Kevin Song
I'm a student at UT (that's the one in Austin) who studies things.
Reading Large Codebases
Your Programming Experience So Far
We give you
- A program specification (i.e. what should this program do?)
- A test suite/test cases
- Some skeleton code
You give us
A few hundred lines of code that pass the tests. You write almost all of this code.
In Practice
Code is written once but read many times.
A lot of programming time is spent understanding existing code and how to interface new code with it.
Example: how do I get the standard library HashMap to do (_____)?
Is it possible to get this programming framework to do (________) or do I have to do it myself?
Reading (existing) code is hard
Why?
Existing code has context.
- Some goal its trying to accomplish
- Things that it assumes are true
- Implicitly encodes an appropriate choice of algorithm (e.g. linear vs binary search).
- Decomposition of problems into smaller pieces
- This is not all evident in the code!!!
Reading code is hard
void traverse(Tree* root, Function* fptr){
for(int i = 0; i < root.numChildren; ++i){
traverse(root->children[i], fptr);
}
*fptr(root);
}Software is
BIG
UTCSH is a big project
Most submissions came in between 400 and 1000 lines of non-comment, non-blank C.
UTCSH
Pintos is a Big project
Pintos has 11.5k lines of C (13.5k if you count headers)
Almost 10x as much code!
UTCSH
Pintos
GCC is a BIG project
About 6 million lines of code (600x more)
(not to scale)
UTCSH
Pintos
GCC
Linux is....
28 million lines of code and counting
UTCSH
Pintos
GCC
Linux
The techniques you can use to understand and work with a 1k LOC project will not scale to 1m LOC!
In other words, it won't scale to real software projects!
Big Ideas
Big Ideas
- You cannot hold every piece of the codebase in your head at once.
You cannot hold the entire codebase in your head at once
You cannot hold the entire codebase in your head at once
Corollary: Knowing how to find something can be almost as valuable as knowing it
Understand your system
Understand your system
void traverse(Tree* root, Function* fptr){
for(int i = 0; i < root.numChildren; ++i){
traverse(root->children[i], fptr);
}
*fptr(root);
}Always have a goal
question
A good question usually has a few key properties:
- The answer can be written down in just a few sentences
- Gives you some insight into the context of the system
What's in this code?
Do I need to rewrite this function?
How does this function relate to these data structures?
What does this function assume about its inputs? Are these assumptions valid?
I want to know more about struct Block
I want to know how each of struct Block's members are used in various functions
Other Good Questions
- What are the primary data structures of this program? What do they represent?
- I want to make a summary version of this documentation that I can use later (cheatsheet)
- What are the main functions that I can call in order to work with <thing>?
- Under what conditions is it safe/unsafe to do X?
Jump In
(Once you're ready)
Reading code is hard
This means it's easy to fool yourself into thinking it's much trickier than it is
Example: Linked Lists
Reading Large Codebases
By Kevin Song
Reading Large Codebases
- 191
