(  _`\       ( )_        ( )                            ( )    
        | |_) )  _ _ | ,_)   ___ | |__   _   _   _    _    _ __ | |/') 
        | ,__/'/'_` )| |   /'___)|  _ `\( ) ( ) ( ) /'_`\ ( '__)| , <  
        | |   ( (_| || |_ ( (___ | | | || \_/ \_/ |( (_) )| |   | |\`\ 
        (_)   `\__,_)`\__)`\____)(_) (_)`\___x___/'`\___/'(_)   (_) (_)

Patchwork is a tool that allows you to recursively add and remove diffs from a
project. It provides a simple interface for moving your code between various
different states. It's primary role is to be a software configuration
management tool (whatever the fuck that is) rather than to replace / be a
version control system. To change the feature set of the software rather and to
act as an alternative to version control's branching.

To give you an idea of how patchwork actually works, there is a file called
patchwork.ogv in the repo. It's a video that'll play in VLC. It shows creating
two patches, one depending on another.

There isn't much point elaborating on what it is. I don't even know what it is
myself. This readme is more of a rambling blog post than the description and
usage of this tool.

--

I absolutely love operating systems. They're the core of a modern computing.
They interact with the hardware, abstract it, provide a consist platform for
business logic and are full of intricate interesting algorithms. When I grew up
using Windows they were the user interface too. The single most common aspect
of all machines shaping how the user worked. The baseline, the commodity, the
one central, important interface on which everything sat. I used to dream of
being a Windows kernel developer. They were the equivalent of rock stars in my
mind.

Weird isn't it? What people think. Just plane fucking bizarre. After my degree
I was messing with Linux and obviously all that Windows kernel developer stuff
evaporated along with that stuff about the link between the OS and user
interface. However I do love operating systems still. I even started one
taken from an example on the internet. It's called Farmix and it's here on
github. It's a piece of crap though.

It's not going to be unique, just a boring half implemented *NIX clone
designed to show someone at Sun Microsystems (Oracle) so I can get a job
working on Solaris (if it still exists) ten years from now.

All I want is it to be as clean, small and easy to read as MINIX and also
as functional as Linux.

Not possible you say? It can't be simple and feature rich? It can't be easy to
read and heavily optimised? I had this idea. I start thinking about
writing a patching tool. In git you have a basic simplistic operating system and
you apply series of patches to it to get more features out. A 'faster
scheduler' patch. On top of that? A POSIX patch providing full capabilities and
a whole series of patches in between, managed by a program that allows me to
switch them on and off as simply as I want. The simplicity of MINIX at the base.
The feature set of Linux in a few key strokes. Being able to drill in and out
of complexity would make my OS easy to study and un-daunting to work with.

And this is it, in this github repository. The result of the idea. See what
makes my operating system unique (or will do) is not the operating system
itself but the tool that manipulates it and controls it allowing me to have
my cake and eat it. I can switch on arrays of patches to strongly manipulate the
way the OS works. The algorithms used, If it is realtime, if it's compatible
with Linux drivers. The patching system allows me to support _any_ direction
I want to take the software and allows me to pursue every route at once. In
one slightly drunk revelation I actually realised my operating system kernel
would be _more_ free than Linux itself. For years people complain about Linus's
OOM Killer. This feature in their memory management that causes it to kill
apps at random. Whilst anyone can modify the Linux kernel, add to it, maintain
their own copy, maybe steer it a certain way, Linux at it's heart had a team
to control the base line of features. They have a generic kernel to a certain
low extent. With Farmix, you don't even need that. I can accept any patch from
anyone and the patching tree allows everyone to build the operating system they
want. Oracle could provide steamlined kernel algorithms for a dedicated database
server and the user could also patch in some other commodity interface. Such
a level of granularity and control gives you the power of a dedicated machine
with widespread understanding of how the OS works and how to use its tools.

So you got a few trees of diffs that need to be applied on and off and when
changes are made to the code you need to test them as well, in all possible
states. Having simplicity and features in my software does mean a fuck ton of
complexity in managing the project behind the scenes but isn't that the job
of the project or the team? To manage complexity. You gotta do more work behind
the scenes to provide so much power but it is doable.

Regardless of me not being able to sell the idea to my friends I can't see
any reason not to try and write the program. I should finish it to see it
through to completion.

So I get halfway through writing patchwork and I kind of think to myself, this
isn't _that_ difficult, there isn't _that_ much to it.

In fact, it isn't really much more than a wrapper for diff and patch.

I stop working on the software about a month ago. There's something 'off' about
this whole project. I don't know why the idea is flawed per say at this point
but patchwork as a program is definitely a steaming pile of useless bullshit.
Fundamentally I can't see what it does. I forced myself to finish it this week
just to wrap it up.

I thought that even with the existence of #ifdefs in C code in git repos, there
is a clearly proof that branches aren't the solution to everything. The
problem is that #ifdefs are in the code are visible. Hidden diffs mean you
don't know what you're reading or how it lines up with your code.

But I don't really know anymore. I haven't used the product properly yet and I
don't think I will. Half of it doesn't work because I can't see it being useful
but I can't put my finger on the problem. Can anyone else?

I was telling this idea to an interviewer last week. He asked if it was like Quilt.
Something similar, written by Andrew Morton, ironically a Linux kernel developer.
The more I read into it, the more it looks identical to my software. In fact
his looks more sophisticated than mine because you can edit patches within the chain
although I don't think his patches are designed to stick around forever. I guess
the problem is the way the patches, although easily applied, hide the complexity of
the software in a way that #ifdef's don't.

In conclusion to my blog post, sometimes you have a good idea and sometimes you
don't. Sometimes you think you're on to something cool and then it fizzles
out. I took this project to completion for the sake of completion but I doubt it'll
make it out of alpha. I'm branding this a failure. A positive, at least I tried
something failure. Maybe something out of ashes? Probably not. Here's to the next
hair brained, open source, idea :-)

It's funny what people think.

----
Phillip Taylor. 5th May 2011