JuLC Compiler Design

A progressive guide to the JuLC compiler architecture — from high-level overview to internal implementation details. Intended for anyone from first-time contributors to seasoned compiler engineers.

1. What is JuLC?

JuLC (Java UPLC Compiler) compiles a subset of Java into Untyped Plutus Lambda Calculus (UPLC) — the on-chain execution language for Cardano smart contracts. It lets Java developers write Cardano validators using familiar syntax, types, and tooling (IDEs, Gradle, JUnit) while producing the same on-chain bytecode as Haskell-based Plutus or Aiken.

┌─────────────────┐          ┌─────────────────┐          ┌──────────────┐
│   Java Source    │  ──────► │  JuLC Compiler   │  ──────► │  UPLC Program │
│  (@Validator)    │          │  (julc-compiler)  │          │  (on-chain)   │
└─────────────────┘          └─────────────────┘          └──────────────┘

What JuLC is:

• A source-level compiler (Java source code in, UPLC bytecode out)
• A subset compiler — only a safe, deterministic subset of Java is allowed
• A Plutus V3 compiler targeting the Conway era

What JuLC is NOT:

• Not a JVM bytecode compiler (it reads .java files, not .class files)
• Not a full Java compiler (no classes, inheritance, exceptions, threads, I/O)
• Not a general-purpose transpiler (output is specifically UPLC for Cardano)

The Java Subset

JuLC accepts a functional subset of Java designed for on-chain safety:

Supported	Not Supported
record types	class with mutable fields
sealed interface + pattern matching	Class inheritance
for-each loops, while loops	for(;;) C-style loops, do-while
if/else, switch expressions	try/catch, throw
Optional<T>	null
BigInteger, byte[], String, boolean	float, double, arrays
Static methods	Instance methods, this, super
Immutable variables (final semantics)	Reassignment (except loop accumulators)

---

2. How Cardano Smart Contracts Work

Cardano uses the Extended UTXO (eUTxO) model. Each unspent transaction output (UTxO) can optionally be locked by a validator script. To spend that UTxO, a transaction must provide a redeemer (the “proof” or “action”) and the validator script must return true.

┌──────────────────────────────────────────────────────────────┐
│                        Transaction                           │
│                                                              │
│  Inputs:                        Outputs:                     │
│  ┌──────────────────┐           ┌─────────────────┐         │
│  │ UTxO (locked by   │           │ UTxO (new, may   │         │
│  │  validator script)│           │  be locked too)  │         │
│  │ + Redeemer        │           └─────────────────┘         │
│  └────────┬─────────┘                                        │
│           │                                                  │
│           ▼                                                  │
│  ┌──────────────────────────────────────┐                    │
│  │  Validator Script (UPLC program)      │                    │
│  │  Input: ScriptContext (tx data)       │                    │
│  │  Output: true (accept) / false (fail) │                    │
│  └──────────────────────────────────────┘                    │
└──────────────────────────────────────────────────────────────┘

The ScriptContext passed to every validator contains:

• TxInfo — the full transaction (inputs, outputs, fees, minting, signatories, etc.)
• Redeemer — the action data provided by the transaction submitter
• ScriptInfo — identifies which script purpose triggered this execution (spending, minting, etc.)

Script Purposes

Purpose	When Triggered	Java Annotation
Spending	Consuming a UTxO locked at a script address	@SpendingValidator
Minting	Minting or burning tokens under this policy	@MintingValidator
Withdrawing	Withdrawing staking rewards	@WithdrawValidator
Certifying	Publishing delegation certificates	@CertifyingValidator
Voting	Casting governance votes	@VotingValidator
Proposing	Submitting governance proposals	@ProposingValidator

---

3. From Java to On-Chain: The 10,000-Foot View

The JuLC compilation pipeline transforms Java source through three intermediate representations:

Java Source                     (human-readable)
     │
     │  JavaParser
     ▼
Java AST                        (abstract syntax tree — JavaParser nodes)
     │
     │  PirGenerator
     ▼
PIR (Plutus Intermediate Rep)   (typed, named variables, high-level constructs)
     │
     │  UplcGenerator
     ▼
UPLC (Untyped Plutus LC)        (untyped, De Bruijn indices, minimal)
     │
     │  FlatEncoder + CborEncoder
     ▼
On-chain bytecode               (binary, embedded in transactions)

Why three stages?

• Java AST → PIR: Translates Java constructs (records, loops, method calls) into a typed lambda calculus with Let bindings, LetRec for recursion, and DataMatch for pattern matching. Types are preserved for correct encode/decode insertion.

• PIR → UPLC: Erases types, converts named variables to De Bruijn indices, lowers Let to function application, LetRec to the Z-combinator, and DataMatch to tag-based dispatch. This is the “compilation” step.

• UPLC → Binary: Serializes using FLAT encoding (compact bit-level format) wrapped in CBOR. This is what goes on-chain.

---

Level 2: Module Architecture

4. Project Modules

JuLC is organized into focused modules. Here are the key ones grouped by role:

Core (the foundation)

Module	Role	Key Types
julc-core	UPLC AST, constants, serialization	Term (10 variants), DefaultFun (102 builtins), PlutusData, Program
julc-ledger-api	Java records for all Cardano V3 ledger types	ScriptContext, TxInfo, TxOut, Value, Address, Credential + 35 more

Compiler (the main event)

Module	Role	Key Types
julc-compiler	Java source → UPLC compilation (36 Java files)	JulcCompiler, PirGenerator, PirTerm, PirType, UplcGenerator
julc-stdlib	On-chain standard library (13 libraries, ~65 methods)	StdlibRegistry, ListsLib, MapLib, ValuesLib, etc.

Runtime (execution)

Module	Role	Key Types
julc-vm	VM SPI — pluggable execution backend	JulcVm, JulcVmProvider, EvalResult
julc-vm-scalus	Default VM backend (wraps Scalus CEK machine)	ScalusVmProvider
julc-vm-java	Future pure-Java VM backend	(in development)

Developer Experience

Module	Role
julc-onchain-api	Annotations (@Validator, @Entrypoint, @Param, @OnchainLibrary) + off-chain stubs for IDE support
julc-testkit	ValidatorTest base class for JUnit-based validator testing
julc-testkit-jqwik	Property-based testing support
julc-annotation-processor	Java annotation processor — compiles validators at build time
julc-gradle-plugin	Gradle plugin wrapping the annotation processor
julc-blueprint	Plutus blueprint (CIP-57) generation
julc-cli	Command-line compiler interface

Integration

Module	Role
julc-cardano-client-lib	Integration with cardano-client-lib (transaction building)
julc-e2e-tests	End-to-end integration tests (CIP-113, etc.)
julc-examples	Example validators and library code
julc-bom	Bill of Materials for dependency management

Advanced / Experimental

Module	Role
julc-analysis	Static analysis tooling
julc-decompiler	UPLC → human-readable decompilation
julc-benchmark	Performance benchmarking
julc-bls	BLS12-381 cryptographic operations
julc-vm-truffle	GraalVM Truffle-based VM backend
julc-playground	Web-based compiler playground

---

5. Module Dependency Graph

                          ┌────────────────┐
                          │   julc-bom     │  (Bill of Materials)
                          └────────────────┘

                    ┌─────────────────────────┐
                    │       julc-core          │
                    │  Term, DefaultFun,       │
                    │  PlutusData, Program     │
                    └────────┬────────────────┘
                             │
              ┌──────────────┼──────────────────┐
              │              │                  │
              ▼              ▼                  ▼
     ┌────────────┐  ┌──────────────┐  ┌───────────────┐
     │  julc-vm   │  │julc-ledger-  │  │julc-onchain-  │
     │  (VM SPI)  │  │    api       │  │    api        │
     └─────┬──────┘  │ (40 types)   │  │ (Annotations) │
           │         └──────┬───────┘  └───────┬───────┘
           │                │                  │
     ┌─────┴──────┐  ┌─────┴──────────────────┴────────┐
     │julc-vm-    │  │         julc-compiler             │
     │  scalus    │  │  (Java → PIR → UPLC pipeline)     │
     └────────────┘  │  36 files, 7 packages              │
                     └──────────┬───────────────────────┘
                                │
              ┌─────────────────┼────────────────────┐
              │                 │                    │
              ▼                 ▼                    ▼
     ┌──────────────┐  ┌───────────────┐   ┌───────────────┐
     │ julc-stdlib  │  │ julc-testkit  │   │julc-annotation│
     │ (13 libs)    │  │(ValidatorTest)│   │  -processor   │
     └──────────────┘  └───────────────┘   └───────────────┘
                                │
                       ┌────────┴────────┐
                       │ julc-gradle-    │
                       │   plugin        │
                       └─────────────────┘

Key dependency rules:

• julc-core has no dependency on other julc modules (only cbor-java)
• julc-compiler depends on julc-core and julc-ledger-api but not on julc-vm
• julc-vm depends only on julc-core
• julc-stdlib provides StdlibRegistry used by julc-compiler at compile time
• julc-testkit composes julc-compiler + julc-vm for test-time compile-and-execute

---

6. Key Data Types Across Modules

UPLC Terms (julc-core)

The on-chain language has 10 term variants:

sealed interface Term {
    record Var(NamedDeBruijn name)                    // Variable (De Bruijn indexed)
    record Lam(String name, Term body)                // Lambda abstraction
    record Apply(Term function, Term argument)        // Function application
    record Force(Term term)                           // Force polymorphic term
    record Delay(Term term)                           // Delay evaluation (thunk)
    record Const(Constant value)                      // Constant value
    record Builtin(DefaultFun fun)                    // Built-in function (102 total)
    record Error()                                    // Halt execution
    record Constr(long tag, List<Term> fields)        // Constructor (Plutus V3)
    record Case(Term scrutinee, List<Term> branches)  // Case match (Plutus V3)
}

PIR Terms (julc-compiler)

The intermediate representation adds types and high-level constructs:

sealed interface PirTerm {
    record Var(String name, PirType type)                              // Named, typed variable
    record Let(String name, PirTerm value, PirTerm body)              // Let binding
    record LetRec(List<Binding> bindings, PirTerm body)               // Recursive let (loops)
    record Lam(String param, PirType paramType, PirTerm body)         // Typed lambda
    record App(PirTerm function, PirTerm argument)                    // Application
    record Const(Constant value)                                      // Constant
    record Builtin(DefaultFun fun)                                    // Builtin
    record IfThenElse(PirTerm cond, PirTerm then_, PirTerm else_)    // Conditional
    record DataConstr(int tag, PirType type, List<PirTerm> fields)   // Data constructor
    record DataMatch(PirTerm scrutinee, List<MatchBranch> branches)  // Pattern match
    record Error(PirType type)                                        // Typed error
    record Trace(PirTerm message, PirTerm body)                      // Debug trace
}

PIR Types (julc-compiler)

sealed interface PirType {
    // Primitives
    record IntegerType()
    record ByteStringType()
    record StringType()
    record BoolType()
    record UnitType()
    record DataType()           // Raw untyped Data (escape hatch)

    // Containers
    record ListType(PirType elemType)
    record PairType(PirType first, PirType second)
    record MapType(PirType keyType, PirType valueType)
    record OptionalType(PirType elemType)
    record ArrayType(PirType elemType)    // PV11, CIP-156

    // Functions
    record FunType(PirType paramType, PirType returnType)

    // Algebraic data types
    record RecordType(String name, List<Field> fields)
    record SumType(String name, List<Constructor> constructors)
}

Plutus Data (julc-core)

The universal on-chain value encoding:

sealed interface PlutusData {
    record ConstrData(long tag, List<PlutusData> fields)
    record MapData(List<Map.Entry<PlutusData, PlutusData>> entries)
    record ListData(List<PlutusData> items)
    record IntData(BigInteger value)
    record BytesData(byte[] value)
}

Every Java value eventually becomes PlutusData on-chain. The compiler inserts encode/decode operations at type boundaries.

---

Level 3: The Compilation Pipeline

7. Pipeline Overview

JulcCompiler.compile() orchestrates a 24-step pipeline. Here is the complete flow:

Java Source(s)
     │
     ▼
┌─────────────────────────────────────────┐
│  1. Parse (JavaParser → AST)            │
│  2. Validate (SubsetValidator)          │
│  3. Library check                       │
│  4. Annotated class discovery           │
│  5. Script purpose detection            │
│  6. Type registration (ledger + user)   │
│  7. @Param field detection              │
│  8. Static field detection              │
│  9. Entrypoint discovery                │
│ 10. Parameter validation                │
│ 11. Library compilation (multi-pass)    │
│ 12. Compose stdlib + library lookups    │
│ 13. Symbol table setup                  │
│ 14. Helper method PIR generation        │
│ 15. Entrypoint PIR generation           │
│ 16. Helper method wrapping (Let)        │
│ 17. Static field wrapping (Let)         │
│ 18. Library method wrapping (Let/LetRec)│
│ 19. Validator wrapping (ScriptContext)   │
│ 20. @Param wrapping (outer lambdas)     │
│ 21. UPLC generation                     │
│ 22. Optimization (6 passes, fixpoint)   │
│ 23. Program creation (PlutusV3)         │
│ 24. ParamInfo creation                  │
└─────────────────────────────────────────┘
     │
     ▼
  CompileResult(program, params, diagnostics)

The pipeline can be divided into four major phases:

Phase	Steps	Input → Output
Frontend	1-10	Java source → validated AST + metadata
Middle-end	11-18	AST → PIR term tree
Backend	19-22	PIR → optimized UPLC
Output	23-24	UPLC → serialized Program

---

8. Phase 1: Parsing and Validation

Parsing

JuLC uses JavaParser to parse Java source into an AST:

StaticJavaParser.getParserConfiguration()
    .setLanguageLevel(ParserConfiguration.LanguageLevel.JAVA_21);
CompilationUnit cu = StaticJavaParser.parse(source);

The parser handles Java 21 features: records, sealed interfaces, pattern matching in switch, instanceof patterns.

SubsetValidator

After parsing, SubsetValidator walks the AST to reject unsupported Java constructs. It extends JavaParser’s VoidVisitorAdapter and collects multiple errors:

SubsetValidator rejects:
  try/catch, throw, synchronized, for(;;), do-while,
  null, this, super, new T[], float/double, class inheritance

SubsetValidator allows:
  for-each, while, break (in loops), records, sealed interfaces,
  switch expressions, instanceof patterns, Optional<T>

Each rejection includes a suggestion pointing to the on-chain alternative:

ERROR: null is not supported on-chain
Suggestion: Use Optional<T> to represent absence

Annotation Discovery

The compiler discovers:

1. The validator class — annotated with @SpendingValidator, @MintingValidator, etc.
2. The entrypoint method — annotated with @Entrypoint
3. @Param fields — deployment-time parameters
4. Static fields — compile-time constants

---

9. Phase 2: Type Registration

Before PIR generation can resolve types, all record and sealed interface types must be registered.

Two-Stage Registration

Stage 1: Ledger Types (LedgerTypeRegistry)

Pre-registers ~40 Cardano ledger types in 4 tiers (dependency order):

Tier 1: Leaf records       → TxOutRef, Value, IntervalBound
Tier 2: Sealed interfaces  → Credential, OutputDatum, ScriptInfo, Vote, DRep
Tier 3: Composite records  → Address, TxOut, TxInInfo, Interval
Tier 4: Top-level          → TxInfo (16 fields), ScriptContext

Governance types (Conway era): Vote, DRep, Voter, StakingCredential, Delegatee, TxCert, GovernanceAction, ProposalProcedure, Committee, ScriptPurpose.

Stage 2: User Types (TypeRegistrar)

Processes all compilation units (validator + libraries) together:

1. Collect all record and sealed interface declarations
2. Validate no duplicate type names
3. Build dependency graph (field types create edges)
4. Topological sort (Kahn's algorithm)
5. Register in dependency order

Circular dependencies are detected and reported as errors.

TypeResolver

TypeResolver maps Java types to PIR types:

Java Type	PIR Type
int, long, BigInteger	IntegerType
byte[], PubKeyHash, TxId, PolicyId, …	ByteStringType
boolean	BoolType
String	StringType
void	UnitType
PlutusData and subtypes	DataType
List<T> / JulcList<T>	ListType(resolve(T))
Map<K,V> / JulcMap<K,V>	MapType(resolve(K), resolve(V))
Optional<T>	OptionalType(resolve(T))
User records	RecordType(name, fields)
User sealed interfaces	SumType(name, constructors)

---

10. Phase 3: PIR Generation

This is the heart of the compiler. PirGenerator (2,147 lines) transforms Java AST nodes into PIR terms, assisted by three extracted helper classes.

Architecture After ADR-018 Refactoring

┌──────────────────────────────────────────────────────────┐
│                    PirGenerator (2,147 lines)             │
│  Entry points: generateMethod(), generateExpression()    │
│  Owns: SymbolTable, TypeResolver, StdlibLookup           │
│                                                          │
│  Delegates to:                                           │
│  ┌────────────────────────────────────────────────┐      │
│  │  AccumulatorTypeAnalyzer (432 lines)            │      │
│  │  Pure AST analysis — detects loop accumulator   │      │
│  │  types (List vs Map) from usage patterns        │      │
│  └────────────────────────────────────────────────┘      │
│  ┌────────────────────────────────────────────────┐      │
│  │  TypeInferenceHelper (282 lines)                │      │
│  │  Read-only type inference —                     │      │
│  │  resolveExpressionType, inferPirType,           │      │
│  │  inferBuiltinReturnType                         │      │
│  └────────────────────────────────────────────────┘      │
│  ┌────────────────────────────────────────────────┐      │
│  │  LoopBodyGenerator (531 lines)                  │      │
│  │  Loop body compilation — 5 paths:               │      │
│  │  single/multi-acc × break/no-break + zero-acc   │      │
│  │  Pack/unpack accumulators, nested loop handling  │      │
│  └────────────────────────────────────────────────┘      │
│                                                          │
│  Also uses:                                              │
│  ┌────────────────────────────────────────────────┐      │
│  │  TypeMethodRegistry (905 lines)                 │      │
│  │  Instance method dispatch (~50 methods)         │      │
│  └────────────────────────────────────────────────┘      │
│  ┌────────────────────────────────────────────────┐      │
│  │  PirHelpers (356 lines)                         │      │
│  │  wrapDecode, wrapEncode, list utilities          │      │
│  └────────────────────────────────────────────────┘      │
│  ┌────────────────────────────────────────────────┐      │
│  │  PirHofBuilders (244 lines)                     │      │
│  │  HOF PIR builders (map, filter, any, all, find) │      │
│  └────────────────────────────────────────────────┘      │
└──────────────────────────────────────────────────────────┘

Expression Compilation

Java Expression	PIR Term
42	Const(Integer(42))
true	Const(Bool(true))
"hello"	Const(String("hello"))
x	Var("x", type)
a + b	App(App(Builtin(AddInteger), a), b)
a == b	App(App(Builtin(EqualsInteger), a), b) (type-aware)
!x	IfThenElse(x, false, true)
cond ? a : b	IfThenElse(cond, a, b)
new Point(x, y)	DataConstr(0, RecordType, [x, y])
point.x()	wrapDecode(HeadList(SndPair(UnConstrData(point))), fieldType)

Statement Compilation

Java Statement	PIR Term
int x = 5; rest...	Let("x", Const(5), rest)
return expr	generateExpression(expr) (becomes final term)
someCall(); rest...	Let("_", someCall, rest) (evaluate and discard)
if (c) { t } else { e }	IfThenElse(c, t, e)
switch (s) { case A a -> ... }	DataMatch(s, [branches...])

Binary Operators (Type-Aware)

Op	Integer	String	ByteString	Data
+	AddInteger	AppendString	AppendByteString	—
==	EqualsInteger	EqualsString	EqualsByteString	EqualsData
<	LessThanInteger	—	LessThanByteString	—

---

11. Phase 4: Loop Desugaring

Since UPLC has no loops, all loops are transformed into recursive LetRec patterns.

For-Each Loop

long sum = 0;
for (var item : items) {
    sum = sum + item;
}

Becomes:

LetRec([loop__forEach__0 = \xs \acc ->
    IfThenElse(NullList(xs),
        acc,                                        // base case: return accumulator
        Let(item, wrapDecode(HeadList(xs), elemType),
            loop__forEach__0(TailList(xs), acc + item)))  // recursive case
], loop__forEach__0(items, 0))                      // initial call

While Loop

long n = x;
while (n > 0) {
    n = n - 1;
}

Becomes:

LetRec([loop__while__0 = \n ->
    IfThenElse(n > 0,
        loop__while__0(n - 1),    // recursive case
        n)                         // base case: return accumulator
], loop__while__0(x))

Five Compilation Paths

Path	Accumulators	Break?	Strategy
A	1	No	Simple LetRec with single accumulator parameter
B	1	Yes	Break returns accumulator directly; continue recurses
C	2+	No	Pack accumulators into Data list tuple
D	0	—	Unit accumulator, discard result
E	2+	Yes	Tuple packing + break-aware body

Accumulator Type Refinement

AccumulatorTypeAnalyzer distinguishes ListType vs MapType accumulators by scanning the loop body for evidence:

Evidence Found	Inferred Type
mkNilPairData() initialization	MapType
fstPair()/sndPair() on cursor element	MapType
mkCons() with mkPairData() items	MapType
mkNilData() without pair evidence	ListType

---

12. Phase 5: UPLC Generation

UplcGenerator lowers PIR to UPLC by erasing types and eliminating high-level constructs.

Key Transformations

Let → Application:

Let(x, val, body)  →  Apply(Lam("x", body'), val')

LetRec → Z-combinator (strict fixed-point):

Z = \f -> (\x -> f (\v -> x x v)) (\x -> f (\v -> x x v))

LetRec([name = body], expr)
→ Apply(Lam(name, expr'), Apply(Z, Lam(name, body')))

IfThenElse → Force/Delay (lazy branches):

IfThenElse(c, t, e)
→ Force(Apply(Apply(Apply(Force(Builtin(IfThenElse)), c'),
                    Delay(t')),    ← prevent eager evaluation
              Delay(e')))

DataConstr → ConstrData:

DataConstr(tag, type, [f1, f2])
→ ConstrData(tag, MkCons(encode(f1), MkCons(encode(f2), MkNilData)))

DataMatch → Tag dispatch:

DataMatch(scrutinee, branches)
→ Let(pair, UnConstrData(scrutinee),
    Let(tag, FstPair(pair),
      Let(fields, SndPair(pair),
        IfThenElse(tag == 0, branch0(fields),
          IfThenElse(tag == 1, branch1(fields),
            Error)))))

De Bruijn Indexing: Named variables become integer indices counting lambda binders outward:

\x -> \y -> x   →   Lam("x", Lam("y", Var(2)))
                                          ^^^ cross y(1) then x(2)

Force Counts for Polymorphic Builtins

Forces	Builtins
2 (∀ a b)	FstPair, SndPair, ChooseList
1 (∀ a)	IfThenElse, Trace, MkCons, HeadList, TailList, NullList, …
0 (monomorphic)	Arithmetic, comparisons, crypto, encode/decode

---

13. Phase 6: Optimization

UplcOptimizer runs 6 passes in a fixpoint loop (max 20 iterations, stops when unchanged):

Pass 1: Force/Delay Cancellation

Force(Delay(t)) → t

Undoes the lazy wrapping when the result is immediately forced.

Pass 2: Constant Folding

AddInteger(3, 4) → 7
EqualsInteger(8, 8) → True

Supports: AddInteger, SubtractInteger, MultiplyInteger, EqualsInteger, LessThanInteger, LessThanEqualsInteger, EqualsByteString, AppendByteString.

Pass 3: Dead Code Elimination

Apply(Lam(x, body), val) → body   (when x unused in body, val has no side effects)

Trace calls are considered side-effecting and preserved.

Pass 4: Beta Reduction

Apply(Lam(x, body), arg) → substitute(body, arg)   (when x used once, arg is "simple")

“Simple” = Const, Var, Builtin, or Force of these. Complex args are not inlined.

Pass 5: Eta Reduction

Lam(x, Apply(f, Var(1))) → f   (when x not free in f, f is a value)

Pass 6: Constr/Case Reduction

Case(Constr(tag, fields), branches) → Apply(branches[tag], fields...)

---

14. Phase 7: Wrapping and Encoding

Validator Wrapping

ValidatorWrapper adds ScriptContext decoding and bool→unit/error conversion:

\scriptContextData ->
  let ctxFields = SndPair(UnConstrData(scriptContextData))
  let redeemer = HeadList(TailList(ctxFields))       // field 1
  let result = validate(redeemer, scriptContextData)
  in IfThenElse(result, Unit, Error)

For spending validators with datum (3-param), the wrapper also extracts the datum from ScriptInfo.SpendingScript.

@Param Wrapping

Each @Param field adds an outer lambda:

\param1__raw -> Let(param1, UnIData(param1__raw),
  \param2__raw -> Let(param2, UnBData(param2__raw),
    <validator body>))

Serialization

The final UPLC term is wrapped in Program.plutusV3(term) and serialized:

1. FLAT encoding — compact bit-level binary format
2. CBOR wrapping — standard Cardano on-chain format

---

Level 4: Compiler Internals

15. The PIR Intermediate Representation

PIR serves as the typed bridge between Java and UPLC. It preserves type information that UPLC lacks, enabling the compiler to insert correct encode/decode operations.

Why PIR Exists

UPLC is too low-level to compile to directly:

• No named variables (De Bruijn indices only)
• No types (everything is untyped)
• No Let bindings (only lambda application)
• No loops (only recursion via combinators)
• No pattern matching (only tag extraction)

PIR adds these as first-class constructs, making compilation from Java AST straightforward:

Java Construct	PIR Construct	UPLC Lowering
Variable declaration	Let(name, value, body)	Apply(Lam(name, body), value)
Loop	LetRec([binding], body)	Z-combinator application
Record construction	DataConstr(tag, type, fields)	ConstrData(tag, fieldList)
Pattern matching	DataMatch(scrutinee, branches)	Tag dispatch chain
Method call	App(function, argument)	Apply(function, argument)
Conditional	IfThenElse(c, t, e)	Force(Apply(Apply(Apply(Force(IfThenElse), c), Delay(t)), Delay(e)))

---

16. The Type System

The Data Encoding Problem

On the Cardano ledger, all values are encoded as Data — a universal 5-constructor representation:

Data Constructor	Encodes	Encode Builtin	Decode Builtin
I(integer)	int, long, BigInteger	IData	UnIData
B(bytestring)	byte[], hash types	BData	UnBData
Constr(tag, fields)	Records, sealed interfaces, booleans, Optional	ConstrData	UnConstrData
Map(pairs)	Map<K,V>	MapData	UnMapData
List(items)	List<T>	ListData	UnListData

The compiler must insert encode/decode wrappers at every type boundary. Two key helpers in PirHelpers.java:

wrapDecode(data, targetType) — Extract typed value from raw Data:

IntegerType     → UnIData(data)
ByteStringType  → UnBData(data)
ListType        → UnListData(data)
MapType         → UnMapData(data)
BoolType        → EqualsInteger(FstPair(UnConstrData(data)), 1)
StringType      → DecodeUtf8(UnBData(data))
DataType        → data  (pass through)

wrapEncode(value, type) — Wrap typed value back to Data:

IntegerType     → IData(value)
ByteStringType  → BData(value)
BoolType        → IfThenElse(value, ConstrData(1,[]), ConstrData(0,[]))
StringType      → BData(EncodeUtf8(value))
ListType        → ListData(value)
MapType         → MapData(value)
DataType        → value  (pass through)

Record and Sum Type Encoding

// Record: tag 0, fields in order
record Point(int x, int y) {}
// Point(10, 20) → Constr(0, [I(10), I(20)])

// Sealed interface: ascending tags per variant
sealed interface Shape {
    record Circle(int radius) implements Shape {}    // tag 0
    record Rect(int w, int h) implements Shape {}    // tag 1
}
// Circle(5) → Constr(0, [I(5)])
// Rect(3,4) → Constr(1, [I(3), I(4)])

Boolean Encoding

Booleans map to Constr(1, []) (True) and Constr(0, []) (False), matching the Haskell Plutus convention.

---

17. Data Encoding: The Bridge Between Java and Plutus

Understanding encode/decode insertion is key to understanding the compiler. Here’s when each happens:

Situation	Direction	Example
Record field access	Decode	point.x() → UnIData(HeadList(fields))
Record construction	Encode	new Point(x, y) → ConstrData(0, [IData(x), IData(y)])
List element access	Decode	list.head() → wrapDecode(HeadList(list), elemType)
List prepend	Encode	list.prepend(elem) → MkCons(wrapEncode(elem), list)
Method parameter	Pass-through	Entrypoint params are raw Data
@Param decode	Decode	@Param int fee → UnIData(param__raw)
HOF lambda argument	Decode	list.map(x -> ...) → unwrap x from Data
HOF lambda result	Encode	list.map(x -> x+1) → wrap result to Data

---

18. PirGenerator and Its Extracted Helpers

Responsibility Split (ADR-018)

Before the ADR-018 refactoring, PirGenerator was a 3,369-line monolith. It was decomposed into focused classes:

Class	Lines	Responsibility
PirGenerator	2,147	Core: expressions, statements, method calls, record access, control flow
LoopBodyGenerator	531	Loop body compilation across 5 paths (single/multi-acc × break/no-break)
AccumulatorTypeAnalyzer	432	Pure AST analysis: detect accumulator types from usage patterns
TypeInferenceHelper	282	Read-only type queries: resolve expression types, infer PIR types
TypeMethodRegistry	905	Instance method dispatch: 50+ methods across 11 type categories
PirHelpers	356	Static utilities: wrapDecode, wrapEncode, list operations
PirHofBuilders	244	HOF PIR construction: map, filter, any, all, find, foldl, zip

How They Interact

PirGenerator
  │
  ├── calls ──► AccumulatorTypeAnalyzer.refineAccumulatorTypes()
  │              (before loop compilation, to determine accumulator types)
  │
  ├── calls ──► TypeInferenceHelper.resolveExpressionType()
  │              (during expression compilation, for type-aware dispatch)
  │
  ├── calls ──► LoopBodyGenerator.generateSingleAccBody() / etc.
  │              (loop body compilation; LoopBodyGenerator calls BACK to
  │               PirGenerator for nested expressions/statements)
  │
  ├── calls ──► TypeMethodRegistry.resolve(type, method, args)
  │              (instance method dispatch: list.head(), map.get(), etc.)
  │
  ├── calls ──► PirHelpers.wrapDecode() / wrapEncode()
  │              (encode/decode insertion at type boundaries)
  │
  └── calls ──► PirHofBuilders.buildMap() / buildFilter() / etc.
                 (HOF lambda compilation for list.map(), list.filter(), etc.)

---

19. Method Call Dispatch

PirGenerator.generateMethodCall() implements a 5-level cascading dispatch. After the ADR-018 refactoring, each level is a named method:

private PirTerm generateMethodCall(MethodCallExpr mce) {
    // Level 1: BigInteger.valueOf(n) → identity
    result = tryBigIntegerConstant(mce, scope, method, args);
    if (result != null) return result;

    // Level 2: PlutusData.fromPlutusData() → identity
    result = tryFromPlutusDataIdentity(mce, scope, method, args);
    if (result != null) return result;

    // Level 3: (Type)(Object) cast patterns
    result = tryPlutusDataCast(mce, scope, method, args);
    if (result != null) return result;

    // Level 4: Stdlib static methods (Builtins.*, ListsLib.*, Math.*)
    result = tryStaticStdlibCall(mce, scope, method, args);
    if (result != null) return result;

    // Level 5: Instance methods via TypeMethodRegistry
    return resolveInstanceMethodCall(mce, scope, method, args);
    // Falls through to: record field access → helper method → error
}

Instance Method Examples

Java Call	Dispatch	PIR Output
list.head()	TypeMethodRegistry → ListType.head	wrapDecode(HeadList(list), elemType)
list.map(x -> x+1)	TypeMethodRegistry → ListType.map → PirHofBuilders	LetRec fold with lambda
map.get(key)	TypeMethodRegistry → MapType.get	LetRec pair list search
value.lovelaceOf()	TypeMethodRegistry → “Value.lovelaceOf” (named dispatch)	Nested map lookup
n.abs()	TypeMethodRegistry → IntegerType.abs	IfThenElse(n < 0, 0-n, n)

---

20. The Standard Library

StdlibRegistry

StdlibRegistry (in julc-stdlib) provides PIR term builders for ~65 methods. These are methods that cannot be compiled from Java source because they require low-level PIR/UPLC constructs (lambdas as values, LetRec builders, raw builtin chaining).

Categories:

Category	Examples
Raw builtins	Builtins.headList(), Builtins.iData(), Builtins.sha2_256()
HOF builders	ListsLib.map(), ListsLib.filter(), ListsLib.foldl()
Math delegates	Math.abs(), Math.max(), Math.min()
Factory methods	Optional.of(), Optional.empty(), PubKeyHash.of()

13 On-Chain Libraries

These are @OnchainLibrary-annotated Java classes compiled from source:

Library	Methods	Focus
ListsLib	17+	List operations + HOFs (map, filter, any, all, find, foldl, zip)
MapLib	9	Map operations (lookup, insert, delete, keys, values)
ValuesLib	9	Multi-asset Value operations (add, subtract, compare)
ContextsLib	13	ScriptContext/TxInfo field extraction helpers
OutputLib	8	TxOut querying (outputsAt, lovelacePaidTo, etc.)
MathLib	8	Math utilities (abs, pow, divMod, quotRem, expMod)
IntervalLib	5	Time interval operations
CryptoLib	3	ECDSA, Schnorr, RIPEMD-160
ByteStringLib	8	ByteString manipulation
BitwiseLib	10	Bitwise operations
AddressLib	3	Address/credential utilities
BlsLib	—	BLS12-381 operations (PV11)
NativeValueLib	—	Native Value operations (PV11)

StdlibLookup Chain

When the compiler encounters SomeClass.method(args):

CompositeStdlibLookup
  ├── 1. StdlibRegistry        (builtins, HOFs, math — ~65 methods)
  └── 2. LibraryMethodRegistry  (compiled @OnchainLibrary methods)

First match wins. If neither matches, the compiler falls through to helper method lookup or reports an error.

---

21. Library Compilation

Library methods may depend on each other across files. LibraryCompiler (extracted from JulcCompiler in ADR-018) uses a multi-pass retry strategy:

Pass 1: Try compiling all library CUs
  → LibA succeeds (no dependencies)
  → LibB fails (depends on LibA, not yet available)

Pass 2: Try remaining CUs
  → LibB succeeds (LibA now in registry)

Pass 3: No progress → report errors for any remaining

Topological Sorting

Before wrapping library methods as Let bindings around the validator:

1. Build dependency graph: method A depends on method B if A’s PIR body references B
2. Kahn’s algorithm sorts dependencies first
3. Self-recursive methods (detected via PirHelpers.containsVarRef()) use LetRec instead of Let

---

Level 5: Execution and Testing

22. The Virtual Machine

SPI Architecture

The VM uses Java’s ServiceLoader pattern for pluggable backends:

// SPI interface
public interface JulcVmProvider {
    EvalResult evaluate(Program program, PlutusLanguage language,
                        ExBudget budget, PlutusData... args);
    int priority();  // higher = preferred
}

// Facade (auto-discovers best provider)
JulcVm vm = new JulcVm();
EvalResult result = vm.evaluate(program, scriptContext);

Available Backends

Backend	Module	Priority	Status
Scalus	julc-vm-scalus	50	Default, wraps Scalus 0.16.0 (Scala)
Pure Java	julc-vm-java	—	In development
Truffle	julc-vm-truffle	—	Experimental (GraalVM)

EvalResult

sealed interface EvalResult {
    record Success(Term result, ExBudget budget)
    record Failure(String message)
    record BudgetExhausted(ExBudget consumed)
}

---

23. The Testing Framework

ValidatorTest Base Class

julc-testkit provides ValidatorTest for JUnit-based testing:

class MyValidatorTest extends ValidatorTest {
    @Test
    void testValidator() {
        // Compile
        var result = compile("MyValidator.java source...");
        assertNotNull(result.program());

        // Evaluate
        var evalResult = evaluate(result.program(), testScriptContext);
        assertTrue(evalResult instanceof EvalResult.Success);
    }
}

Test Pyramid

┌─────────────────────────────┐
│     E2E Tests               │  julc-e2e-tests (CIP-113, on-chain deploy)
│     (Yaci DevKit)           │  Requires external devnet
├─────────────────────────────┤
│   Integration Tests         │  julc-compiler tests (compile + evaluate)
│   (Compile → Evaluate)      │  3,442+ tests, 0 failures
├─────────────────────────────┤
│     Unit Tests              │  Individual component tests
│  (TypeResolver, SymbolTable,│  (PirHelpers, UplcOptimizer, etc.)
│   SubsetValidator, etc.)    │
└─────────────────────────────┘

Standard Test Pattern

@Test
void testFeature() {
    var source = """
        @Validator
        public class Test {
            @Entrypoint
            public static boolean validate(PlutusData redeemer, PlutusData ctx) {
                // Test logic that should return true
                return someCondition;
            }
        }
        """;

    // Compile and evaluate in one step
    compileAndAssertTrue(source);
}

Golden File Testing

GoldenUplcTest captures compiled UPLC hex for 8 representative validators. After any refactoring, golden files must match byte-for-byte — ensuring that internal restructuring produces identical output.

---

24. End-to-End: A Validator’s Journey

Let’s trace a simple validator through the entire pipeline:

Source

@SpendingValidator
public class AlwaysSucceeds {
    @Entrypoint
    public static boolean validate(PlutusData redeemer, PlutusData scriptContext) {
        int x = 5;
        int y = x + 3;
        return y == 8;
    }
}

Step 1: Parse → Java AST

JavaParser produces a CompilationUnit with one class declaration containing one method with three statements.

Step 2: Validate

SubsetValidator walks the AST. No rejected constructs.

Step 3: Type Registration

No user-defined types. LedgerTypeRegistry pre-registers ScriptContext, TxInfo, etc.

Step 4: PIR Generation

Lam("redeemer", DataType,
  Lam("scriptContext", DataType,
    Let("x", Const(Integer(5)),
      Let("y", App(App(Builtin(AddInteger), Var("x", IntegerType)), Const(Integer(3))),
        App(App(Builtin(EqualsInteger), Var("y", IntegerType)), Const(Integer(8)))))))

Step 5: Validator Wrapping

Adds ScriptContext decoding and bool→unit/error:

Lam("__scriptContextData", DataType,
  Let("__ctxFields", SndPair(UnConstrData(Var("__scriptContextData"))),
    Let("__redeemer", HeadList(TailList(Var("__ctxFields"))),
      Let("__result",
        App(App(entrypoint, Var("__redeemer")), Var("__scriptContextData")),
        IfThenElse(Var("__result"), Const(Unit), Error)))))

Step 6: UPLC Generation

Let → Apply(Lam(...), ...), variables → De Bruijn indices, IfThenElse → Force/Delay.

Step 7: Optimization

• Constant folding: AddInteger(5, 3) → 8
• Constant folding: EqualsInteger(8, 8) → True
• Beta reduction: inline single-use variables
• Dead code elimination: remove unused bindings
• Force/Delay cancellation: Force(Delay(True)) → True

After optimization, the validator becomes equivalent to \ctx -> Unit (always succeeds).

Step 8: Serialization

Program.plutusV3(optimizedTerm) → FLAT encoding → CBOR wrapping → hex string ready for on-chain submission.

---

Appendices

Appendix A: File Index

julc-compiler (36 files)

Top-level:

File	Lines	Role
JulcCompiler.java	1,221	Main pipeline orchestrator — 24-phase compilation
LibraryCompiler.java	138	Library compilation sub-pipeline
CompileResult.java	—	Compilation result (program + diagnostics + params)
CompilerException.java	—	Fatal compiler error
CompilerOptions.java	—	Compilation options
LibrarySourceResolver.java	—	Classpath scanning + transitive BFS library resolution

pir/ — PIR generation subsystem:

File	Lines	Role
PirGenerator.java	2,147	Core Java AST → PIR transformer
LoopBodyGenerator.java	531	Loop body compilation (5 paths × break/no-break)
AccumulatorTypeAnalyzer.java	432	Accumulator type analysis (pair list detection)
TypeInferenceHelper.java	282	Read-only type inference
TypeMethodRegistry.java	905	Instance method dispatch (~50 methods across 11 types)
PirHelpers.java	356	wrapDecode/wrapEncode + list utilities
PirHofBuilders.java	244	HOF PIR builders (map, filter, any, all, find, foldl, zip)
PirTerm.java	—	PIR term AST (12 variants)
PirType.java	—	PIR type system (13+ variants)
PirFormatter.java	—	PIR pretty-printing
PirSubstitution.java	—	PIR variable substitution
StdlibLookup.java	—	Functional interface for stdlib resolution
CompositeStdlibLookup.java	—	Chains multiple StdlibLookup instances

resolve/ — Type resolution:

File	Role
TypeResolver.java	Java → PIR type mapping
TypeRegistrar.java	Topological type registration (Kahn’s algorithm)
SymbolTable.java	Scope stack for variable/method management
LedgerSourceLoader.java	Dynamic ledger type loading from META-INF
LibraryMethodRegistry.java	Compiled library method storage + typed coercion
ImportResolver.java	Import resolution

Other packages:

File	Role
codegen/ValidatorWrapper.java	ScriptContext decoding + bool→unit/error wrapping
codegen/DataCodecGenerator.java	Data codec generation
desugar/LoopDesugarer.java	For-each/while → LetRec transformation
desugar/PatternMatchDesugarer.java	Switch/instanceof → DataMatch transformation
error/CompilerDiagnostic.java	Diagnostic record (level, message, location)
error/DiagnosticCollector.java	Structured error collection
uplc/UplcGenerator.java	PIR → UPLC lowering
uplc/UplcOptimizer.java	6-pass UPLC optimizer with fixpoint iteration
validate/SubsetValidator.java	Java subset enforcement
util/MethodDependencyResolver.java	Method dependency graph construction
util/StringUtils.java	String utilities (Levenshtein distance)

Other Key Files

File	Role
julc-core/.../Term.java	UPLC term AST (10 variants)
julc-core/.../DefaultFun.java	102 Plutus builtin functions
julc-core/.../PlutusData.java	Universal on-chain data encoding
julc-stdlib/.../StdlibRegistry.java	PIR term builders for ~65 stdlib methods
julc-vm/.../JulcVmProvider.java	VM SPI interface
julc-testkit/.../ValidatorTest.java	Testing base class

---

Appendix B: Glossary

Term	Definition
UPLC	Untyped Plutus Lambda Calculus — the on-chain execution language
PIR	Plutus Intermediate Representation — typed bridge between Java and UPLC
De Bruijn index	Variable indexing where variables reference lambda binders by distance (1 = innermost)
Data	Universal on-chain value encoding (5 constructors: Constr, Map, List, I, B)
CEK machine	Count-Evaluate-Kont — the Plutus virtual machine
ScriptContext	Ledger data passed to every validator (TxInfo + redeemer + script info)
eUTxO	Extended Unspent Transaction Output — Cardano’s accounting model
Force/Delay	UPLC constructs for lazy evaluation (needed because UPLC is strict/call-by-value)
Z-combinator	Strict fixed-point combinator enabling recursion in UPLC
wrapDecode	PirHelpers method to extract a typed value from raw Data
wrapEncode	PirHelpers method to wrap a typed value back into Data
SumType	Tagged union — a sealed interface with record variants
RecordType	Product type — a record with named typed fields
LetRec	Recursive let binding — used for loops and self-recursive functions
Accumulator	Variable modified across loop iterations (packed into Data tuples for multi-acc)
Kahn’s algorithm	Topological sort used for type registration and library ordering
StdlibLookup	Interface for resolving static method calls to PIR terms
TypeMethodRegistry	Registry mapping (PirType, method) pairs to instance method handlers
@OnchainLibrary	Annotation marking a class as a reusable on-chain library
@Param	Annotation marking a validator field as a deployment-time parameter
@Entrypoint	Annotation marking the main validator method
FLAT	Binary encoding format for UPLC programs on-chain
SOPs	Sums of Products — Plutus V3 constructor/case terms
Golden test	Test that compares output byte-for-byte against a saved reference file
Conway era	Current Cardano era with governance features (CIP-1694)