=head1 Complete Programming
This document describes a set thougths and workflows that evolved from the
development of the Kephra editor.
The name is a reminder that we reach for the impossible,
because software is never complete. So please relax, enjoy and focus.
I strongly dislike strict rules imposed on me and make fun of people that produce
theories with three letter acronyms that try to be the answer to everything.
But now I am standing here, trying to "sell" you yet another programming methodology.
That IS irony, isn't it?
At least I have just a two letters acronym to sound even more important.
=head3 Basic Ideas
Well, several reasons brought me here.
1) I always wanted a sane balance between the old bureaucratic waterfall method,
and the modern extreme or agile programming, which is a bit too shortsighted
and accumulates additional work as the project grows.
It should be possible to combine sound planning, practicability and quick results.
2) All methodologies I know of overlook several aspects of the product.
That includes even "documentation driven development",
which gave important impulses for the creation of CP.
3) Documentation coverage and depth matters a great deal and writing it also
helps the developer.
4) The awesome powers of prototypes are known but still not fully used.
5) Writing tests is no silver bullet. It is used for several competing purposes
and in an unnatural way. That's why a lot of programer hate it.
6) New tools like hg or git allow new workflows which solve pending problems.
=head3 Main Goal
Highest aim of CP is the conscience the code is produced with. As result you get
=item a superb overall user experience
=item quality code
=item transparent project planning and status
=item room for experiments and changes without trouble
=item Honest and brief communication is key.
=item Every action is communication and has a productive and lasting (yet changable) result.
=item All code, data and metadata form one project and go into one repository.
=item They are all equally important, even if they arranged in a structure.
=item Respect (or change) the structure and avoid any duplication.
=item Don't plan new feature, document it from users point.
=item Note the stage it should be implemented in and its technical requirements.
=item These low level functionalities are tested first in a prototype.
=item Than follows a usability prototype.
=item When implementing a feature, envision and do all aspects at once.
=item Only after something went into the main programm a test is needed.
=item We rely heavily on branches and decentral version control systems.
Lets start with a small meditation, because Complete Programming (CP) relies on
embodying a mindset that is still a bit alien to some. In one word it is holistic -
meaning seeing the full (complete) picture. That includes the views of the user,
the programmer, the designer, tester and project manager.
That sounds demanding, but you do it already when you really care about something.
As a consequence we count an ugly icon, missing documentation or an badly
designed internal API equally as a bug as a crashing function. CP is designed
to deliver that high standard, without packing too much burden on the programmer.
Firstly: all aspects of a feature are planned at once - so they're in tune with
each other. And they are also get imlepemented together - when details are fresh
in your memory. And since that also includes comments and documentation, you will
later know what you have done and why, saving a lot of guesswork.
Secondly: we reduce duplicate work and duplicate data where possible. CP will
compel you to think about a problem several times, to ensure you understood its
implications. But every time you will produce a useful part of the project.
Every vital information goes into one well structured repository (again holistic!).
Your highten clarity and the low duplication also improve the quality of the
communication, that will mainly take place through your common work (repository).
And thirdly: as hinted, the development model knows several decoupled, well
defined stages that correlate with the natural flow of ideas into materialisation.
Any stage has only one goal that is set according to its possibilities and with
respect to its unavoidable limits. That's one reason why work gets less heavy.
(Joyful labour is important to get the deeper focus we need to achieve quality.)
The other reason: some stages allow free minded hacking and bring back the fun
you missed for so long. (Play and work balanced - that is holistic too).
The next chapter describe each stage in detail.
Initial planning in CP is mostly metaplanning. Of course you have to think about
the classic five questions:
* general vision of the program with use cases
* approx. feature list
* small roadmap (what featture into what stage)
* on which technology to rely (with fallback alternatives)
* what team we need (especially in corporate setting)
Take your time and write down answers in a structured and well readable manner,
because this gets already shipped into the repository as user docs. In a sense -
development has already started. Of course these docs will be presented to your
boss as the project proposal - better they are convincing and well formulated.
Please don't overdo your first plan, since most information you will gather along
the way. CP means that new information will get stored quickly in the appropriate
places - not that work targets change every week. CP knows well defined workflows
which spare the team from some agreements people call planning. Some planning will
be replaced by asynchronous interactions with results that are immediately visible
to all project member. But still there are meetings, a boss and something similar
After the rough goal is set and before any "real" coding begins - its time for
the announced metaplanning. Outline the abstraction layers of the program and
plan when to introduce them during developement. This is most critical, because
changes on these cause the most growth pain. Write some placesholder now into the
code base for all eyes to see and all devs to remember. The compiler will tell
you, if it's coherent. It also helps you to find the right names for important
details, so later they will be used consistantly in the docs, code and all the rest.
Second part of metaplanning is: Adapt this document, which is part of the programers
documentation, to the needs of the project. Maybe your team is smaller and several
roles will be played by one person or maybe your program has no GUI and you have
to redefine completeness of a feature. Maybe you want to tweak the role of tests -
by all means these are not the ten commandments.
And please give us a feedback when your tweaks seem to work well.
After the metaplanning do a small roadmap. In CP thats basically a list of several
planned stages of growth. Assign the features you already know about to a stage,
where it can be developed with least effort and all needed abstractions and support
libs exist. Try to group related features in one stage - so changes become more
managable. If your understanding is clear enough - assign the features to
substages that will be handled similar to a sprint.
Savor the magic of a new beginning and avoid the trap of doing things as usual.
As you saw, CP starts always by writing documentation, forwhy its usually better
to think before doing anything. Putting it down also helps to reflect on it.
And knowing that others will rely on it, should motivate you toward precision.
The first thoughts about anything are too vague to build a good program from.
Trial and error and even more thinking have to follow, before it gets useful.
So why not take what you get in the first minutes, put it in a readable form,
(which helps you to understand it deeper) and minimize with the result the effort
to write the docs after you did coding. Practically that is one trick in CP to
do more and better quality with less work. As your clairvoyance is most probably
not fully developed, you need to readjust the doc safter implementation.
Read details about that in chapter L<Main Program> and L<Itersations>.
Some devs feel writing docs is unproductive, a waste of time or should be done
by people more talented to do that. Even if good UI design is "self-explanatory",
missing docs are crippling the usability and not findable features are not
distinguishable from not existing ones. Thatswhy only documented features, libs
and API's are quality work. And since you wrote them, you should document them.
It would take more time to explain it anyway, especially to a none-techie. And
switching to the users perspective - that's what you need to do for writing good
docs - you also need to develope good software.
The reason why docs are considered in the "agile" field as less important,
(agile manifesto says: "Working software over comprehensive documentation")
is because they confuse it with overhead. In CP we only document software parts as
a first step toward their implementation. Planned ones are just noted in a roadmap.
Meetings get more productive, when everybody gets all information and there is
arguing about now to be made changes in the docs that will go online instantly.
Managers as well as customers will get high level user docs. They are the basis
of most communication. That's because the struggle for satisfying agreements,
a common language and a good product are the same thing. Unifying them makes just
sense, improves transparency, demands honesty and reduces overhead.
User docs, developer docs and comments, as well as the code itslef are means of
communication between people. They have to cooperate. All these parts should have
a common language and define key words and phrases in the exactly same way.
Less friction here is an enormous productivity boost - a further reason,
why they gotta be written altogether and by the same people.
One of the things Steve Jobs got right was: "Start with envisioning the user
experience and then work your way through the technicalities.". CP does that by
beginning with a bit of user documentation - followed by an explorative phase,
where developers find out how to realize it by writing prototypes.
These are small, quickly written programs that are test beds for very few things.
Usually do programer enjoy building them, because they demand less restrictions
than huge programs and allow much more fast changes and creativity.
This way ideas get stable without harming the "real code" and coders have more fun.
And if a prototype turns out to be a blind alley, it can be deleted without remorse.
Some may now intervene: "Why not just fork an experimental/feature branch,
if we using hg, git or arch anyway? Afterwards it can be just merged back by the
software and we don't have to write that code again in the main program.".
Well, like said - dealing with a huge program is much more hairy and distracting -
compared with that fresh clean start prototypes provide. And beside that -
the code that will flow back is not as much compared with the
many trials and errors that might pollute the history of the main project branch
if you merge it there. There is still rebase, that cleans it up, but you might
preserve that history in the right place, since there you learned your lessons.
Prototypes also serve an educational purpose. Programer often look up how they
did something in their own programs. That is much easier with nicely commented,
sorted and indexed prototypes that sit and wait unchanged in their own branch -
presenting just the essence how to solve one particular problem. Its a kind of
documentation from programmer for programer. But this works only if a solution
is also tidied up after its found. When you later want to test another idea just
clone or copy a similar prototype and you already got the needed minimal framework
supporting that kind of function.
A completed prototype has also a role similar to a test. It demonstrates that a
chunk of code works that way. You can't practically achieve that while have having
a lot dependencies that can introduce a failure into an otherwise working piece
of code. And even beside that - in CP we have a way to write prototypes,
where you can do heart transplant to the production stream without greater trouble.
CP knows two types of prototypes that are living in two seperate branches.
That's because the main program has to bridge two endpoints which are too different
to think about and optimize for at once. Theset are the hardware or the near
hardware software layer and the user. For the one side we have functional
prototypes and for the second use case prototypes.
=head4 Functional Prototypes
After starting with a users wish, we go to the opposite end of the spectrum,
to the building blocks of a program like computing a formula or writing a file.
Because it doesn't make sense to plan on details if we don't know if it can be
built or how. I mean experienced programmer know a lot, but there are far too
many fast changing libraries, techniques and domains to know them all. And even
if you use other peoples work - code can be buggy and docs might be unclear,
missing, lying or simply out of date. So only when you built it and it's running
you can be sure it runs. Therefor we do functional prototypes (FP).
A FP has only one file and should be as short as possible. It can be used to
check out how third party libraries work, but none of the projects libs should
be used if possible. FP need no exceptions, no logging and alike.
Just solve a core problem that is needed to implement a feature.
If there are two ways to implement, try both. Usually that's not much more work
and you can later make wiser decisions. Especially in discussion with the lead
programmer that knows the whole structure and maybe sees more implications of
these alternatives. That became only possible, because well commented FP are
easier to grok than main program code.
=head4 Use Case Prototypes
Back to the users perspective. Same basic rules apply, except now you may use
libs from the main project or some altered/simplified derivatives. Without these
you can't write this kind of programs quickly because they are far more complex
then functional prototypes. And it will be a good stress test for the libs too.
Since this will be a program with only one feature or a set of related ones,
you are free to model it to an ideal state, that maybe wouldn't fit to main branch.
That enables a realstic decision making process, how to wrench it into the whole,
or maybe even to alter the main thing to get it shipped this way.
This creative phase is also the perfct time to make the artwork (UI and icons)
and to figure out the key and mouse controls and possible options.
=head3 Main Program
Before we get to the new prepackaged feature, waiting to be integrated -
lets back up for a while - to the humble beginnings of the project.
As stated several times: CP doesn't recognizes a clear distinction between writing
documentation or building the program. Docs are been written as a first step of
programming and we write code to document ideas. That might include writing comments,
that marks and explains what is expected to happen at that place. But I mean also
creating files, inserting classes, methods, attributes - even variables,
that have a name and purpose we already could agree upon.
First of all this makes ideas very, very concrete. If you see your ideas like that,
you get a much better sense how the program will be look like. Spotting design
flaws or misleading names becomes easier - plus the compiler tells if the plan is
even doable. Don't underestimate the power of names. They are equally important as
the logical structure, helping to communicate the inner workings and the intentions
of the architect. In CP we spend some time just contemplating dummy code to find
the right hierarchies and names.
Maybe other people don't call that programming, but we certainly do.
Secondly: having this dummy code in place (just do it for the current stage),
team member will comprehend the current goals and where to put their work into
much clearer than from text desctiptions or boring UML-diagrams. Sources will tell
the current state of planning and also communicate changes. This also means:
if the architect changes his mind - he has to clean up the mess the change created.
Of course nobody is allowed to check in code that breaks tests.
Only completed features with their tests as one commit are acceptable. This means:
at this point we use short lived feature branches, that pick the the code from a
use case prototype and bring it into the program.
* tested code
* documentation (user, programmer, comments)
* logic (main program)
* supply libs
The rise of Extreme Programming (XP) gave also rise to Test Driven Development (TDD),
where you start the implementation by building tests. The rationale behind this:
You have to think about your goal by defining it with a small program. After that
it's easier to write the code and the new failing test motivates you to do so.
And as soon you brake a function or specification, the testing suite will yell
at you and you are spared of a very time consuming bug search 10 month later,
when angry user yell at you, because the program deleted their data but you already
forgot half the details of the program.
Problem with this theory: its so hard to write tests out of the blue, because your
notion of this task might be very vague at first. But even if you have a billiant
idea, there are still a lot of details to be added and altered while you grok
the algorithm, its use cases and its role in the larger scheme fully. And having
to rewrite the code and the tests several times is frustrating. Besides - TDD
is not even doable for GUI and other areas where the supporting structure has to
be built first to even run a test. A startling read about even more prescribed
madness in software testing is Elisabeth Hendricksons paper: "better testing —
worse quality?" at L<http://testobsessed.com/wp-content/uploads/2011/04/btwq.pdf>
But the other benefits of testing are still as worthwhile as describes by TDD people.
That's why in CP we write the tests hopefully once, after the dust is settled.
Because tests are so precise, writing them down will give us a precious opportunity
to rethink the code one last time from a logical and technical perspective.
All the other issues were already solved and we are now free to give special
attention to the corner cases we may have overlooked so far. Yet another case of
getting better results with less effort in CP.
One point that is still experimental, but worth considering - putting the tests
into the code in the same file as the code it tests. The test suite will then be
just a different "starter" for all this code, but no seperate dir tree of code.
That would be really holistic. Writing tests would be easier, even if good editors
can display several files beside each other and test would functions as an
additional comment, explaining what we we want to achieve here. In fact to the
parser of the main program test will appear as comments. This might be hard to
realize in some languages, but not problem in Perl 6 and even Perl 5 should have
at least one way to do that.
See more in L<lib/Kephra/Versioning.pod>