RML Compiler

This project is a compiler for Rowset Manipulation Language (RML), an experimental SQL-like language designed to query and manipulate CSV files.
The compiler performs lexical analysis, syntax analysis, semantic analysis, intermediate representation (IR) generation, multi-pass optimizations and native x64 code generation.
Further details can be found in assignment-docs.

RML Language Overview

RML uses a ternary select operator to query data. While it relies on dynamic typing at runtime, it enforces static type checking during compilation.

Primitive Types:

• number: IEEE 754 double-precision floating-point values.
• boolean: Standard true or false literals.
• string: Character arrays.
• rowset: Encapsulated rows and columns, reducible to primitive types under specific conditions.

Basic Syntax:

select <column_list> from <product_space> where <condition>;

An RML program must contain at least one select statement.

---

Compiler Design

1. Frontend (Flex/Bison): Parses the source code, checks for syntax and builds the initial abstract syntax tree composition.
2. Semantic Analyzer: Resolves variable scopes using hierarchical symbol tables and validates parameter types for language runtime functions.
3. IR & IC Generation: Translates expression trees into a stack-machine Intermediate Code (IC) format. Outputs include a JSON AST report and a human-readable linear IC text file.
4. Native x64 Code Generation: Translates the linear stack-machine IC directly into native x64 instructions. The compiler loads these instructions into an executable memory block and runs it. It also interfaces with RML runtime library to support operations like console output (print).

---

Usage

cmake -S . -B build && cmake --build build
cd build

# copy input source file and csv files

# run
./project4 <optional optimization parameter> <rml module file name>

Optimization Flags

Combine the following flags using bitwise addition to enable specific compiler optimizations. For example, -p15 enables all optimizations, while -p0 disables them completely.

Flag	Optimization	Description
-p1	Dead Code Elimination	Skips scanner loops and aggregations if the where condition statically evaluates to false.
-p2	Ineffective Expressions	Eliminates expression parts within comma operators that lack side-effects.
-p4	Invariant Code Motion	Optimizes the loop execution template for dynamically calculatable idempotent statements.
-p8	Peephole Optimization	Eliminates trivial conditional jumps immediately following a boolean constant push.

RML Code Examples

1. List the students who took the course code C05 and scored below 85 Present the course ids and names along with student's name name, and the score.

select C.coursename as cousename,
B.firstname + B.lastname as name,
B.sex=="F"?"Female":"Male" as sex,
A.grade as grade
from grades A, students B, courses C
where B.sid==A.sid && number(A.grade)<85;

2. Report the course codes, mean grades for the courses having the mean less than 85

select meangrades.cid as cid, meangrades.meangrade as meangrade from
(select C.cid as cid, (select mean(number(G1.grade)) as grade from grades G1 where G1.cid==C.cid) as meangrade from courses C where (select count() as cnt from grades G2 where G2.cid==C.cid)>0) meangrades
where number(meangrades.meangrade)<85;

3. Report the minimum success by mean observed in courses

select min(number(meangrades.meangrade)) as minmeangrade from
(select C.cid as cid, (select mean(number(G1.grade)) as grade from grades G1 where G1.cid==C.cid) as meangrade from courses C where (select count() as cnt from grades G2 where G2.cid==C.cid)>0) meangrades
where true;

4. Report the courses having the lowest mean grade

select meangrades.cid as cid, meangrades.coursename as coursename, meangrades.meangrade as meangrade from
(select C.cid as cid, C.coursename as coursename, (select mean(number(G1.grade)) as grade from grades G1 where G1.cid==C.cid) as meangrade from courses C where (select count() as cnt from grades G2 where G2.cid==C.cid)>0) meangrades
where number(meangrades.meangrade)==
(select min(number(meangrades2.meangrade)) as minmeangrade from
(select C.cid as cid, (select mean(number(G1.grade)) as grade from grades G1 where G1.cid==C.cid) as meangrade from courses C where (select count() as cnt from grades G2 where G2.cid==C.cid)>0) meangrades2
where true);

5. The list of the courses with no attendant

select C.cid as cid, C.coursename as coursename from courses C
where (select count() as cnt from grades G where G.cid==C.cid)==0;