README

The code in this directory makes up the "git data miner," a simple hack
which attempts to figure things out from the revision history in a git
repository. 


INSTALLING GITDM

gitdm is a python script and doesn't need to be proper installed like other
normal programs. You just have to adjust your PATH variable, pointing it to
the directory of gitdm or alternatively create a symbolic link of the script
inside /usr/bin.

Before actually run gitdm you may want also to update the configuration file
(gitdm.config) with the needed information.


RUNNING GITDM

Run it like this:

   git log -p -M [details] | gitdm [options]

Alternatively, you can run with:

   git log --numstat -M [details] | gitdm -n [options]

The [details] tell git which changesets are of interest; the [options] can
be:

	-a	If a patch contains signoff lines from both Andrew Morton 
		and Linus Torvalds, omit Linus's.

	-b dir	Specify the base directory to fetch the configuration files.

	-c file Specify the name of the gitdm configuration file.  
	   	By default, "./gitdm.config" is used.

	-d	Omit the developer reports, giving employer information
         	only.

	-D	Rather than create the usual statistics, create a file (datelc.csv)
	    providing lines changed per day, where the first column displays
	    the changes happened only on that day and the second sums the day it
	    happnened with the previous ones. This option is suitable for
	    feeding to a tool like gnuplot.

	-h file	Generate HTML output to the given file

	-l num	Only list the top <num> entries in each report.

    -n  Use --numstat instead of generated patches to get the statistics.

	-o file	Write text output to the given file (default is stdout).

    -p prefix Dump out the database categorized by changeset and by file type.
	    It requires -n, otherwise it is not possible to get separated results.

	-r pat	Only generate statistics for changes to files whose 
	   	name matches the given regular expression.

	-s	Ignore Signed-off-by lines which match the author of 
		each patch.

	-t	Generate a report by type of contribution (code, documentation, etc.).
	    It requires -n, otherwise this option is ignored silently.


	-u 	Group all unknown developers under the "(Unknown)"
 	        employer. 

    -x file  Export raw statistics as CSV.

    -w  Aggregate the data by weeks instead of months in the
        CSV file when -x is used.

	-z 	Dump out the hacker database to "database.dump".

A typical command line used to generate the "who write 2.6.x" LWN articles
looks like:

    git log -p -M v2.6.19..v2.6.20 | \
	gitdm -u -s -a -o results -h results.html

or:

    git log --numstat -M v2.6.19..v2.6.20 | \
	gitdm -u -s -a -n -o results -h results.html

CONFIGURATION FILE

The main purpose of the configuration file is to direct the mapping of
email addresses onto employers.  Please note that the config file parser is
exceptionally stupid and unrobust at this point, but it gets the job done.  

Blank lines and lines beginning with "#" are ignored.  Everything else
specifies a file with some sort of mapping:

EmailAliases file

	Developers often post code under a number of different email
	addresses, but it can be desirable to group them all together in
	the statistics.  An EmailAliases file just contains a bunch of
	lines of the form:

		alias@address  canonical@address

	Any patches originating from alias@address will be treated as if
	they had come from canonical@address.

    It may happen that some people set their git user data in the
    following form: "joe.hacker@acme.org <Joe Hacker>". The 
    "Joe Hacker" is then considered as the email... but gitdm says
    it is a "Funky" email. An alias line in the following form can
    be used to alias these commits aliased to the correct email
    address:

        "Joe Hacker" joe.hacker@acme.org


EmailMap file

	Map email addresses onto employers.  These files contain lines
	like:

		[user@]domain  employer  [< yyyy-mm-dd]

	If the "user@" portion is missing, all email from the given domain
	will be treated as being associated with the given employer.  If a
	date is provided, the entry is only valid up to that date;
	otherwise it is considered valid into the indefinite future.  This
	feature can be useful for properly tracking developers' work when
	they change employers but do not change email addresses.


GroupMap file employer

	This is a variant of EmailMap provided for convenience; it contains
	email addresses only, all of which are associated with the given
	employer.

VirtualEmployer name
    nn% employer1
    ...
end

	This construct (which appears in the main configuration file)
    	allows causes the creation of a fake employer with the given
    	"name".  It directs that any contributions attributed to that
    	employer should be split to other (real) employers using the given
    	percentages.  The functionality works, but is primitive - there is,
	for example, no check to ensure that the percentages add up to
    	something rational.

FileTypeMap file

	Map file names/extensions onto file types.  These files contain lines
	like:

		order <type1>,<type2>,...,<typeN>

		filetype <type> <regex>
		...

	This construct allows fine graned reports by type of contribution
	(build, code, image, multimedia, documentation, etc.)

	Order is important because it is possible to have overlapping between
	filenames.  For instance, ltmain.sh fits better as 'build' instead of
	'code' (the filename instead of '\.sh$').  The first element in order
	has precedence over the next ones.


OTHER TOOLS

A few other tools have been added to this repository:

  treeplot
	Reads a set of commits, then generates a graphviz file charting the
	flow of patches into the mainline.  Needs to be smarter, but, then,
	so does everything else in this directory.

  findoldfiles
	Simple brute-force crawler which outputs the names of any files
	which have not been touched since the original (kernel) commit.

  committags
	I needed to be able to quickly associate a given commit with the
	major release which contains it.  First attempt used 
	"git tags --contains="; after it ran for a solid week, I concluded
	there must be a better way.  This tool just reads through the repo,
	remembering tags, and creating a Python dictionary containing the
	association.  The result is an ugly 10mb pickle file, but, even so,
	it's still a better way.

  linetags
	Crawls through a directory hierarchy, counting how many lines of
	code are associated with each major release.  Needs the pickle file
	from committags to get the job done.


NOTES AND CREDITS

Gitdm was written by Jonathan Corbet; many useful contributions have come
from Greg Kroah-Hartman.

Please note that this tool is provided in the hope that it will be useful,
but it is not put forward as an example of excellence in design or
implementation.  Hacking on gitdm tends to stop the moment it performs
whatever task is required of it at the moment.  Patches to make it less
hacky, less ugly, and more robust are welcome.

Jonathan Corbet
corbet@lwn.net