Unify patterns in UCS and UPS

[chengluyu]

The Updated UCS and UPS Compilation Pipelines

• Term: If now contains SimpleSplit, which represents splits using nested patterns. Nested patterns are represented by Pattern, which is also used by pattern definitions.
• Elaborator: Since the elaboration of split is implemented by several mutually recursive methods, I placed them in a trait called SplitElaborator to control the length of Elaborator and maintain modularity.
  • At this step, we report semantic errors that do not require context, such as mixing do and then, as well as some obvious unreachable cases, like having more branches after an else.
• Lowering is now responsible for most of the work.
  • SimpleSplits in UCS expressions are expanded to Splits using SplitCompiler, then normalized by Normalization, and finally lowered to Block.
  • The naïve compilation (generation of unapply and unapplyStringPrefix is now done by calling SplitCompiler for each PatternDef in Lowering.
  • The efficient compilation (done by Instantiator and Compiler) is invoked by SplitCompiler when it sees @compile annotated on patterns.
  • At this step, we perform all the checks on arguments and parameters.

[LPTK]

Where is this unit from?

[chengluyu]

I guess it is because the pattern was parsed as part of the annotation, and nothing followed it until the closing parenthesis. As a result, the annotation was not applied to anything. Therefore, the parenthesis is treated as an empty one, which means it becomes unit.

[LPTK]

Is the then parsed as part of the annotation? 🤯

[chengluyu]

if str is @compile ((" " | "\t") ~ rest) then trimStart(rest) else str

The pattern is parsed as part of the annotation. The parser wants to parse a term next, but found then.

[chengluyu]

I just finished the final major refactor of this PR. The latest commit successfully declutters the Normalization class.

For historical reasons, Normalization not only checked the number of pattern parameters and reported errors, but also performed the efficient and backtracking-free pattern compilation. This made the code look very bloated.

Now, we check the number of arguments in the pattern and call efficient pattern compilation in the process of expanding splits with nested patterns into splits with flattened patterns (handled by SplitCompiler). Since SplitCompiler is also responsible for the naïve compilation of patterns, this refactoring not only reduces duplicate code but also makes the error messages about patterns in UPS and UCS more consistent.

Later today I will do the final minor wrap-up for this PR.