Currently, the type inference mechanism is very simple in that in infers types for a variable based only on the last seen assignment to the variable. This has shortcomings for scenarios like the following, where the type may differ based on the branch taken.

x = ""

if y == 0:
    x = True

x # inferred as `Literal[True]`, should be `Literal[""] | True`

To make the type checker more accurate in these situations, this PR introduces a primitive form of data-flow analysis heavily based on Pyright and, by extension, the TypeScript compiler. This includes constructing a code-flow graph as an additional lowering step and using it to determine types as part of the type checker. While under active development, this functionality will be gated behind the --experimental_use_code_flow_analysis flag.

Tasks:

• Support constructing a code-flow graph from the current HIR
  • if statements
  • for loops
  • List/dict comprehensions
  • Assignments for params (will do in follow-up)
• Use code-flow graph as part of type inference
  • if statements
  • for loops
• Caching mechanism data-flow analysis results
  • Effective type results (will do in follow-up)
  • Type of reference at specific flow nodes
• Add --experimental_use_code_flow_analysis flag