (ignoring header files), the header is assumed to belong to that category as well. Similarly, if there is a body file (e.g., ".c") that has the same name as the header file, then presumably the header file is of the same language. Finally, a header file with the keyword ``class'' is almost certainly not a C header file, but a C++ header file; otherwise it's assumed to be a C file.

None of the SLOC counters fully parse the source code; they just examine the code using simple text processing patterns to count the SLOC. In practice, by handling a number of special cases this seems to be fine. Here are some notes on some of the language counters; the language name is followed by common extensions in parentheses and the SLOCCount name of the language in brackets:

1. Ada (.ada, .ads, .adb) [ada]: Comments begin with "--".
2. Assembly (.s, .S, .asm) [asm]: Assembly languages vary greatly in the comment character they use, so my counter had to handle this variance. The assembly language counter (asm_count) first examines the file to determine if C-style ``/*'' comments and C preprocessor commands (e.g., ``#include'') are used. If both ``/*'' and ``*/'' are in the file, it's assumed that C-style comments are being used (since it is unlikely that both would be used as something else, say as string data, in the same assembly language file). Determining if a file used the C preprocessor was trickier, since many assembly files do use ``#'' as a comment character and some preprocessor directives are ordinary words that might be included in a human comment. The heuristic used is as follows: if #ifdef, #endif, or #include are used, the C preprocessor is used; or if at least three lines have either #define or #else, then the C preprocessor is used. No doubt other heuristics are possible, but this at least seems to produce reasonable results. The program then determines what the comment character is by identifying which punctuation mark (from a set of possible marks) is the most common non-space initial character on a line (ignoring ``/'' and ``#'' if C comments or preprocessor commands, respectively, are used). Once the comment character has been determined, and it's been determined if C-style comments are allowed, the lines of code are counted in the file.
3. awk (.awk) [awk]: Comments begin with "#".
4. C (.c) [ansic]: Both traditional C comments (/* .. */) and C++ (//) style comments are supported. Although the older ANSI and ISO C standards didn't support // style comments, in practice many C programs have used them for some time, and the C99 standard includes them. The C counter understands multi-line strings, so comment characters (/* .. */ and //) are treated as data inside strings. Conversely, the counter knows that any double-quote characters inside a comment does not begin a C/C++ string.
5. C++ (.C, .cpp, .cxx, .cc) [cpp]: The same counter is used for both C and C++. Note that break_filelist does try to separate C from C++ for purposes of accounting between them.
6. C# (.cs): The same counter is used as for C and C++. Note that there are no "header" filetypes in C#.
7. C shell (.csh) [csh]: Comments begin with "#".
8. COBOL (.cob, .cbl) [cobol]: SLOCCount detects if a "freeform" command has been given; until such a command is given, fixed format is assumed. In fixed format, comments have a "*" or "/" in column 7 or column 1; any line that's not a comment, and has a nonwhitespace character after column 7 (the indicator area) is counted as a source line of code. In a freeform style, any line beginning with optional whitespace and then "*" or "/" is considered a comment; any noncomment line with a nonwhitespace characeter is counted as SLOC.
9. Expect (.exp) [exp]: Comments begin with "#".
10. Fortran 77 (.f, .f77, .F, .F77) [fortran]: Comment-only lines are lines where column 1 character = C, c, *, or !, or where ! is preceded only by white space.
11. Fortran 90 (.f90, .F90) [f90]: Comment-only lines are lines where ! is preceded only by white space.
12. Haskell (.hs) [haskell]: This counter handles block comments {- .. -} and single line comments (--); pragmas {-# .. -} are counted as SLOC. This is a simplistic counter, and can be fooled by certain unlikely combinations of block comments and other syntax (line-ending comments or strings). In particular, "Hello {-" will be incorrectly interpreted as a comment block begin, and "{- -- -}" will be incorrectly interpreted as a comment block begin without an end. Literate files are detected by their extension, and the style (TeX or plain text) is determined by searching for a \begin{code} or ">" at the beginning of lines. See the Haskell 98 report section on literate Haskell for more information.
13. Java (.java) [java]: Java is counted using the same counter as C and C++.
14. lex (.l) [lex]: Uses traditional C /* .. */ comments. Note that this does not use the counter as C/C++ internally, since it's quite legal in lex to have "//" (where it is NOT a comment).
15. LISP (.cl, .el, .scm, .lsp, .jl) [lisp]: Comments begin with ";".
16. ML (.ml, .mli, .mll, mly) [ml]: Comments nest and are enclosed in (* .. *).
17. Modula3 (.m3, .mg, .i3, .ig) [modula3]: Comments are enclosed in (* .. *).
18. Objective-C (.m) [objc]: Comments are old C-style /* .. */ comments.
19. Pascal (.p, .pas) [pascal]: Comments are enclosed in curly braces {} or (*..*). This counter has known weaknesses; see the BUGS section of the manual page for more information.
20. Perl (.pl, .pm, .perl) [perl]: Comments begin with "#". Perl permits in-line ``perlpod'' documents, ``here'' documents, and an __END__ marker that complicate code-counting. Perlpod documents are essentially comments, but a ``here'' document may include text to generate them (in which case the perlpod document is data and should be counted). The __END__ marker indicates the end of the file from Perl's viewpoint, even if there's more text afterwards.
21. PHP (.php, .php[3456], .inc) [php]: Code is counted as PHP code if it has a .php file extension; it's also counted if it has an .inc extension and looks like PHP code. SLOCCount does not count PHP code embedded in HTML files normally, though its lower-level routines can do so if you want to (use php_count to do this). Any of the various ways to begin PHP code can be used (<? .. ?>, <?php .. ?>, <script language="php"> .. </script>, or even <% .. %>). Any of the PHP comment formats (C, C++, and shell) can be used, and any string constant formats ("here document", double quote, and single quote) can be used as well.
22. Python (.py) [python]: Comments begin with "#". Python has a convention that, at the beginning of a definition (e.g., of a function, method, or class), an unassigned string can be placed to describe what's being defined. Since this is essentially a comment (though it doesn't syntactically look like one), the counter avoids counting such strings, which may have multiple lines. To handle this, strings which started the beginning of a line were not counted. Python also has the ``triple quote'' operator, permitting multiline strings; these needed to be handled specially. Triple quote stirngs are normally considered as data, regardless of content, unless they were used as a comment about a definition.
23. Ruby (.rb) [ruby]: Comments begin with "#".
24. sed (.sed) [sed]: Comments begin with "#". Note that these are "sed-only" files; many uses of sed are embeded in shell scripts (and are categorized as shell scripts in those cases).
25. shell (.sh) [sh]: Comments begin with "#". Note that I classify ksh, bash, and the original Bourne shell sh together, because they have very similar syntaxes. For example, in all of these shells, setting a variable is expressed as "varname=value", while C shells use the use "set varname=value".
26. TCL (.tcl, .tk, .itk) [tcl]: Comments begin with "#".
27. Yacc (.y) [yacc]: Yacc is counted using the same counter as C and C++.

Much of the code is written in Perl, since it's primarily a text processing problem and Perl is good at that. Many short scripts are Bourne shell scripts (it's good at short scripts for calling other programs), and the basic C/C++ SLOC counter is written in C for speed.

I originally named it "SLOC-Count", but I found that some web search engines (notably Google) treated that as two words. By naming it "SLOCCount", it's easier to find by those who know the name of the program.

SLOCCount only counts physical SLOC, not logical SLOC. Logical SLOC counting requires much more code to implement, and I needed to cover a large number of programming languages.

Definition of SLOC

This tool measures ``physical SLOC.'' Physical SLOC is defined as follows: ``a physical source line of code (SLOC) is a line ending in a newline or end-of-file marker, and which contains at least one non-whitespace non-comment character.'' Comment delimiters (characters other than newlines starting and ending a comment) are considered comment characters. Data lines only including whitespace (e.g., lines with only tabs and spaces in multiline strings) are not included.

To make this concrete, here's an example of a simple C program (it strips ANSI C comments out). On the left side is the running SLOC total, where "-" indicates a line that is not considered a physical "source line of code":

 1    #include <stdio.h>
 -    
 -    /* peek at the next character in stdin, but don't get it */
 2    int peek() {
 3     int c = getchar();
 4     ungetc(c, stdin);
 5     return c;
 6    }
 -    
 7    main() {
 8     int c;
 9     int incomment = 0;  /* 1 = we are inside a comment */
 -    
10     while ( (c = getchar()) != EOF) {
11        if (!incomment) {
12          if ((c == '/') && (peek() == '*')) {incomment=1;}
13        } else {
14          if ((c == '*') && (peek() == '/')) {
15               c= getchar(); c=getchar(); incomment=0;
16          }
17        }
18        if ((c != EOF) && !incomment) {putchar(c);}
19     }
20    }

Robert E. Park et al.'s Software Size Measurement: A Framework for Counting Source Statements (Technical Report CMU/SEI-92-TR-20) presents a set of issues to be decided when trying to count code. The paper's abstract states:

This report presents guidelines for defining, recording, and reporting two frequently used measures of software sizeŃ physical source lines and logical source statements. We propose a general framework for constructing size definitions and use it to derive operational methods for reducing misunderstandings in measurement results.

Using Park's framework, here is how physical lines of code are counted:

1. Statement Type: I used a physical line-of-code as my basis. I included executable statements, declarations (e.g., data structure definitions), and compiler directives (e.g., preprocessor commands such as #define). I excluded all comments and blank lines.
2. How Produced: I included all programmed code, including any files that had been modified. I excluded code generated with source code generators, converted with automatic translators, and those copied or reused without change. If a file was in the source package, I included it; if the file had been removed from a source package (including via a patch), I did not include it.
3. Origin: You select the files (and thus their origin).
4. Usage: You selects the files (and thus their usage), e.g., you decide if you're going to include additional applications able to run on the system but not included with the system.
5. Delivery: You'll decide what code to include, but of course, if you don't have the code you can't count it.
6. Functionality: This tool will include both operative and inoperative code if they're mixed together. An example of intentionally ``inoperative'' code is code turned off by #ifdef commands; since it could be turned on for special purposes, it made sense to count it. An example of unintentionally ``inoperative'' code is dead or unused code.
7. Replications: Normally, duplicate files are ignored, unless you use the "--duplicates" or "--crossdups" option. The tool will count ``physical replicates of master statements stored in the master code''. This is simply code cut and pasted from one place to another to reuse code; it's hard to tell where this happens, and since it has to be maintained separately, it's fair to include this in the measure. I excluded copies inserted, instantiated, or expanded when compiling or linking, and I excluded postproduction replicates (e.g., reparameterized systems).
8. Development Status: You'll decide what code should be included (and thus the development status of the code that you'll accept).
9. Languages: You can see the language list above.
10. Clarifications: I included all statement types. This included nulls, continues, no-ops, lone semicolons, statements that instantiate generics, lone curly braces ({ and }), and labels by themselves.

Thus, SLOCCount generally follows Park's ``basic definition'', but with the following exceptions depending on how you use it:

1. How Produced: By default, this tool excludes duplicate files and code generated with source code generators. After all, the COCOMO model states that the only code that should be counted is code ``produced by project personnel'', whereas these kinds of files are instead the output of ``preprocessors and compilers.'' If code is always maintained as the input to a code generator, and then the code generator is re-run, it's only the code generator input's size that validly measures the size of what is maintained. Note that while I attempted to exclude generated code, this exclusion is based on heuristics which may have missed some cases. If you want to count duplicates, use the "--autogen", "--duplicates", and/or "--crossdups" options. If you want to count automatically generated files, pass the "--autogen" option mentioned above.
2. Origin: You can choose what source code you'll measure. Normally physical SLOC doesn't include an unmodified ``vendor-supplied language support library'' nor a ``vendor-supplied system or utility''. However, if this is what you are measuring, then you need to include it. If you include such code, your set will be different than the usual ``basic definition.''
3. Functionality: I included counts of unintentionally inoperative code (e.g., dead or unused code). It is very difficult to automatically detect such code in general for many languages. For example, a program not directly invoked by anything else nor installed by the installer is much more likely to be a test program, which you may want to include in the count (you often would include it if you're estimating effort). Clearly, discerning human ``intent'' is hard to automate.

Otherwise, this counter follows Park's ``basic definition'' of a physical line of code, even down to Park's language-specific definitions where Park defined them for a language.

Miscellaneous Notes

There are other undocumented analysis tools in the original tar file. Most of them are specialized scripts for my circumstances, but feel free to use them as you wish.

If you're packaging this program, don't just copy every executable into the system "bin" directory - many of the files are those specialized scripts. Just put in the bin directory every executable documented here, plus the the files they depend on (there aren't that many). See the RPM specification file to see what's actually installed.

You have to take any measure of SLOC (including this one) with a large grain of salt. Physical SLOC is sensitive to the format of source code. There's a correlation between SLOC and development effort, and some correlation between SLOC and functionality, but there's absolutely no correlation between SLOC and either "quality" or "value".

A problem of physical SLOC is that it's sensitive to formatting, and that's a legitimate (and known) problem with the measure. However, to be fair, logical SLOC is influenced by coding style too. For example, the following two phrases are semantically identical, but will have different logical SLOC values:

   int i, j;  /* 1 logical SLOC */

   int i;     /* 2 logical SLOC, but it does the same thing */
   int j;

If you discover other information that can be divided up by data directory children (e.g., the license used), it's probably best to add that to each subdirectory (e.g., as a "license" file in the subdirectory). Then you can modify tools like get_sloc to add them to their display.

I developed SLOCCount for my own use, not originally as a community tool, so it's certainly not beautiful code. However, I think it's serviceable - I hope you find it useful. Please send me patches for any improvements you make!

You can't use this tool as-is with some estimation models, such as COCOMO II, because this tool doesn't compute logical SLOC. I certainly would accept code contributions to add the ability to measure logical SLOC (or related measures such as Cyclomatic Complexity and Cyclomatic density); selecting them could be a compile-time option. However, measuring logical SLOC takes more development effort, so I haven't done so; see USC's "CodeCount" for a set of code that measures logical SLOC for some languages (though I've had trouble with CodeCount - in particular, its C counter doesn't correctly handle large programs like the Linux kernel).

SLOCCount License

Here is the SLOCCount License; the file COPYING contains the standard GPL version 2 license:

=====================================================================
SLOCCount
Copyright (C) 2000-2001 David A. Wheeler (dwheeler, at, dwheeler.com)

This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
=====================================================================

While it's not formally required by the license, please give credit to me and this software in any report that uses results generated by it.

This document was written by David A. Wheeler (dwheeler, at, dwheeler.com), and is (C) Copyright 2001 David A. Wheeler. This document is covered by the license (GPL) listed above.

The license does give you the right to use SLOCCount to analyze proprietary programs.

Related Tools

One available toolset is CodeCount. I tried using this toolset, but I eventually gave up. It had too many problems handling the code I was trying to analyze, and it does a poor job automatically categorizing code. It also has no support for many of today's languages (such as Python, Perl, Ruby, PHP, and so on). However, it does a lot of analysis and measurements that SLOCCount doesn't do, so it all depends on your need. Its license appeared to be open source, but it's quite unusual and I'm not enough of a lawyer to be able to confirm that.

Another tool that's available is LOCC. It's available under the GPL. It can count Java code, and there's experimental support for C++. LOCC is really intended for more deeply analyzing each Java file; what's particularly interesting about it is that it can measure "diffs" (how much has changed). See A comparative review of LOCC and CodeCount.

CCCC is a tool which analyzes C++ and Java files and generates a report on various metrics of the code. Metrics supported include lines of code, McCabe's complexity, and metrics proposed by Chidamber & Kemerer and Henry & Kafura. (You can see Time Littlefair's comments). CCCC is in the public domain. It reports on metrics that sloccount doesn't, but sloccount can handle far more computer languages.

Submitting Changes

The GPL license doesn't require you to submit changes you make back to its maintainer (currently me), but it's highly recommended and wise to do so. Because others will send changes to me, a version you make on your own will slowly because obsolete and incompatible. Rather than allowing this to happen, it's better to send changes in to me so that the latest version of SLOCCount also has the features you're looking for. If you're submitting support for new languages, be sure that your chnage correctly ignores files that aren't in that new language (some filename extensions have multiple meanings). You might want to look at the TODO file first.

When you send changes to me, send them as "diff" results so that I can use the "patch" program to install them. If you can, please send ``unified diffs'' -- GNU's diff can create these using the "-u" option.