Lecture 001

Programming language:

• we have many programming language because they are for different applications

  • some care about floating point (Taichi)
  • some care about parallism
  • some care about arrays
  • some care about persistence
  • some care about data analysis (SQL)
  • some care about fast develop time (Python)
  • some care about deployment (javascript)
  • some care about timely resource management (system programming, Golang)

• About new programming language

  • new programming language is not hard to develop, and not costly, can be changed fast since nobody use it
  • widespread languages are slow to change because training programmers is time consuming
  • languages are invented when there is a new domain, hardware, or need
  • new languages syntax looks like average language (easier to train)
  • there is no universal accepted metrics for language design

Histroy: we used to write byte code. But we developed "speedcode" as first compiler and then FORTRAN I by John Backus. Before 1958, 50% of code was written in FORTRAN.

Five Phasis:

1. Lexical Analysis: divide program text into tokens.
2. Parsing: diagramming sentence by generating parse tree
3. Semantic Analysis: understand "meaning" (ie. catch language inconsistencies and type check)
4. Optimization: make program run faster, reduce power, network, memory by doing
5. Code Generation: language translation

For FORTAN, the semantic analysis part is small and the rest is evenly distributed accross different phasis.

For today's compiler, we have big semantic alalysis phase and very large optimization phase. Other phasis are small.

Lexical Analysis

Lexical Analysis: divide program text into tokens.

Parsing

Semantic Analysis

Inconsistency example:

Jack said Jerry left his assignment at home

// Do you know what "his" is refering to?
// Answer: No

This is a problem in semantic analysis

To avoid this issue we use scope:

{
  int Jack = 3;
  {
    int Jack = 4;
    cout << Jack; // It will print 4
  }
  cout << Jack; // It will print 3
}

Optimization

Consider the following code

X = Y * 0

While this could be optimized to

X = 0

However, this optimization does not always hold. It only holds for integers, but invalid for floating point since NaN * 0 = NaN != 0

Code Generation

Generate lower level code.