Clean Code

Notes about Clean Coding


Last Updated: February 02, 2021 by Pepe Sandoval



Want to show support?

If you find the information in this page useful and want to show your support, you can make a donation

Use PayPal

This will help me to create more stuff and fix the existent content... or probably your money will be used to buy beer


Clean code

My personal notes from Uncle Bob clean code lectures and video episodes, they have been paraphrased or written in my own words just for my personal reference, in some cases adding my personal opinions about certain topics

Intro to clean coding

  • We make a mess because we want to go fast but the only way to go fast is to go well

  • Adding people doesn’t make things go fast because it is the old code that is actually training the new people so the code slows programmers

  • No one writes clean code at first attempt, once the code works it most likely be a mess, after that it is when you need to clean it

  • Having code that works is only half the job. It will take you half of your time to make it work, clean it will take the other half

  • You are not done when it works, you are done when it is right

  • Clean code does one thing well

  • You must write code that other people can maintain, making it work is not enough

  • It is more important that your peers understand the code your write than the computer understand the code you wrote
    • If I have code that doesn't work but I understand it, someone probably will be able to make it work,
    • If I don’t understand the code but works perfectly as soon as we need to add extra functionality or the requirement changes the code is useless
  • Unclean code characteristics:
    • Rigidity: A system is rigid when that system requires us to make many changes to many different places in order to fix a single bug or add a new feature, this rigidity usually makes the system unpredictable (It is hard to know what will we break when we change something)
    • Inseparability: An inseparable system is a system when the various parts of that system cannot be independently used in some other system.
    • Opacity: A system is opaque when it is badly structured so it is hard to read, hard to change and hard understand the intention of the code or how it works
  • Clean code allows the reader to exit early when reading it

Names & Naming Convention

  • Use intention-reveling names, choose a name that reveal the intention of the thing you are trying to name. In other words stuff should describe itself for example it is better to have variable that describe itself than a variable followed by a comment with a description

  • Choose names that communicate your intentions in the code and are pronounceable

  • Encodings like Hungarian Notation are a thing of the past, they just obfuscate the code
    • If you want to keep track of the types in your variables make it explicit but without abbreviations or prefixes
  • Declare explanatory variables, the purpose is to explain what its content is

  • Boolean variables or functions/methods that return a Boolean should be written like predicates (is_empty, is_pending, is_active, etc)

Try to do

elapsedTimeInDays = 100
window_size_in_months = 7
number_of_pass_scenarios = 0
is_terminated = True
name = "John"

Avoid

d = 100 # Elapsed time in days
wsim = 7
qty_pass_s = 0
terminate = True
s_name = "John"
  • Function names should be verbs because they do things, while Classes, variables or attributes should be nouns/noun-phrases and Enums usually are adjectives because they tend to be names or object descriptors
  • Name of a variable should be proportional to the size of the scope that contains it, long scopes need long names. This is the reason i, j or d are good a variable names if for example they are used on a short loop. Variable names should be short if they are in small scope and long if they are in big scope
    • Local variables will have short names and global variables will have long names
  • Name of functions and classes should be inversely proportional to the size of the scope that contains it:
    • Functions/Classes names should be short if they have a big scope while long and descriptive if they have short scope
    • The more specific class/function the longer the name, while the more general class/function the shorter name
    • Classes and functions that are public and called from many places (this mean they have long scopes) have nice short convenient names (open(), close(), serve(), etc.)

Try to do

from enum import Enum
class TransactionStatus(Enum):
    PENDING = auto()
    CLOSED = auto()
    CANCELLED = auto()

if transaction.is_cancelled():
    cancel_reason = transaction.get_cancellation_reason()
    ...

Avoid

# TODO

Functions & Functions implementation

  • A function should be small, do one thing, do it well and do it only.
    • Function should be small (4-6 lines or around 10 lines maximum) breaking down your thoughts and making things small takes time but not doing it will cost other people or your future self more time
    • Functions should do one thing, one thing means you cannot meaningfully extract another function from it.
      • If a function contains code and you can extract that code to create another function the original function is doing more than one thing
      • if a function does only one thing it means every line of the function is at the same level of abstraction, we don't want a function that crosses levels of abstraction
      • Code that goes up and down the abstraction levels is not clean code
      • We want to separate the levels of abstraction into different functions and/or classes
      • Create functions that do one thing only, try to Extract Until You Drop
  • Function-call overhead is in most scenarios a thing of the past we should take advantage of the power of our current hardware to partition our software for readability first
  • A set of functions that manipulate a common set of variables is a class
    • A large function is usually just a class with a lot of smaller functions inside

Structure and Functions Structure

  • Two or three arguments per function, if you already have 5-6 things to pass into a function, if they are so cohesive (related to each other) to be passed into a function together maybe they should be a class

  • Try to avoid passing Booleans to functions create separate functions, when you pass a Boolean you are saying that your function has two paths or does two things, one for the True case and another for the False case

  • No output arguments if possible in functions use ONLY the return value to pass data out of a function

  • Avoid functions that can take None or null or expects one of these invalid values because we are saying the function handles two possible states, one for the null case and one for the non-null case, write to functions one that takes no arguments (for null case) and one that takes one single argument (for non-null case)
  • Try to avoid defensing programming, checking for null/None checks and error code checks, try to write code that prevents you from passing null/None values
    • Try to limit defensive programming to public APIs, here it is needed because we don't know what people are going to pass into my functions
  • The Stepdown Rule: Important stuff at the top and the more specific or detail oriented stuff at the bottom, avoiding backward references. Every functions calls children function which then call their own children's function, we order functions in the order they are called and in the order of their hierarchy

The Stepdown Rule

  • Use exceptions over return codes, keep the code inside a try block small or better yet make it just be a function call
    • If a function can throw an exception just put the call to that function inside a try-catch, don’t add more code before or in the try-catch

Dependency Structures

  • If a module A calls a function from module B directly we have two types of dependencies:
    • Runtime/Flow of control dependency: Module must be B loaded and ready in memory when we execute module A
    • Source code Dependency: module A has some type of include/import of module B
  • If we use an Object Oriented (OO) paradigm we can manage the source code dependency by making A dependent of an interface and module B derives from the interface, this allows modules to be deployed separately. Module B is a plugin to A
    • Dynamic linking objects exists so we can independently deploy aesthetics modules (like the GUI) from core functionality and data managers (code that interacts with a database for example)
    • Things like .jar files and DLLs are dynamic linking objects which means everything's is loaded and linked at runtime, they exists so you can independently deploy chunks/modules of your system

Invert source code dependency

  • Switch statements creates an dependency magnet (Fan-out problem), they are the opposite of independent deployability because usually each case in the switch statement is likely to have a dependency of an external module, so the source code dependencies point in the same direction as the flow control. In summary the switch statement depends on all the downstream modules

Switch statement dependency

  • To avoid switch/if-elif-else dependencies problems we can use Polymorphism or move the statement to a sections of the code that we know won't grow
  • Dependency Inversion: We should be able to draw a line that separates the core app functionality (App partition) from the low-level details(Main partition)
    • App partition a bunch of different modules, this should not have dependencies from the Main partition
    • Main partition keep it small and with limited subdivision, this part depends on the application. This section is a plugin to the application
  • Runtime and source code dependencies should point towards the direction of the application

  • There should be only a few entry points from main into the application, let main do rest of the work with factories implementations and strategies patterns (abstract interfaces or skeletons will be in the app side) and nothing that happens in the application should affect the main partition

App & Main partitions

  • switch statements that live in the main partition usually are OK as long as all the cases stay in the main partition as well

  • The system should be composed of independently deployable plugins (main partition), all of those plugins should know about the central core of the application partition but the application partition shouldn't know anything about the plugins at all

  • Sequence, selection (branching) and iteration are the invariants (stuff that never change) in software, they are the fundamental blocks of software

  • The main SW development paradigms are 3: Structured, Functional and Object-oriented

Separation Structures

  • Functional programming is the paradigm where we cannot change the state of variables, this means no assignment statements so therefore no side effects
    • Instead of setting a bunch of values into variables, you pass those values as arguments into functions. Instead of looping over a set of variables, you recurse through a set of function arguments.
    • Functions always return the same value back, with you giving the same input. - because the value of a function depends only on its input arguments, and not on any other state in the system
  • Side effect refers to code (or function) that changes the state of the system:
    • Functions like open or new have side effects because they leave a file opened or leave a block of memory allocated, before calling the function the system had a state that was using certain amount of memory after calling the function it has a state in which it uses more memory
    • Side effect functions come in pairs (open/close, new/free, lock/unlock, set/get, etc.)
    • Contain side effect code so you don’t forget to close, free, unlock, etc.
  • When functions must be called in a order we have temporal coupling or when we depend on one thing happening before or after another E.x. when you use a database you are coupled to the fact that you have opened a DB connection then do some stuff and finally close the connection
class FileCommand():
    def process(to_process_file):
        print("Processing of {} happens here".format(to_process_file))

def open_file(my_file, my_file_command):
    my_file.open()
    my_file_command.process(my_file)
    my_file.close()


open_file(open("my_file.txt", "w"), FileCommand())
  • Command and query separation: Functions that change state should return nothing while functions that return values should not change state
    • Commands change state of the system, and returns nothing, these can throw exceptions. E.x. setters are commands
    • A Query returns value of a computation or the current state of system but it won't change the state. E.x getters are queries
  • Avoid functions that change state (Commands) AND return values, this violates command query separation

Try to do

class Auth():
    def login(username, pass):
        # Call code that returns if login was possible here
        login_successful = True
        if not login_successful:
            raise Exception

authorize = Auth()
authorize.login(username, password)
user = authorize.get_user(username)

Avoid

class Auth():
    def login(username, pass):
        # Call code that returns if login was possible here
        login_successful = True
        if login_successful:
            return User(username)
        else:
            return None

authorize = Auth()
user = authorize.login(username, password)
  • Tell Don't Ask: You should tell objects to do the work not asking what their state is or whether they could perform the work
    • Avoid Asking for the state of an object then make decision on behalf of that object
    • Objects should know their own states and make their own decisions
    • Minimize query functions
  • Law of Demeter:
    • It's dangerous that a method has too much knowledge of the system or the entire navigation structure of the system (we don't want a single method to be coupled to a lot of parts of the system)
    • Want we want are methods that tell neighbor objects what we need to have done and depend on them to propagate that message outwards to the appropriate destinations
    • You may call methods of objects that are:
      1. Passed as arguments to your method
      2. Created locally in your method
      3. Instance variables of the class of your method
      4. Global variables
    • You may NOT call methods of an object if that objects was returned by a previous method call

Try to do

my_object.doSomething()

Avoid

my_object.getX().getY().getZ().doSomething()
  • Structured programming: every algorithm can be written out of 3 structures (sequence, selection if-else, iteration loops)
    • The basic structures used as building block of modules and systems have a single entrance at the top and an exit at the bottom

Error Handling

  • Error handling is important but If it obscures logic it is wrong
  • Write error handling first before writing the rest of the code
  • Prefer exceptions over error codes
  • Exceptions should be part of the scope of the class that throws it and be named with as much specific information of the class error
  • Exceptions should derive from a base exceptions like Runtime
  • Exceptions names should be so precise they don't need message
  • null/None is not an error is a value when tempted to return those check if using exceptions is better
  • We use exceptions for things that aren't expected
  • if try blocks are in function it should be the first thing before any declarations and should have very few lines or better yet a single function call

Py-Stack Example

# python -m unittest py_stack.py

from unittest import TestCase
from abc import ABC

class Stack(ABC):
    def get_size(self):
        pass
    def is_empty(self):
        pass
    def push(self, element):
        pass
    def pop(self):
        pass
    def top(self):
        pass
    def find(self, to_search_element):
        pass

    class Overflow(RuntimeError):
        pass
    class Underflow(RuntimeError):
        pass
    class IllegalCapacity(RuntimeError):
        pass
    class Empty(RuntimeError):
        pass

class BoundedStack(Stack):
    class ZeroCapacityStack(Stack):
        def get_size(self):
            return 0
        def is_empty(self):
            return True
        def push(self, element):
            raise Stack.Overflow()
        def pop(self):
            raise Stack.Underflow()
        def top(self):
            raise Stack.Empty()
        def find(self, to_search_element):
            return None

    def __init__(self, capacity):
        self.capacity = capacity
        self.size = 0
        self.elements = []

    @staticmethod
    def make(capacity):
        if capacity < 0:
            raise BoundedStack.IllegalCapacity()
        elif capacity == 0:
            return BoundedStack.ZeroCapacityStack()

        return BoundedStack(capacity=capacity)

    def get_size(self):
        return self.size

    def is_empty(self):
        return (self.size == 0)

    def push(self, element):
        if self.size >= self.capacity:
            raise BoundedStack.Overflow()

        self.elements.append(element)
        self.size += 1

    def pop(self):
        if self.is_empty():
            raise BoundedStack.Underflow()

        element = self.elements.pop(self.size-1)
        self.size -= 1
        return element

    def top(self):
        if self.is_empty():
            raise BoundedStack.Empty()
        return self.elements[self.size-1]

    def find(self, to_search_element):
        for i, e in enumerate(self.elements):
            if to_search_element == e:
                return i
        return None

class StackTest(TestCase):
    def setUp(self):
        self.my_stack = BoundedStack.make(2)

    def test_just_create_stack_is_empty(self):
        self.assertIsNotNone(self.my_stack)
        self.assertTrue(self.my_stack.is_empty())
        self.assertEqual(0, self.my_stack.get_size())

    def test_after_one_push_stack_size_must_be_one(self):
        self.my_stack.push(1)
        self.assertEqual(1, self.my_stack.get_size())
        self.assertFalse(self.my_stack.is_empty())

    def test_after_one_push_and_one_pop_stack_size_must_be_empty(self):
        self.my_stack.push(1)
        self.my_stack.pop()
        self.assertTrue(self.my_stack.is_empty())

    def test_when_pushed_past_limit_stack_overflows(self):
        self.my_stack.push(1)
        self.my_stack.push(1)
        # self.assertRaises(BoundedStack.Overflow, self.my_stack.push, 1) # Also valid
        with self.assertRaises(BoundedStack.Overflow):
            self.my_stack.push(1)

    def test_when_pop_past_zero_stack_underflows(self):
        # self.assertRaises(BoundedStack.Underflow, self.my_stack.pop) # Also valid
        with self.assertRaises(BoundedStack.Underflow):
            self.my_stack.pop()

    def test_when_one_is_pushed_one_is_popped(self):
        self.my_stack.push(1)
        self.assertEqual(1, self.my_stack.pop())

    def test_when_one_and_two_are_pushed_two_and_one_are_popped(self):
        self.my_stack.push(1)
        self.my_stack.push(2)
        self.assertEqual(2, self.my_stack.pop())
        self.assertEqual(1, self.my_stack.pop())

    def test_when_creating_stack_with_neg_size_must_raise_IllegalCapacity(self):
        with self.assertRaises(BoundedStack.IllegalCapacity):
            BoundedStack.make(-1)

    def test_when_creating_stack_with_zero_size_any_push_should_overflow(self):
        with self.assertRaises(BoundedStack.Overflow):
            local_stack = BoundedStack.make(0)
            local_stack.push(1)

    def test_when_one_is_pushed_one_is_on_top(self):
        self.my_stack.push(1)
        self.assertEqual(1, self.my_stack.top())

    def test_stack_is_empty_top_raises_empty(self):
        with self.assertRaises(BoundedStack.Empty):
            self.my_stack.top()

    def test_with_zero_capacity_stack_top_raises_empty(self):
        with self.assertRaises(BoundedStack.Empty):
            local_stack = BoundedStack.make(0)
            local_stack.top()

    def test_given_stack_with_one_two_pushed_find_one(self):
        self.my_stack.push(1)
        self.my_stack.push(2)
        self.assertEqual(0, self.my_stack.find(1))
        self.assertEqual(1, self.my_stack.find(2))

    def test_given_stack_with_no_two_find_two_must_return_None(self):
        self.my_stack.push(1)
        self.my_stack.push(3)
        self.assertIsNone(self.my_stack.find(2))
  • Open close principle: a module should be open for extension but closed for modification, you should be able to extend the behavior of a module without modifying that module, this is usually implemented with base classes and derivatives

  • Science is a set of theories, conjectures, hypothesis that cannot be proven correct, experiments can prove them false but never true, we have not proven that airplanes can fly, we have just surround with so many tests, it doesn't matter we don’t prove it correct, we can do the same with software by surrounding with tests, we demonstrate our software is not incorrect

  • Why would we write code but don't write tests for every line of code? why don't we test the individual lines we write?

Any fool can write code that a computer can understand but it takes a good programmer to write code a human can understand :: Martin Fowler

Form, Explicit Code & Comments

  • Purpose of a comment is to explain the code that cannot be explained by the code itself

  • Comments are to compensate for our failure to express ourselves in code, so every comment represents a failure and you will certainly fail, that's OK but consider the comment as a unfortunate necessity

  • Don't comment first, try everything else first, use a comment as last resort

  • Comments are a lot of times bad because we don't maintain them

  • Comments don't make up for bad code, try to clean the code if you fail, that's fine write a comment and move on but REALLY try to clean the code first

  • Explain yourself with code not with comments, make the code express itself

  • Never comment about code that is somewhere else

  • Reserve comments for local stuff, if we write comments that are not local when some other part of the code is changed it may invalidate our comment

  • Use comments to document public APIs and generic functions/methods that other programmers will use

  • Try to have if statements and while loops of only one line

  • Big projects do NOT imply big files (Smaller is better)
    • Most files should be 200 lines or less with a max file size of 500 lines
  • Use blank lines to separate topics or processes in code. Things that are related to each other should be closed to each other
    • Separate variable declaration from functional code in a method
    • Separate with blank lines parts of processes in a function
    • If a group of variables are related to same topic group them together but separate them from others that are used for a separate topic or process
  • Length of lines of code should be of any length as long as you don't have to scroll right to see (max 100 or 120 characters)

Classes & Data Structures

  • Classes are containers of methods/functions that hide the implementation of the data they manipulate

  • Minimize getters and setters in class and try to maximize cohesion

  • Don't just expose variables through simple getters, abstract information you want to expose

  • Methods operate on data but tell you nothing about how that data is implemented, when they expose data they exposed in the most abstract way

Try to do

class Car():
    def __init__(self):
        self.fuel = 0
    def get_fuel_percent():
        pass

class ElectricCar(Car):
    def calc_fuel_electric_car(self):
        # You would implement the particularity of
        # getting fuel percent for electric car here
        return self.fuel*2 + 1
    def get_fuel_percent():
        return self.calc_fuel_electric_car()

class NuclearCar(Car):
    def calc_fuel_nuclear_car(self):
        # You would implement the particularity
        # of getting fuel percent for Nuclear car here
        return self.fuel*3.1416 + 1000
    def get_fuel_percent():
        return self.calc_fuel_electric_car()

Avoid

class Car():
    def __init__(self):
        self.gallons_of_gas = 0
    def get_gallons_of_gas():
        pass
"""
You need to forget get_gallons_of_gas because
it doesn't make sense for other types of cars, so
you end up with a useless class hierarchy
"""
class ElectricCar(Car):
    def __init__(self):
        self.battery_level = 0
    def get_battery_percent():
        return self.battery_level

class NuclearCar(Car):
    def __init__(self):
        self.energy_level = 0
    def get_nuclear_energy_level():
        return self.energy_level
  • Data Structures are containers of data with no cohesive methods

  • A data structure is very different from a class because it has a bunch of public variables and virtually no methods while classes should have a bunch of public methods and try to have no public variables

  • Data structures can have methods but they should only manipulate individual variables they don't manipulate cohesive groups of variables like methods of a class do

  • Methods of a data structure expose implementation they don't hide it nor abstract it like classes

  • You can't tell a data structure to do anything you can only ask it questions

We use classes and objects when it's types that are more likely to be added we use switch statements and data structures when it's methods that are more likely to be added

Data Structure and Classes

Boundaries

  • Boundaries separate things that are concrete from things that are abstract in our system:
    • Boundary that separates Main from the core App
    • Boundary that separates View from Models
    • Boundary that separates the Database from Domain objects
  • For each boundary one side is concrete implementation the other side is abstract definitions. E.x. Main, DB & View are concrete while App, Models, Domain are abstract,

  • All the source code dependencies point from concrete to abstract (imports are in concrete side). E.x. Main, DB & View depend on the App, Models & Domain

Concrete & Abstract

  • Impedance Mismatch: When we define classes in the application side with methods these usually don't look like database tables or database schema this is the
    • Rows in a database is not an object is a data structure
    • Databa se do not contain objects they contain data structures
    • You cannot for a data structure to be an object
    • Object Relational Mapper (ORM) aren't actual Object Relational Mapper they just fills data structures, they take a database row or other source of data as inputs and fills attributes or data structures
  • It is the responsibility of the interface between the DB and the APP to convert the data structures that live in the database to business objects that app uses and the layer needs to do this without letting the app know it exists
    • On the app side we have interfaces that define data access methods, the app objects uses those interfaces
    • On the other side of the boundary (DB, View or Main) we want classes that derive from the interfaces and implement the data access method by interrogating data structures that are fetched from database
  • Views should know about the app, the app should know nothing about the views

Boundaries

Often the more concrete side will be composed of data structures that use switch statements

Test Driven Development (TDD)

  • Test Driven Development is a discipline that has the following rules which are followed in a endless loop:
    1. You are not allowed to create any production code until you have at least one test that fails because the production code doesn't exists
    2. You are not allowed to continue writing a test that you already know will fail and not compiling or running is a failure. In other words as soon as you realize you have written a test that fails you must stop written the test
    3. You are not allowed to write more production code than the one needed to make the previous test pass
    4. Every change you add to the test make it more specific or test more specific things while every change you add to production code makes it more general

Disciplines are arbitrary or have arbitrary components and behaviors that are driven by a substantial motive. Arbitrary means in simple words it is just there because of a guess that has proven to work based on experience and it doesn't cause problems

  • We can't clean code until we eliminate the fear caused by changing and breaking the code

  • Tests eliminate fear so they let you clean code

  • If you want a flexible system, get a suite of tests that you trust

  • A test suite that doesn't allow you to make a decision or give you enough confidence to know your code works is useless

  • Giving priority to tests, makes a system testable (and also decoupled)

  • Inheritance is the tightest coupling we have so we want to have as little inheritance as possible or inherit as few implemented things as possible

  • The tests must be like code examples of how to use your system/app, they are short snippets of the code that explain how one part of the system works. E.x. if you want to create an object then there should be a test that does that and tells you what you have to do, so tests if implemented correctly will give you all the low-level detail of how the system works

  • Test Driven Development Examples:
    • General: You need to code a feature or create some code that does something, so you start writing a test, once you realize you can run it and it fails you stop and write code that makes that test pass, once the test passes you continue writing other test for the thing you are creating until you are able to finish a test suite that proves that what you are creating works or until you realize the test will fail, in that case you go back to writing production code to try to make the test pass pass and repeat that pattern
    • Specific: I want a function to add two numbers so I write a line for a test, oh it doesn't compile because I don't have the function declared, I'll write some code to make that compile, now it compiles, now let me continue with the test to prove my code really adds two numbers, now it fails because I don't have the implementation of the function so I go back to implement that...
  • Mutation testing: is an computationally expensive technique of testing in which a test suite that pass is run and the coverage is recorded then a mutation tool changes/mutates something in the code (change a == for a != for example`), then runs the test suite again, measures the coverage expecting one or more test failures, if a test pass it means a mutation is alive, these mutation are counted and reported since they point we have either a bad test, a bad code or bad coverage

  • Every creative effort on the planet is done iteratively, we don't write perfect code the first time. It requires refinement and rework

  • If a single change in production code breaks the tests there is something wrong with the tests, they are too coupled to production code

Test Driven Development General Process

  1. You have a design session, in this we define the classes, some general methods and how the objects will interact with each other, also which objects contain other objects, etc
  2. Start development with a test, we kind of ignore the design we just use it as a guideline but we don't actually follow it. The process may lead us to the conclusion our design was not needed or it is not the right design
  3. We follow the Fail-Pass-Refactor Cycle
    1. Write a test that Fails
    2. Write production code that makes the test Pass
    3. Check for duplicate code or whether there is something we need to Refactor, if that's the case, refactor it otherwise go back to first step

Test and Test code are just as important as production code

  • Try to get 100% coverage even knowing it is not possible
  • GUIs are separated in two layers, presenter and view
    • The presenter gets information from app objects and constructs data structures, the presented knows the state of the system by querying all that information, understand how to fetch data and format dates, currencies, strings and other items to put correct values in data structures
    • The view renders the data structures created by the presenter

Expectations in a SW project

The way we imprint professionalism into the SW development profession is by setting high expectations like:

Release good quality code

  • This means release code you know it works
  • Release code that has been tested at all levels (unit-integration-system)
  • Release code that is clean and understandable
  • Release code that is well organized

Always be ready to deploy

  • This doesn't mean it will have enough features but should be ready to be released and work (testing is done, documentation and all the work needed to make current features work is done)

Stable productivity

  • We should not get slower as time passes
  • We should be able to produce features at the same rate at the beginning middle or end of project
  • The reason we slow is because we make a mess

Inexpensive Adaptability

  • It must be cheap and easy to make changes to the system, because the main purpose of software is that must be changeable
  • in more realistic words: the cost of the change is proportional to the scope of the change
  • If doing a minimal change like changing a label on the system costs me a lot we are not fulfilling this expectation

Continuous improvement of the code

  • Code should get better with time, the design and the architecture should get better with time
  • Boy's scout rule: Check in the code a little better that you checked it out

Fearless Competence

  • This means don't be afraid of the code, don't be afraid to clean it, to improve it and restructure it
  • Avoid the "if ain't broke don't fix" mentality
  • If we choose change the code in a way to minimize the risk of breaking it instead of changing it to improve the general structure then the code will just rot and get worse with time, the productivity will slow to a point it is not possible to make a change without breaking something
  • To be fearless you need confidence to have confidence you need feedback mechanisms that run fast, are easy to run and tell you everything is OK, in other words tests
  • Fragile test problem: When test are so coupled to the system that if you change something in the system the whole test suite break
  • To avoid the fragile test problem we need to consider tests are part of the system so need to be designed as part of it with the same techniques to have a good design. They need to be decoupled from the system

  • The definition of bad design is a design that when you make a change in one part that change breaks a lot of things in another part

Extreme Quality / QA must find nothing

  • QA is not the tool to find bugs, they should find nothing
  • QA belongs at the beginning of the process
  • QA is the one that tells you how the system should work

We cover to each other

  • We cover for each other when a player goes down
  • To make sure someone can cover for you is your responsibility and to do this you need to walk someone through what you do
  • Pair enough (1 hour a day or a couple times a week) to be able to cover for each other

Honest estimates / Say "NO"

  • The most honest estimate: I don't know is the right but less useful answer
  • Define the shape of what we don't know
  • PERT (Project Evaluation In Real Time): You give 3 numbers best-case, nominal-case/usual-case and worst-case
  • A coder is hired because he knows how to code and works very hard to say "YES" but most importantly a coder knows when the answers is "NO" or "I don't know"

Architecture

What is Architecture

  • Architecture includes the grand shape of the system and the teeniest low level design, includes the most abstract of the module interfaces and the most concrete of the method implementations

  • The Architecture is the shape the system takes in order to meet its use case

  • Architecture is not about the tools we use like (IDE, GUIs, language, dependencies, frameworks, databases, etc.) it is about usage and communicates intent, architecture tells you how a system it is being used, it's about the use cases

  • A good architecture is a SW structure that allows you to defer critical decisions about tools and frameworks (GUIs, webservers, DBs etc) for as long as possible, defer them to the point where you've got enough information to actually make them or not make them at all, allows you to defer majors decisions until the last responsible moment. A good architecture decouples you from tools, frameworks & databases

  • Architecture needs to be focused on use cases not on software environment

  • Make the GUI and database a plugin into the business rules

  • Using a framework comes with a great commitment to the framework

  • Architecture must be flexible because as we implement it, the code and modules starts imposing pressure on the architecture which changes the shape of the architecture and this change continues every time a new feature is added

  • The architecture is a living thing, it changes with the system on a day to day basis

  • Good architecture results in good structured SW in which the effort to build it, to maintain it and the struggle to continue to add new stuff to it is minimized while the functionality and flexibility is maximized

  • The goal of SW architecture is to minimize the effort and resources required to build and maintain a software system

  • Quality of design can be measured by measuring the amount of effort it takes to meet the needs of a customer, if effort increases design quality decreases (it is bad) while if effort remains constant of shrink with time then design is good

  • We want to go fast and become overconfident we will be able to deal with the mess later, so the mess builds slowly but inevitably and eventually it is so big it dominates everything about the project

  • We need to build in a steady, consistent, clean way so that so that we can continue to go as fast as possible, which may not be very fast but at least it is not going to slow down

  • The key to go fast is to not slow down

  • A discipline may feel slow even if it is actually speeding you up. A good discipline may feel slow but will speed you up

  • SW has two values the value of what it does and the value of its structure but we tend to focus on what the SW has to do (satisfy the requirements) and if we do that we are maximizing the fact that it works today at the expense of making it flexible and being able to change so when the requirements change (and they will, they always change) the cost to change it will be enormous and if the cost is to big that will just kill the project

  • Programmers know we need to build a structures that supports the requirements, while stakeholders only know/see the requirements, programmers need to communicate the importance of structure

  • We (Programmers) tend to be bad at emotional conformation and relating emotionally to people

  • Behavior is urgent structure/architecture is important, because it has an extreme long term value while the requirements are urgent because they have a short term value

  • Focus on important and urgent, also on the important but not urgent

  • Architecture should express the intent of the application

  • The web, databases, frameworks and other system components are just I/O devices from the SW developers point of view

Architecture

Use Cases

  • Description of how the user interacts with the system in order to achieve a specific goal
  • We need to abstract the uses cases from UI or web based interactions like (buttons, screens, pop-ups, droboxes, checkboxes, input fields, etc)
  • Use cases need to be independent of the delivery mechanisms (the web, a console, a GUI, etc)
  • Use case must be an algorithm for the interpretation of input data and generation of output data
  • We need to make the uses cases as agnostic as possible about delivery mechanisms, frameworks, databases, corporate standards, etc
  • A use case talks about the data and the commands that go into the system and the way the system response
    • Data a list of descriptions of the data
    • Primary course: refers to the happy path, it shows you what happens if nothing goes wrong
    • Exception course/Extension course: refers to are the possible exceptions that can occur and shows what happens on those cases

Use Cases

Input-Process-Output Models

  • Use cases talk about system objects which usually fall in the following categories:
    • Business objects (entities):
      • Objects that contain Application Independent Business Rules
      • Objects have methods that perform actions that are valid in any applications the object can be used in. E.x. a product object, which can be used on an order system, inventory management system or online catalog app
      • Perform actions you will do even if there were no computer, the manual steps. E.x. in a interest calculator app do the math operations to calculate the interest is a business rule you would do it by hand or with traditional calculator even if you don't have a computer)
      • These objects capture the agnostic behaviors of the uses cases
    • User Interface objects (boundaries):
      • Objects that accept input data and deliver output data to the user
      • Isolate the uses cases from the delivery mechanism
      • Boundary objects communicate the delivery mechanisms with the interactors
    • Use Case objects (controls/interactors/transactors/transaction)
      • Objects that contain Application Specific Business Rules they are all about automation, these are the thing you need to do because there is a computer like the coordinating the calls of a sequence of functions in a specific order
      • These objects capture the application specific behaviors of the uses cases
      • These are the objects that have application specific logic that call application agnostic logic that lives in the entities

entities-boundaries-interactors

  • Object Oriented Use case driven: Interactor/Control Object (object that represents a use case, which is a application specific business rule) guides/coordinates the execution of the entities, which represent the actual business rules controlled by the interactor, then there is data in and out of the system, the data movement of data is done through interfaces or boundary objects, interactor uses the input boundary objects to push data across the entities and the methods on the boundaries allow data to go in and out of the system
    • Entities execute the real business rules the uses cases or interactors coordinate the business rules
    • The interactor uses the output boundary to send data out while it implements the input boundary interface
  • Object Oriented Use case general process:
    1. The delivery mechanism is just a user interface like forms on web page which are submitted to a web server and the web server is the one that creates a request model which is just a raw data structure that holds the input data, doesn't know where it comes from (for example the web),
    2. Request model it gets passed through the input boundary to the interactor
    3. The interactor interprets the data, coordinates and uses the entities to perform operations or transform the data
    4. Then the interactor gathers the result data from the entities and builds a result model which is just raw output data that gets passes out through the output boundary

Request and result models are just plain data structures or objects

Object interactor-entities-boundary example

  • The presenter jobs is to reorganize the response model (raw data) to be ready to be displayed, response model holds objects (e.x. has a currency object, date object, etc) the presenter grabs those objects and transform them to be displayable (e.x. to a string)
  • View model is a data structure that contains information about the data and data itself ready to be displayed (for or example strings properly formatted), if a button must be grayed the presenter will set a Boolean in the view model that indicates this, the view just takes the data out of the view model and puts it into the display

Presenter example

  • Interactors and entities should know nothing about the delivery mechanisms, they should be plain old objects operating on data
  • Interactors take in request models, produce response models, pass the data through the boundaries, controllers are the ones that pass data into the interactors while the presenters are the ones that take data and present it to the view
  • Boundaries are a set of interfaces:
    • Input data interface First set is uses by the controllers but implemented by the interactors, accepting request model data structures
    • Output data interface Second set is uses by the interactor but implemented by the presenter accepting response model data structures

Another Presenter example

  • We don't want the things that perform business rules/operations coupled to database language like SQL, we need and entity gateway that has an interface (not implemented methods) for every query we want to perform, then we have the entity gateway implementation that has the knowledge on how a specific database works, if we have an SQL database all the SQL will be in the implementation

  • Interactor objects uses the database abstraction (entity gateway interface methods) on the application side, the implementation of those abstractions

DB and entity gateway example

  • An object is something that contains behavior, not necessarily data, if it contains data most of the times user should not know about this

Nice Architecture example

MVC

  • Model View Controller (MVC): input-process-output model intended to be uses in small scale, you have an MVC for a button or a textbox, not a whole webpage
    • Model: know how to perform business rules operations, does data transformations
    • Controller: knows everything about inputs and restructures input into commands for the model
    • View: Knows how to display model data, present it
      • View registers a callback with the model, when model changes it tells the view 'hey redisplay me because I changed'

MVC example

Manage a SW project

  • The reason we manage is because mismanagement cost a lot (meaning, produce the wrong product, create a product of inferior quality, work 80 hours a week, etc)
  • Good (quality), Fast (Time to Market), Cheap (Cost effectiveness), Done... Pick only 3 the fourth you will not get it.
    • A good manager knows how good, how fast how cheap, what will be done to deliver the project to the best possible outcome (good enough) to do this we need data
  • Agile is a way to produce data to destroy hope, to know how messed up things really are

  • Dates are fixed because the obey business, for example it has to be done by November because we ran out of money in December

  • Requirements change all the time because they users/customers don't know what they want, the more we talk about what they want they kind of get a better idea of what they want but they really don't know until they see something execute that is not what they want so then they say "oh yeah that is not what I wanted"

  • We need to get things in front of customers very early so they can tells us "oh that's not what I wanted, I know that is what I said but now that I see it I realize it is not what I want at all", we need to start this feedback cycle fast

  • Waterfall model in which we analyze, then develop then implement doesn't work because once you start implementing you realize requirements were not clear and design won't really help solve the problems we want so the only solution is to do everything at the same time
    • We never know when design and analysis is done because we don't have a clear definition of done for these stages
  • On Agile we do exploration phase that involves doing analysis,design and implementation the first 3-4 sprints to be able to estimate how much we can get done in average in a sprint and extrapolate when we will actually be done, we will realize we wont be able to do everything by the end of date so we make adjustments

  • Doubling staff does not double throughput because old people slow down due to they need to help ramp up new people and expecting new people to get faster by an arbitrary date (in a month for example) is risky because you really don't know that, you don't have data to know that and even if you have data it is not reliable because varies from person to person

  • We must focus on never ever deliver something you stakeholder don't need

  • The problem with estimates is that we need to tell a moron (computer) how to do something

  • Agile is not to go fast is to get the bad news as early as possible

  • Scrum is a business facing process, then there are the engineering practices and the team behaviors

  • The purpose of an iteration is to produce data and measure how much can get done, an iteration doesn't fail because even if no points are done that is actually a result with data

  • Write acceptance test that defines done for every story so a programmer knows he is done when an acceptance test for that story pass, so acceptance test must be done by midpoint of the sprint

  • Seniors developers should spend most of their time pairing with other people, mid level people should spend some time pairing with junior people

  • If we have story that we don't know how long it should take, we can define a spike, which is a story to try to figure out how long it should take you to finish the first story

  • You need to try as best and as hard as you can to follow clean code practices and rules but sometimes the rules cannot hold all the time so we violate the rules

References

Want to show support?

If you find the information in this page useful and want to show your support, you can make a donation

Use PayPal

This will help me to create more stuff and fix the existent content... or probably your money will be used to buy beer