------ IN PROGRESS ------

Elise: Schema-Specialized Execution Model

/eˈliːs/ → pronounced like “eh-LEES”

Grammar Rules, Todos

Overview

Elise is a strongly-typed, schema-driven language for processing structured data efficiently. Programs are compiled against schemas rather than specific datasets, separating code correctness from data correctness. Execution can be optimized or unsafe depending on user requirements.

File Types

Only .csv files are supported for now.

Extension	Purpose
.eli	Source code
.elt	Schema / type definitions for input data
.csv	Input data file
.elc	Compiled, schema-specialized bytecode with metadata

Execution Modes

1. Safe Direct Execution

elise --mode=run --source-code=sample.eli --data=data.csv --data-schema=data.elt

• Compiles in-memory (no .elc output)

• Performs full runtime validation of input data against schema

• Executes immediately

Safety: High

Performance: Medium

2. Execute Pre-compiled (Optimized, Safe)

Step 1 — Build an executable

elise --mode=build --source-code=sample.eli --data-schema=data.elt --output=program.elc

Step 2 — Execute

elise --mode=exec --executable=program.elc --data=data.csv

• Uses precompiled schema-specialized bytecode

• Performs minimal runtime validation (structural + parsing checks)

• Executes optimized pipeline

Safety: High

Performance: High

3. Unsafe Execution (Maximum Performance)

Step 1 — Build an executable

elise --mode=build --source-code=sample.eli --data-schema=data.elt --output=program.elc

Step 2 — Execute

elise --mode=exec --executable=program.elc --data=data.csv --unsafe-assume-valid

• Requires precompiled .elc

• Skips runtime validation

• Executes fastest possible path

Use case: trusted, prevalidated, or immutable data

Safety: None ⚠️

Performance: Maximum

4. Validation-Only Step

elise --mode=validate --data=data.csv --data-schema=data.elt

• Full scan of data to ensure strict schema compliance

• Can be used before unsafe execution

Technical features

Distributed Pipelines & Loop fusion

For sequential data transformations compiler can decide whether to run in in single of multi thread.

Example (pseudo code):

pipe big_dataset
    map(parse)
    map(transform)
    reduce(sum)

We can also optimize sequential calls by fusing it into one loop.

Before (pseudo code):

pipe numbers
    map(mul(_, 2))
    filter(.gt(_, 10))
    sum

After (pseudo code):

sum = 0
for n in numbers {
    x = n \* 2
    if x > 10 {
        sum += x
    }
}

Constant folding

Pseudo code:

add(mul(2,3), 4)

Compile-time result:

10

Compile-Time Execution

Allow functions to run during compilation.

Example (pseudo code):

primes(generate-primes(1000))

The compiler computes primes and embeds them during compilation so in runtime there is no computation at all.