The Programmer's Brain
by Felienne Hermans
Last updated
by Felienne Hermans
Last updated
Your brain responds in a predictable way when it encounters new or difficult tasks. This unique book teaches you concrete techniques rooted in cognitive science that will improve the way you learn and think about code. In The Programmer’s Brain: What every programmer needs to know about cognition you will learn:
Fast and effective ways to master new programming languages
Speed reading skills to quickly comprehend new code
Techniques to unravel the meaning of complex code
Ways to learn new syntax and keep it memorized
Writing code that is easy for others to read
Picking the right names for your variables
Making your codebase more understandable to newcomers
Onboarding new developers to your team
Learn how to optimize your brain’s natural cognitive processes to read code more easily, write code faster, and pick up new languages in much less time. This book will help you through the confusion you feel when faced with strange and complex code, and explain a codebase in ways that can make a new team member productive in days!
Confusion = part of programming.
When you learn a new programming language, concept, or framework
New ideas might scare you
When reading unfamiliar code or code that you wrote a long time ago, you might not understand what the code does or why it was written the way it is
Whenever you start to work in a new business domain, new terms and jargon can all bump into each other in your brain
3 types of confusion in code
Lack of knowledge
Lack of easy-to-access information
The information about exactly how toBinaryString() works is not readily available
Needs to be found somewhere else in the code
Lack of processing power in the brain
You cannot oversee all the small steps that are being executed
If you need to understand all the steps, you may likely use a memory aid like intermediate values of variables shown here
An overview of the three cognitive processes that this book covers: STM, LTM, and working memory.
Information coming into your brain.
Information that proceeds into your STM
Information traveling from the STM into the working memory, where it’s combined with information from the LTM (arrow 4).
Working memory is where the information is processed while you think about it.
Lack of knowledge -> Issue in long-term memory
Lack of information -> Issue in short-term memory
Lack of processing power -> Issue in working memory
Long-term memory
First cognitive process that is used while programming is long-term memory
This can store your memories for a very long time
Like the hard drive of your brain
Short-term memory
Short-term memory is used to briefly hold incoming information
The estimates differ, but most scientists agree that just a few items fit in short-term memory, and certainly not more than a dozen
Like the computer's RAM or a cache that can be used to temporarily store values
Working memory
The actual “thinking,” however, happens in working memory
Like the processor of the brain
Use long-term memory when remembering meanings of keywords for example
Use short-term memory to store some of the info you encounter
Use working memory : trying to mentally execute the code / understand what is happening
That process is called tracing — the mental compiling and executing of code.
When tracing very complex programs, you might feel the need to note down the values of variables, either in the code or in a separate table.
The fact that your brain feels the need to store information externally can be a sign that your working memory is too full to process more information.
Research indicates that almost 60% of programmers’ time is spent understanding code, rather than writing code.[1] Thus, improving how quickly you can read code, without losing accuracy, can help you improve your programming skills substantially.
"Programs must be written for people to read and only incidentally for machines to execute.” - Structure and Interpretation of Computer Programs by Harold Abelson, Gerald Jay Sussman, and Julie Sussman (MIT Press, 1996)
Reading code is done for a variety of reasons:
to add a feature
to find a bug
or to build an understanding of a larger system
Limited capacity of your short-term memory :
STM stores information that you read or hear for a brief period of time
Research indicates it cannot hold information for more than 30 seconds
Limitation of size : just a few slots available for information
7 + or - two
More research indicates that its capacity could be smaller : just 2 to 6 things
Experts’ memories differ from beginners’
De groot chess experiment
When comparing the performance of average players with chess master
He found that expert players were much better at recreating the chess setups than average players if not random
Experts grouped info in logical ways -> chunks
He considered “Sicilian opening,” for example, as one chunk,
Which can fit into one slot in short-term memory.
The theory of chunks also adequately explains why both types of players performed equally in experiment 2 (random setup)
The more information you have stored about a specific topic, the easier it is to effectively divide information into chunks.
In 1981 Katherine McKeithen, a researcher at Bell Labs, tried to repeat de Groot’s experiments on programmers :
Main takeaway : beginners will be able to process a lot less code than experts
Use Design Patterns
Having a design pattern present in the code was more helpful for performing maintenance tasks when the programmers knew that the pattern was present in the code
Gaining knowledge about design patterns
Going to improve your chunking ability
Helps you to process code faster
Write comments
When code contains comments
Programmers will take more time to read it
You might think that is a bad thing—that comments “slow you down”—
In fact, this is a sign that comments are being read when programmers read code
High-level comments like “this function prints a given binary tree in order”
can help programmers to chunk larger pieces of code
Low-level comments such as “increment i (by one)” after a line that reads i++;
can create a burden on the chunking process
Leave Beacons
Beacons are parts of a program that help a programmer to understand what the code does like a line of code even part of a line of code, that your eye falls on which "Now I see"
Provide an important signaling service for programmers during the comprehension process
They often act as a trigger for programmers to confirm or refute hypotheses about the source
2 types
Simple beacons :
Self-explaining syntactic code elements : meaningful variable names, operators such as +, >, and && and structural statements such as if, else, and so on can be considered simple beacons too
Compound beacons : larger code structures comprised of simple beacons. Compound beacons provide semantic meaning for
What you already know impacts to a large extent how efficiently you can read and understand code
The more concepts, data structures, and syntax you know
The more code you can easily chunk, and thus remember and process
4 Techniques that will help you memorize programming concepts better and more
Parnin determined that interruptions are, unsurprisingly, quite disruptive to productivity
His study showed that
It typically takes about a quarter of an hour to get back to editing code after an interruption
When interrupted during an edit of a method
Programmers were able to resume their work in less than a minute in only 10% of cases
A great way to learn anything, including syntax, fast is to use flashcards
Simply paper cards or Post-its of which you use both sides
One side has a prompt on it—the thing that you want to learn
The other side has the corresponding knowledge on it
When to use them ?
Use them often to practice
Use Apps like Brainscape, Anki, Quizlet, that allow you to create your own digital flashcards
They will remind you when to practice again
After a few weeks your syntactic vocabulary will have increased
Expanding the set of flashcards
2 good times to add flashcards to your set
When you’re learning a new programming language, framework, or library
You can add a card each time you encounter a new concept
When you’re about to Google a certain concept
That is a signal that you do not yet know that concept by heart
Thinning the set of Flashcards
To keep track of how well you know certain concepts
You can keep a little tally on each card of your right and wrong answers
Demonstrate what knowledge is already reliably stored in your long-term memory
The big problem with your LTM : you cannot remember things for a long time without extra practice
The decay for LTM is not seconds, like that of the STM
But it is still a lot shorter than you might think
After 2 days, just 25% of the knowledge remains in your LTM
Storage of memories in the brain is not based on zeros and ones
BUT it does share a name with how we store information to disk: encoding
Memories in the brain are organized in a network structure
Because facts are connected to large numbers of other facts
Memories are stored with associations, or relationships to each other
You remember the longest if you study over a longer period of time
In contrast with formal education, where we try to cram all the knowledge into one semester
Revisiting your set of flashcards once a month
Will be enough to help your memory in the long run, and it’s also relatively doable!
The biggest takeaway from this section is that the best way that science knows to prevent forgetting is to practice regularly. Each repetition strengthens your memory.
After several repetitions spaced out over a long period, the knowledge should remain in your long-term memory forever.
Don’t try to memorize all of your flashcards in a day
But spread your study out over a longer period of time
Research has shown that actively trying to remember makes memories stronger
2 techniques to strengthen memories :
Retrieval practice : actively trying to remember something
Retrieval is a more formal word for remembering
Elaboration : actively connecting new knowledge to existing memories
How to measure memories strength ?
Storage strength : indicates how well something is stored in long-term memory
The more you study something
The stronger the memory of it is
Until it becomes virtually impossible to ever forget it
Retrieval strength : indicates how easy it is to remember something
Have you already have the feeling where you’re sure you know something
a name, a song, a phone number, the syntax of the filter() function in JavaScript
But you can’t quite recall it; the answer is on the tip of your tongue
Information for which the storage strength is high— and retrieval strength is low
By actively thinking about and reflecting on the information
The process of thinking about information that you have just learned is called elaboration.
Schemata
The ways in which thoughts and the relationships between them are organized in our minds are called schemas, or schemata
When we save memories
the memories can even be changed to adapt themselves to existing schemata
Technique : Elaboration
One thing you can do to strengthen the initial encoding of memories is called elaboration
Elaboration means thinking about what you want to remember
Relating it to existing memories
Making the new memories fit into schemata already stored in your long-term memory
Using elaboration to learn new programming concepts
For example, you might try thinking of related concepts in other programming languages
Thinking of alternative concepts in Python or other programming languages
Or thinking of how this concept relates to other paradigms
What’s the difference between working memory and short-term memory?
Working memory = the short-term memory applied to a problem
Example :
Remember a phone number -> use your STM
Add integers -> use your working memory
What happens in the working memory when you read code?
Like the short-term memory, the working memory is only capable of processing between 2 and 6 things at a time
In the context of working memory, this capacity is known as the cognitive load.
Different types of cognitive load :
Load Type
Description
Intrinsic load
How complex the problem is in itself
Extraneous load
What outside distractions add to the problem
Germane load
Cognitive load created by having to store your thought to LTM
Intrinsic cognitive load is cognitive load caused by features of a problem that the problem contains by nature
Example : A geometry problem in which the lengths of two sides of a triangle are given and the third one needs to be calculated. This can be a problem that is hard to solve, depending on your prior knowledge. However, the problem itself cannot be made any simpler without changing it.
we can think of this type of extraneous load as similar to accidental complexity :
Aspects of a program that make a problem harder than it needs to be
What creates extraneous load is not the same for every programmer
The more experience you have using a certain concept, the less cognitive load it creates for you.
In most cases, refactoring is done to make it easier to maintain the resulting code
But code that is more maintainable in the long run is not always more readable now.
In many cases these refactorings are temporary
only meant to allow you understand the code
can be rolled back once your understanding is solidified
When reading a call to a method with a vague name, you will need to spend some time up front to understand what the method does
Inlining the method lowers your extraneous cognitive load
Within the new context, you might also be able to choose better name for the method
Example - Lambdas : if lambdas are new to you, this code might cause too much extraneous cognitive load
If you feel that you are struggling with the lambda expression :
Simply rewrite the code to use a regular function temporarily
Other examples : Ternary operators
What is easy to read depends on your prior knowledge
There is no shame in helping yourself understand code by translating it to a more familiar form
CODE SYNONYMS ARE GREAT ADDITIONS TO A FLASHCARD DECK
While there’s no shame in changing code temporarily to aid comprehension, this does point to a limitation in your understanding.
There are two ways in which code with a complicated structure can overload the working memory :
You do not know exactly which parts of the code you need to read
read more of the code than is needed
may be more than your working memory is able to process
With code that is highly connected, your brain is trying to do two things at the same time :
Understand individual lines of code
Understand the structure of the code to decide where to continue reading
When you reach the limits of your working memory, you can use a memory aid to help you focus on the right parts of the code.
Print out the code, or convert it to a PDF
Open it on a tablet so you can make annotations digitally
Circle all the variables
Link similar variables
draw lines between occurrences of the same variable
helps you to understand where data is used in the program
Circle all method/function calls
Link methods/functions to their definitions
Circle all instances of classes
Draw a link between classes and their instances
Now have a reference that you can refer to for information about the code’s structure :
Saving you the effort of, for example, having to search for definitions
While also deciphering the meaning of the code
A state table focuses on the values of variables rather than the structure of the code. It has columns for each variable and lines for each step in the code.
How to ?
Make a list of all the variables
Create a table and give each variable its own column
Add one row to the table for each distinct part of the execution of the code.
Rows in the state table represent separate parts of the dependencies
Execute each part of the code and write down the value each variable has afterward in the correct row and column.
Once you’ve prepared the table, work your way through the code and calculate the new value of each variable for each row in the state table.
The process of mentally executing code is called tracing or cognitive compiling.
These techniques focus on different parts of the code :
Dependency graph draws your attention to how the code is organized
The state table captures the calculations in the code
Understanding what types of information variables hold is key to being able to reason about and make changes to code.
The reason that variables are hard to understand is that most programmers do not have a good schema in their long-term memory to relate variables.
With just 11 roles, you can describe almost all variables : (distinguished by Sajaniemi)
Role
Description
Examples
Fixed value
Variable whose value does not change after initialization plays the role of a fixed value.
pi
data read from a file
Stepper
When iterating in a loop : always a variable that is stepping through a list of values
i in for loop
Flag
A variable used to indicate that something has happened or is the case
are is_set
is_available
is_error
Walker
A walker is a variable that traverses a data structure in a way that is unknown before the loop starts.
a search index in a binary tree
Most-recent holder
A variable that holds the latest value encountered in going through a series of values.
store the latest line read from a file : line = file.readline()
Most-wanted holder
Often when you are iterating over a list of values, you are doing that to search for a certain value. The variable that holds that value, or the best value found so far, is what we call a most-wanted holder.
minimum value
maximum value
the first value meeting a certain condition
Gatherer
A variable that collects data and aggregates it into one value.
sum = 0 that uses in a loop
Container
Any data structure that holds multiple elements that can be added and removed.
lists
arrays
trees
Follower
Some algorithms require you to keep track of a previous or subsequent value. A variable in this role is called a follower, and is always coupled to another variable.
A pointer that points to a previous element in a linked list
Organizer
Sometimes a variable has to be transformed in some way for further processing.
Often, they are temporary variables
Store a sorted version of a given list
Temporary
Variables that are used only briefly.
Used to swap data, or to store the result of a computation that is used multiple times.
often called temp or t
For most professional programmers :
The roles in Sajaniemi’s framework will be somewhat familiar (maybe by other names)
Rather than introducing new concepts, the purpose of this list is to give you a new vocabulary to use when discussing variables.
You can create a set of icons corresponding to the 11 roles that a variable can play according to Sajaniemi’s framework, and use them to mark the roles of variables in unfamiliar code.
Text knowledge vs plan knowledge
Text structure knowledge : means knowing the syntactic concepts used in code
Plan knowledge : means understanding the intentions of the code’s creator
4 steps commonly taken when moving from superficial knowledge of a program to deeper understanding are:
Find a focal point
Start your exploration of the code at a certain focal point.
It may be the main() method
Expand knowledge from the focal point.
Look for relationships in the code.
Starting at the focal point, circle all the relevant entities (variables, methods, and classes) that play a role
Understand a concept from a set of related entities.
There are several lessons from the lines highlighted
Is there a method that is called in several places within the slice you’ve highlighted?
That method likely plays a large role in the codebase and warrants further investigation
Once you have investigated the important locations further, you can create a list of all related classes
Understand concepts across multiple entities.
Get a high-level understanding of the different concepts in the code
What are you allowed to do, and what is forbidden?
For example, will an error be thrown if you add a third node or is that up to the user?
Evidence from fMRI about what code does in the brain
Siegmund's findings reliably showed that program comprehension activates five Brodmann areas, all located in the left hemisphere of the brain.
The regions are BA6, BA21, BA40, BA44, and BA47
BA21, BA44, and BA47 are related to natural languages
There are many similarities between reading code and reading natural language, and your ability to learn a natural language can be a predictor of your ability to learn to program.
If you can learn English, you can learn Clojure !!!
There has been a lot of research into effective reading strategies and how to learn them. Strategies for reading comprehension can be roughly divided into these seven categories :
Activating—Actively thinking of related things to activate prior knowledge
Actively thinking about code elements will help your working memory to find relevant information stored in the LTM
That might be helpful in comprehending the code at hand
Monitoring—Keeping track of your understanding of a text
When reading code, it is important to keep track of what you are reading and whether or not you understand it
Determining importance—Deciding what parts of a text are most relevant
What matters is that you think about which parts of the code are likely to have the most influence on the program’s execution.
Inferring—Filling in facts that are not explicitly given in the text
Inferring the meaning of variable names
Visualizing—Drawing diagrams of the read text to deepen understanding
One technique that can be helpful for very complex code of which a deeper understanding is needed is to list all operations in which variables are involved.
Questioning—Asking questions about the text at hand
Asking yourself questions while reading code will help you understand the code’s goals and functionality.
Ex : What are the five most central concepts of the code? (identifier names, themes, classes, or information found in comments), what strategies did you use to identify the central concepts?
Summarizing—Creating a short summary of a text
Writing a summary of code in natural language will help you gain a deeper understanding of what’s happening in that code
Models are simplified representations of reality
The main goal = support you in thinking about a problem and ultimately also solving the problem
How you represent a problem can heavily influence the way you think about it
For example : thinking of customers as a list versus as a collection can influence how you store and analyze customer objects
Mental models are mental representations that we form while thinking of problems.
People can hold multiple mental models that can compete with each other.
Notional machines are abstract versions of how a real computer functions that are used when explaining programming concepts and reasoning about programming.
Sometimes bugs are the result of sloppiness :
For example when you forget to close a file or make a typo in a filename.
More often though, bugs are the result of a mistake in thinking.
Keywords and mental models that are stored in your long-term memory can help you comprehend code Sometimes, when you’ve learned something, that knowledge is also useful in another domain.
This is called transfer.
Transfer happens when information that you already know helps you to do new things.
When you activate your working memory by thinking about a new programming concept
The LTM is also activated and starts a search for relevant information.
Transfer of learning happens when you can apply things that you already know in entirely unfamiliar situations. When people are talking about cognitive science and use the term transfer, they almost always mean transfer of learning.
High and Low-road transfer
Transfer of automatized skills is called low-road transfer.
In programming, low-road transfer might occur if you use Ctrl-C and Ctrl-V in a new editor without thinking about it.
Transfer of more complex tasks is called high-road transfer.
When high-road transfer occurs, you are often aware that it is happening
Ex : you might assume that you need to declare a variable in a new programming language because you know that is the case in most languages.
Near and far transfer :
Near transfer happens when knowledge transfers from domains that are seen as close to each other :
like calculus and algebra
C# and Java
Far transfer when skills transfer between very different domains
like Latin and logic
Java and Prolog
Sometimes existing knowledge helps you learn faster or perform new tasks better. This is called positive transfer.
When existing knowledge prevents you from learning something new, we call this negative transfer.
Wrong assumptions based on previous leraning
Ex : Exception in java vs C#
Similar languages but not identical
Checked Exceptions in java (exceptions checked at compile-time)
No try / catch -> will not compile
People from C# won't realize these exceptions from what the are used to
Those people have the wrong mental model, but they think they have the right one!
You can use positive transfer to learn new things more effectively by actively searching for related information in your long-term memory (for example, by elaboration, as covered earlier in the book).
You may hold misconceptions, which occur when you are sure you are right but are actually wrong
Misconceptions are not always addressed by simply realizing or being told that you are wrong.
For misconceptions to be fixed, you need a new mental model to replace the old, wrong model.
Even if you have learned a correct model, there is always the risk that you will fall back to using the misconception.
Use tests and documentation within a codebase to help prevent misconceptions.
Good names help to activate your LTM to find relevant information that you already know about the domain of the code
Bad names can lead you to make assumptions about the code leading to misconceptions.
4 main reasons why identifier names matter :
Names make up a large part of codebases
Eclipse source code : tokens = 33%, characters = 72%
Names play a role in code reviews
Analyzed over 170 reviews with over 1000 remarks in them
Findings :
1 in four code reviews contained remarks related to naming
Remarks about identifier names occurred in 9% of all code reviews.
cf Miltiadis Allamanis study
Names are the more accessible form of documentation
Serve as an important type of documentation : right inside the codebase
Names can serve as beacons
Help readers make sense of code
Choosing a good name is important
Many different researchers have tried to define what makes a variable name good or bad
Different perspectives on this question
Simon Butler, associate Senior Lecturer at the Open University in the UK, created a list of issues with variable names
Name
Description
Example of a bad name
Capitalization Anomaly
Identifiers should use proper capitalization.
page counter
Consecutive underscores
Identifiers should not contain multiple consecutive underscores.
page__counter
Dictionary words
Identifiers should consist of words, and only use abbreviations when they are more commonly used than the full words.
pag_countr
Number of Words
Identifiers should be composed of between two and four words.
page_counter_
converted_and_
normalized_value
Excessive words
Identifiers should not be composed out of more than four words.
page_counter_
converted_and_
normalized_value
Short Identifier Name
Identifiers should not consist of fewer than eight characters, except for c, d, e, g, i, in, inOut, j, k, m, n, o, out, t, x, y, z.
P, page
Enumeration Identifier Declaration Order
Unless there are clear reasons to do so, enumeration types should be declared in alphabetical order.
CardValue = {ACE, EIGHT, FIVE, FOUR, JACK, KING...}
External Underscores
Identifiers should not begin or end in an underscore.
__page_counter_
Identifier Encoding
Type information should not be encoded in identifier names using Hungarian notation or similar.
int_page_counter
Long Identifier Name
Long identifier names should be avoided where possible.
page_counter_
converted_and_
normalized_value
Naming Convention Anomaly
Identifiers should not combine uppercase and lowercase characters in non-standard ways.
Page_counter
Numeric Identifier Name
Identifiers should not be composed entirely of
numeric words or numbers.
FIFTY
If the same word is used for similar objects across a codebase
it will be easier for the brain to find relevant related information stored in the LTM
Proved by Allamanis work on code reviews
It helps make sense of the names you are reading
Researcher
Perspective
Fits cognition because
Allamanis
Names should be consistent across a codebase
Supports chunking
Butler
Syntactically similar names
Lower cognitive load while processing names
When you read a name, the name firstly will be broken up into separate chunks
The name is then sent to the working memory
At the same time, the long-term memory is searched for information related to the different parts of the variable name
Related information from the long-term memory is also sent to the WM
Names can support your thinking about :
Domain of the code
Word like “customer” will have all sorts of associations in your LTM
A customer is probably buying a product, needs a name and an address, and so on
Programming
Concept like a tree will also unlock information from the LTM
A tree has a root, can be traversed and flattened, and so on
Conventions
Contains information about conventions your LTM is aware of
For example, a variable named j will remind you of a nested loop
Hofmeister (researcher at the University of Passau)
72 professional C# developers
Found that participants found on average 19% more defects per minute when reading programs in which the identifiers were words
Compared to letters and abbreviations.
Study investigating the differences in comprehension between variables written in camelCase and those written in snake_case : https://ieeexplore.ieee.org/abstract/document/5090039 :
Show that the use of camelCase leads to higher accuracy among both programmers and non-programmers
There is a 51.5% higher chance of selecting the right option for identifiers written in the camelCase style.
Study available here : http://oro.open.ac.uk/17007/1/butler09wcreshort_latest.pdf
Select the concepts to include in the name
Domain-specific
Intent of the name : should represent what information the object
holds and what it is used for
When you feel the need to write a comment to explain the name
wording from the comment should probably be included in the variable name
Choose the words to represent each concept
A project lexicon : important definitions are noted and alternatives for synonyms are registered
Can help programmers select names consistently
Construct a name using these words
Constructing a name using the chosen words
Comes down to selecting one of the naming molds
Code Smell
Explanation
Level
Long method
A method consists of many lines of code, performing different calculations.
Method level
Long parameter list
A method has many parameters
Method level
Switch statements
Code contains large switch statements, while polymorphism could also be used to ease the code
Method level
Alternative classes with different interfaces
Two different classes seem different at first glance but have similar fields and methods
Class level
Primitive obsession
Overuse of primitive types in a class
Class level
Incomplete library class
Methods are added to ‘random’ classed instead of to a library class
Class level
Large class
A class has too many methods and fields, and it is not clear what abstraction the class provides.
Class level
Lazy class
A class is doing too little to justify its existence
Class level
Data class
Classes should not contain just data, they should contain methods as well
Class level
Temporary field
Classes should not contain unnecessary temporary fields
Class level
Data clumps
Data are often used in combination belong together and should be stored together in a class or structure
Class level
Divergent change
Code changes should in general be local, preferably to one class. If you have to make many different changes in different places, that indicates a poor structure in the code
Codebase level
Feature Envy
When methods on class A are references a lot from class B, they belong in B and should be moved there.
Codebase level
Inappropriate intimacy
Classes should not be connected to other classes extensively
Codebase level
Duplicated code or code clones
The same or very similar code occurs in multiple different places in a codebase
Codebase level
Comments
Comments should describe why the code is there not what it does
Codebase level
Message chains
Long chains of message calls, where methods call methods call methods, etc
Codebase level
Middle Man
If a class is delegating too much responsibility, should it exist?
Codebase level
Parallel inheritance
When you make a subclass of one class, you need to make a subclass of another. This indicates that the functionality of both classes might belong in one class.
Codebase level
Refused bequest
When classed inherit behavior that they do not use, the inheritance might not be necessary.
Codebase level
Shotgun surgery
Code changes should in general be local to one class. If you have to make many different changes in different places, that indicates a poor structure in the code
Codebase level
Speculative generality
Do not add code to a codebase “just in case”, only add features that are needed.
Codebase level
Knowing what we know about the working memory : we can understand why Long parameter list and Complex switch statements are hard to read :
Both smells are related to an overloaded working memory
Newer research even suggests that the capacity might be as low as 6
While reading code, it will be impossible to hold all the parameters in working memory
as such, the method will be harder to understand
Code smells are parts of code that suffer from structural anti-patterns: the code is correct, but not structured in a way that is easy to process.
Arnaoudova’s six categories of linguistic anti-patterns :
Methods that do more than they say
Methods that say more than they do
Methods that do the opposite than they say
Identifiers whose name says that they contain more than what the entity contains
Identifiers whose name says that they contain less than what the entity contains
Identifiers whose name says that the opposite than the entity contains
List of Linguistic Anti-patterns available here
She studied their occurrence in 7 open-source projects :
11% of setters also return a value in addition to setting a field
2.5% of methods, the method name and the corresponding comment gave opposite descriptions of the working of the method
64% of identifiers starting with ‘is’ turned out not to be Boolean
Measuring Cognitive Load
Cognitive load = the overloading of the working memory
How to measure it ?
PAAS scale : participants are asked to self-rate the cognitive load on this nine-point scale ( from "very, very low mental effort" to "very high mental effort")
Skin based measurement : skin temperature / sweat
Brain based measurement : fMRI scanner
Electroencephalogram
Functional near infra-read spectroscopy (FNIRS)
Results from the fNIRS device :
Presence of linguistic anti-patterns in the source code significantly increases the average oxygenated blood flow that a participant experiences
aka cognitive load that is induced by the snippet is higher
When reading a conflicting name
The wrong information will be presented to you
Ex : reading a function name retrieveElements()
Think of information on functions returning a list of things
Gives you the idea you could sort, filter, or slice the returning element
While that is not true for a single element
Lead to “mischunking”
When you are reading a variable name such as isValid
Assume the variable is a Boolean
No need for your brain to dig deeper
By trying to save energy, your brain has made a wrong assumption
Problem solving means :
Traversing the state space in the optimal way
Reaching the goal state in as few steps as possible
How to ?
Understanding the problem
Devising a plan
Carrying out the plan
LTM can store different types of memory :
Procedural (or implicit) memory showcases
How to do something
Ex : How to run a bike
Declarative (explicit) memory consists of memories we are explicitly aware of (facts you can remember), divided into things :
You have experienced and that are stored in episodic memory
memories of experiences
Ex : -> meeting your spouse
Facts you know that are stored in semantic memory
Memory for meanings, concepts, facts
Ex : 5 x 7 = 35
Research shows that especially experts heavily rely on episodic memory when solving problems : instead of finding a new solution, they rely on solutions that have previously worked for similar problems.
Unlearning :
Having a lot of implicit memory can harm your flexibility
Ex : once you have learned how to touch type on a Qwerty keyboard, learning to use a Dvorak keyboard will be harder than if you had never learned Qwerty
Unlearning implicit memory > learned a second programming language with a syntax quite different from the first programming language that you learned
Moving from C# to Java
Once you have practiced a skill so many times that you can do it without thinking about it, like walking, reading, or tying your shoelaces, we say that you have automatized this skill.
Automatization of programming skills is key to being able to solve larger and more complex problems :
The more implicit memories you have for programming
The easier it will be to solve larger problems
Because you will have more cognitive load to spare to think about the problem
Implicit memories are formed in 3 different phases :
Cognitive phase : when you learn something new
In this phase : a new piece of information needs to be split explicitly in smaller parts and you have to explicitly think about the task at hand
That is why it is called the cognitive phase.
Associative phase : you need to actively repeat the new information until patterns of response emerge.
You see an opening bracket and you get nervous if you do not see the closing one
Autonomous phase : In the autonomous phase (also called the procedural phase) the skill is perfected.
Ex : you have reached the autonomous phase when you index into a list and you always do it correctly, no matter the context, the data type or the operations on the list.
Here you have automatized the skill
Automatization will make you program quicker
By creating a large repository of techniques (skeptics might call them “tricks”)
we can create an ever-growing toolbox of new techniques
Gordon Logan, an American psychologist
Automatization is done by retrieving memories from the episodic part of the long-term memory
When you are confronted with a similar task, rather than reasoning about the task---as someone might do who lacks enough instance memories
you can remember how you did it before and apply the same method again
According to him
automatization is complete when you fully rely on episodic memory and do not use any reasoning at all
Use deliberate practice to improve skills
Use very small tasks and execute them repeatedly, until you have reached perfection
When you are struggling with creating for loops without errors
Deliberately typing 100 for loops
Building these small skills will help you to solve larger problems with greater ease
It frees up cognitive load for larger problems
Just like with flashcards, spaced repetition is key to learning
Set some time aside every day to practice and continue until you can consistently perform the tasks without any effort.
Deliberately study how others have solved problems
Study worked examples
Worked examples explain how to solve the problems in detail
Australian professor John Sweller
Sweller taught children mathematics by having them solve algebra equations
Frustrated by how little they were learning from only working on traditional algebra problems
Experiment : divided the students into two groups, who were both asked to solve typical algebra equations, like a = 7 – 4a, solve for a.
Group 2 : received the algebra equations
but also received worked examples of the equations
Can think of worked examples are something like a recipe which describes in detail the steps that are needed to solve the equations.
Group 2 solved it 5 times faster
Here is the kicker:
Children in Group 2 also performed better on different problems
For which calculations rules could be used which were present in the recipe, like subtracting the same value from both sides of an equation or dividing both sides of the equation by the same
We have seen that explicitly studying code, plus studying the process of how it was created can help you to strengthen your programming skills :
Collaborate with a colleague
start a “code reading club”
can exchange code and its explanation, and learn from each other
Explore github
if you choose a repository of which the domain is at least a bit familiar to you, so there are not too many unfamiliar words and domain concepts causing additional extraneous load and you can focus on the programming itself
Read books / blog post about source code
Different activities while programming :
CDN is a framework to evaluate the cognitive impact of a programming language or code base
Cognitive Dimensions of Notations
CDN describes 5 activities
Searching : looking through a code base, searching for a specific piece of information
Location of a bug
Calls of a certain method or the location of something (variables, ...)
Mainly hard on your short-term memory
Best supported by making notes to offload some of the memories to paper, or to a separate document on your code
Comprehension : reading and executing code to gain an understanding of its functionality
Running test code as complement
Transcription : the activity where you are “just coding”.
You have a concrete plan of what you want to add or change to the code base, and you are going to do just that.
Incrementation : mix of all of the above
adding a new feature
Likely to include both :
searching for the location(s) to add code
comprehending the existing code to understand where to add code and how to do that
if you know the code base very well, your working memory and short-term memory might not be impacted by searching and comprehending the code
Exploration : you are in essence sketching with code.
Have a vague idea of where you want to go
But by the act of programming you are gaining clarity about the domain of the problem and about the programming constructs that you will need to use
About 20 percent of a developer’s time is spent on interrupts (Van Solingen study)
According to Parnin’s study :
The average programmer will have just one uninterrupted two-hour session in a day
It takes about a quarter of an hour to start editing code after an interruption.
When interrupted during an edit of a method, only 10% of times programmers could resume their work in less than a minute.
3 techniques :
Store your mental model :
Nakagawa’s results showed a warm-up period in comprehension activities :
Spent on building a mental model of the code at hand
If parts of the model are stored apart from the code, that can help quickly regain your mental model.
Comments can be an excellent location to leave notes about your mental model.
Help your "Prospective memory"
memory of remembering to do something in the future
Related to planning / problem solving
TODO comments in the part of the code that they are working on to remind them to complete or improve part of the code
Parnin also developed a Visual Studio plugin that allows you to support your prospective memory when you have to stop programming
Allows to add TODO items to your code
and give the TODOs an expiry date
so you do not forget to actually work on tasks
Label subgoals
explicitly write down into what small steps a problem can be divided.
Be interrupted at more convenient times with FlowLight :
a physical light that a developer can place on their desk or on top of their screen
Based on computer-interactions, like typing speeds and mouse clicks
Detects whether the programmer is deeply engaged in a task and experiencing high cognitive load
Cognitive Dimensions of Notations :
Dimension
Description
Error-proneness
In some programming languages it is easier to make a mistake than in other languages (dynamic languages like JS). inconsistent conventions, a lack of documentation or vague names.
Consistency
Another way to examine how people will interact with a programming language or codebase is consistency: how similar are similar things?
Diffuseness
How much room or space a programming construct takes, not only concern the number of lines of code, you could also consider how many chunks the code consists of
Hidden dependencies
indicates to what extent dependencies are visible to the user.
Ex : HTML page with a button controlled by JavaScript, which is stored in a different file.
Provisionality
How easy it is to 'think using the tool'
Viscosity
how hard it is to make changes in a certain system.
Typically, codebases written in dynamically typed languages are a bit easier to change.
Progressive evaluation
how easy it is in a given system to check or execute partial work.Some programming systems allow programmers to do live programming
Role expressiveness
how easy it is to set the role of different code parts in a program. Brackets at the end of a function are an example of role expressiveness.
Closeness of mapping
how close the programming language or the code is to the domain in which problems are solved
Hard mental operations
Some systems require a user to think very hard, to perform hard mental operations, outside of the system. For example, asking users to memorize a large number of parameters to call in the right order is a hard mental operation
Secondary notation
indicates the possibility for the programmer to add extra meaning to code, which is not in the formal specification.
Abstraction
describes whether a user of your system can create their own abstractions that are as powerful as the build-in abstractions.
Visibility
how easy it is to see different parts of a system.
In a codebase, it can be hard to see what classes the codebase consists of, especially if code is divided over different files.
The list of Cognitive Dimensions can be used as a sort of a checklist for a codebase.
Not all dimensions matter for all codebases
Investigating each one and deciding how your codebase is doing regularly will help you maintain the usability
Making changes to a codebase to improve a certain dimension in a codebase is called a Design Maneuver.
Examples :
adding types to a codebase is a Design Maneuver that improves error-proneness
Changing function names to be more in line with the domain of the code is a Design Maneuver that improves Closeness of mapping
Dimension
Helps
Harms
Error-Proneness
Incrementation
Consistency
Searching, Comprehension
Transcription
Diffuseness
Searching
Hidden Dependencies
Searching
Provisionality
Exploration
Viscosity
Transcription, Incrementation
Progressive evaluation
Exploration
Role expressiveness
Comprehension
Closeness of mapping
Incrementation
Hard mental operations
Transcription, Incrementation, Exploration
Secondary Notation
Searching
Abstraction
Comprehension
Exploration
Visibility
Comprehension
A senior developer throws lots of new information at a newcomer
The amount of information is too much to process
Causing high cognitive load
After the introduction, the senior developer asks the newcomer a question or gives the newcomer a task.
The newcomer fails
Because of the high cognitive load
Caused by a combination of a lack of relevant chunks for the domain and/or the programming language, and a lack of automatized skills relevant
One of the reasons that more senior people often struggle with effectively teaching and explaining is the "curse of expertise." Once you have mastered a certain skill sufficiently, you will inevitably forget how hard it was to learn that skill or knowledge.
Limit tasks to one programming activity :
One of the issues in the onboarding scenario above is that you ask the newcomer to perform at least four different activities:
Searching for the right place to implement the feature or for relevant information
Comprehending new source code
Exploring the codebase to gain a better understanding
Incrementing the codebase with a new feature
During an onboarding process, it is best to specifically choose activities in each of the five categories, and have newcomers do them one by one.
Activity
Example use to support onboardees
Exploration
Browsing the codebase to get a general sense of the codebase
Searching
Find a class that implements a certain interface
Transcription
Give the newcomer a clear plan to implement a certain method which has to be implemented
Comprehension
Understand aspects of the code, for example summarize a specific method in natural language
Incrementation
Add a feature to an existing class, including the creation of the plan for the feature.
Support the memory of the onboardee
Support the LTM : explain relevant information
Prepare the onboarding process for newcomers by deeply understanding the relevant information that plays a role when working with the codebase
Separate domain learning from exploring code
If you want a newcomer to understand a certain piece of the code, ask them to understand a piece rather than giving them an implementation task.
For example, ask them to write a summary of an existing class or write down all classes involved in executing a certain feature.
Another technique you can use to onboard newcomers is to read code as a team collaboratively. In Chapter 5, we proposed seven techniques from natural language to be applied to code reading.
Activating. Actively thinking of related things to activate prior knowledge.
Determining Importance. Deciding what parts of text are most relevant.
Inferring. Filling in facts that are not explicitly given in the text.
Monitoring. Keeping track of you understanding of a text.
Visualizing. Drawing diagrams of the read text to deepen understanding.
Questioning. Asking questions about the text at hand.
Summarizing. Creating a short summary of text.
