RFC: Minification

[Boshen]

(placeholder)

[Boshen]

Some facts:

• Mangling is super fast. This works on the symbol table and only affects printing. Turning symbol generation can improve gzip sizes. This accounts for ~30% of size reduction.
• Whitespace removal is super fast. This is just an alternative printer. This accounts for ~50% of size reduction.
• Constant folding is not fast, it's an interpreter.
• Changing expressions and statements is super slow
  • Terser runs this in a loop and stops when nothing changes (a.k.a fixed point).
  • This can be changed with the passes option in terser
  • This is where the last 10% minification size difference comes from, for different minifiers
  • People would opt in for longer minification duration for smaller sizes

[Boshen]

Idea: setup mangle and benchmark with other swc and esbuild. Defer implementation of compress until we are 100% sure on how to do it.