js-confuser-vm 0.1.0 → 0.1.2

package/src/template.ts

@@ -1,141 +0,0 @@
-// Template
-// Compiles a JS code snippet into raw IR bytecode that can be spliced into the
-// parent compiler's bytecode stream at any point before resolveRegisters /
-// resolveLabels run.
-//
-// ── Usage ─────────────────────────────────────────────────────────────────────
-//
-//   const tmpl = new Template(`
-//     function {name}(x, y) {
-//       return x + y;
-//     }
-//   `);
-//
-//   const bc = tmpl.compile({ name: "myHelper" }, parentCompiler);
-//   result.push(...bc);
-//
-// ── How it works ──────────────────────────────────────────────────────────────
-//
-// 1. {name} placeholders are replaced with the caller-supplied string values.
-// 2. A fresh child Compiler is created, inheriting the parent's OP table so
-//    opcode numbers match exactly (including randomizeOpcodes mappings).
-// 3. The child compiles the snippet to raw IR (no passes, no label/register
-//    resolution).
-// 4. Post-processing makes the child's bytecode compatible with the parent:
-//
-//    Labels      — every label string is renamed via parentCompiler._makeLabel()
-//                  so names never collide with existing or future labels.
-//
-//    FnIds       — the child's main scope (fnDescriptors[0]) is mapped to
-//                  targetFnId (default 0).  Any inner functions (closures
-//                  declared inside the template) are appended to
-//                  parentCompiler.fnDescriptors with fresh indices.
-//
-// 5. The main function's entry defineLabel is stripped from the output — it is
-//    a synthetic wrapper added by _compileMain and is not part of the injected
-//    code.  All other instructions (including the implicit RETURN at the end of
-//    the main scope and any inner-function blocks) are returned as-is so the
-//    caller can append them wherever appropriate.
-//
-// ── Limitations (MVP) ─────────────────────────────────────────────────────────
-// • Variables are plain string/number interpolation only — no AST-node
-//   substitution.
-// • Templates that reference upvalue-captured registers from the call site are
-//   not supported (inner functions closing over template-local variables work).
-// • Opcodes with no JS equivalent (JUMP_REG, BXOR used as decode, etc.) cannot
-//   be expressed in a template; write those instruction arrays manually.
-
-import { Compiler } from "./compiler.ts";
-import { DEFAULT_OPTIONS } from "./options.ts";
-import type { Bytecode, Instruction } from "./types.ts";
-
-export class Template {
-  private readonly _source: string;
-
-  constructor(source: string) {
-    this._source = source;
-  }
-
-  // ── String interpolation ──────────────────────────────────────────────────
-  private _interpolate(variables: Record<string, string | number>): string {
-    return this._source.replace(/\{(\w+)\}/g, (match, name) => {
-      if (!(name in variables)) {
-        throw new Error(`Template: missing variable {${name}}`);
-      }
-      return String(variables[name]);
-    });
-  }
-
-  // ── Main entry point ───────────────────────────────────────────────────────
-  /**
-   * Compile the template and return the inner (non-main) function descriptors
-   * and their bytecode blocks, ready to splice into the parent compiler's
-   * instruction stream.
-   *
-   * The template source should declare one or more named functions.  The
-   * top-level ("main") scope of the template is discarded — it exists only as
-   * a syntactic wrapper so that function declarations parse correctly.
-   *
-   * Each inner function is registered in parentCompiler.fnDescriptors with a
-   * fresh fnIdx, and its bytecode block (defineLabel + body instructions) is
-   * returned so the caller can append it to the parent bytecode stream at the
-   * desired location (typically at the end, after all function bodies).
-   *
-   * @param variables       Substitution map for {name} placeholders.
-   * @param parentCompiler  The Compiler whose OP table, label counter, and
-   *                        fnDescriptors are shared.
-   *
-   * @returns
-   *   functions  — ordered list of inner FnDescriptors (index 0 = first named
-   *                function in the template source).  Use .entryLabel and
-   *                ._fnIdx to build MAKE_CLOSURE operands.
-   *   bytecode   — IR bytecode blocks for all inner functions, ready to splice
-   *                after the parent's function bodies.  Does NOT include the
-   *                template's main-scope instructions.
-   */
-  compile(
-    variables: Record<string, string | number>,
-    parentCompiler: Compiler,
-  ): { functions: any[]; bytecode: Bytecode } {
-    // ── 1. Interpolate ────────────────────────────────────────────────────
-    const code = this._interpolate(variables);
-
-    // ── 2. Create child compiler, inherit parent's OP table ───────────────
-    // randomizeOpcodes is disabled — we copy the parent's already-randomized
-    // mapping directly so all opcode numbers are identical.
-    const child = new Compiler({ ...DEFAULT_OPTIONS, randomizeOpcodes: false });
-    child.OP = { ...parentCompiler.OP };
-    child.OP_NAME = { ...parentCompiler.OP_NAME };
-    child.JUMP_OPS = new Set(parentCompiler.JUMP_OPS);
-
-    child._makeLabel = parentCompiler._makeLabel.bind(parentCompiler);
-
-    // Record how many descriptors the parent already has so we can find the
-    // child's main (index = startIdx) and inner functions (startIdx+1 …).
-    const startIdx = parentCompiler.fnDescriptors.length;
-    child.fnDescriptors = parentCompiler.fnDescriptors; // share — inner functions auto-register
-
-    // ── 3. Compile to raw IR (no passes) ──────────────────────────────────
-    child.compile(code);
-
-    // parentCompiler.fnDescriptors[startIdx]   → child's main (discard)
-    // parentCompiler.fnDescriptors[startIdx+1…] → inner helper functions
-    const innerDescs = parentCompiler.fnDescriptors.slice(startIdx + 1);
-
-    // Build bytecode blocks for inner functions only.
-    // child.bytecode was assembled by _compileMain from ALL fnDescriptors
-    // starting at startIdx.  We rebuild it here from the inner descs only.
-    const innerBytecode: Bytecode = [];
-    for (const desc of innerDescs) {
-      innerBytecode.push([
-        null,
-        { type: "defineLabel", label: desc.entryLabel },
-      ] as Instruction);
-      for (const instr of (desc as any).bytecode as Bytecode) {
-        innerBytecode.push(instr);
-      }
-    }
-
-    return { functions: innerDescs, bytecode: innerBytecode };
-  }
-}

package/src/transforms/bytecode/aliasedOpcodes.ts

@@ -1,148 +0,0 @@
-import type { Bytecode, InstrOperand, Instruction } from "../../types.ts";
-import { Compiler, SOURCE_NODE_SYM } from "../../compiler.ts";
-import { nextFreeSlot } from "../../utils/op-utils.ts";
-import { shuffle } from "../../utils/random-utils.ts";
-
-// Opcodes that must not be aliased.
-// Variable-length operand opcodes cannot be statically aliased since the
-// number of this._operand() calls varies at runtime.
-// Infrastructure opcodes (PATCH, TRY_SETUP, TRY_END, DEBUGGER) are excluded
-// because aliasing them would interfere with self-modifying bytecode and
-// exception-handling machinery.
-const DISALLOWED_OP_NAMES = new Set([
-  "MAKE_CLOSURE",
-  "BUILD_ARRAY",
-  "BUILD_OBJECT",
-  "CALL",
-  "CALL_METHOD",
-  "NEW",
-  "PATCH",
-  "TRY_SETUP",
-  "TRY_END",
-  "DEBUGGER",
-]);
-
-// Creates aliased opcodes: duplicate handlers for commonly-used opcodes,
-// optionally with a permuted operand read order in the bytecode stream.
-//
-// For each aliased op, we record an `order` permutation of length `arity`.
-// order[i] = j means: bytecode slot i holds what was originally operand j.
-//
-// Example: LOAD_GLOBAL [dst, nameIdx] with order=[1,0]:
-//   Bytecode stores:  [ALIAS_OP, nameIdx, dst]
-//   Handler reads:    _unsortedOperands = [nameIdx, dst]
-//                     _operands = [_unsortedOperands[1], _unsortedOperands[0]]
-//                               = [dst, nameIdx]   ← original order restored
-//
-// Runs LAST among bytecode transforms (after selfModifying), before resolveLabels.
-export function aliasedOpcodes(
-  bc: Bytecode,
-  compiler: Compiler,
-): { bytecode: Bytecode } {
-  // Build a map of base opcode value → name, excluding disallowed ops
-  const baseOpValueToName = new Map<number, string>();
-  for (const [name, val] of Object.entries(compiler.OP)) {
-    if (DISALLOWED_OP_NAMES.has(name)) continue;
-    baseOpValueToName.set(val as number, name);
-  }
-
-  // ── Step 1: count frequency and determine arity for each eligible base opcode ─
-  // We scan the actual post-transform bytecode so frequency reflects what's
-  // really left (specialized/macro ops already consumed their share).
-  const opStats = new Map<number, { freq: number; arity: number | null }>();
-
-  for (const instr of bc) {
-    const op = instr[0];
-    if (op === null || !baseOpValueToName.has(op)) continue;
-
-    const arity = instr.length - 1;
-    if (arity < 1) continue; // 0-operand opcodes have nothing to permute
-
-    const existing = opStats.get(op);
-    if (!existing) {
-      opStats.set(op, { freq: 1, arity });
-    } else {
-      if (existing.arity !== arity) {
-        // Inconsistent arity → variable-length; skip
-        existing.arity = null;
-      }
-      existing.freq++;
-    }
-  }
-
-  // ── Step 2: sort by frequency descending, keep only consistent-arity ops ────
-  const candidates = Array.from(opStats.entries())
-    .filter(([, s]) => s.arity !== null)
-    .sort(([, a], [, b]) => b.freq - a.freq);
-
-  if (candidates.length === 0) return { bytecode: bc };
-
-  // ── Step 3: assign free slots, build order permutations ─────────────────────
-  // aliasMap: originalOp → aliasOp (only the winning alias per original op)
-  const aliasMap = new Map<number, number>();
-  const aliasedOps: Compiler["ALIASED_OPS"] = {};
-
-  for (const [originalOp, stats] of candidates) {
-    const aliasOp = nextFreeSlot(compiler);
-    if (aliasOp === -1) break;
-
-    const arity = stats.arity!;
-
-    // Build a permutation of [0 .. arity-1].
-    // For arity >= 2: shuffle until we get a non-identity permutation so the
-    // operand order is actually different (makes the alias more confusing).
-    // For arity == 1: only one permutation exists ([0]); still useful as a clone.
-    let order: number[];
-    if (arity >= 2) {
-      const identity = Array.from({ length: arity }, (_, i) => i);
-      let attempts = 0;
-      do {
-        order = shuffle([...identity]);
-        attempts++;
-      } while (attempts < 20 && order.every((v, i) => v === i));
-    } else {
-      order = [0];
-    }
-
-    aliasMap.set(originalOp, aliasOp);
-    aliasedOps[aliasOp] = { originalOp, order };
-
-    const originalName = compiler.OP_NAME[originalOp] ?? `OP_${originalOp}`;
-    compiler.OP_NAME[aliasOp] = `ALIAS_${originalName}_${order.join("_")}`;
-  }
-
-  compiler.ALIASED_OPS = aliasedOps;
-
-  if (aliasMap.size === 0) return { bytecode: bc };
-
-  // ── Step 4: rewrite bytecode ─────────────────────────────────────────────────
-  const result: Bytecode = [];
-
-  for (const instr of bc) {
-    const op = instr[0];
-    if (op === null || !aliasMap.has(op)) {
-      result.push(instr);
-      continue;
-    }
-
-    const aliasOp = aliasMap.get(op)!;
-    const { order } = aliasedOps[aliasOp];
-    const originalOperands = instr.slice(1) as InstrOperand[];
-
-    // Guard: if arity changed (shouldn't happen after the consistency check),
-    // fall back to the original instruction.
-    if (originalOperands.length !== order.length) {
-      result.push(instr);
-      continue;
-    }
-
-    // Rearrange operands: new slot i receives original operand order[i].
-    const newOperands = order.map((i) => originalOperands[i]);
-
-    const newInstr: Instruction = [aliasOp, ...newOperands];
-    (newInstr as any)[SOURCE_NODE_SYM] = (instr as any)[SOURCE_NODE_SYM];
-    result.push(newInstr);
-  }
-
-  return { bytecode: result };
-}

package/src/transforms/bytecode/concealConstants.ts

@@ -1,52 +0,0 @@
-import { Compiler } from "../../compiler.ts";
-import type * as b from "../../types.ts";
-
-export function concealConstants(
-  bytecode: b.Bytecode,
-  compiler: Compiler,
-): {
-  bytecode: b.Bytecode;
-} {
-  const newBytecode: b.Bytecode = [];
-
-  for (const instr of bytecode) {
-    const [op, ...operands] = instr;
-
-    const hasContant = operands.some(
-      (o) =>
-        o !== undefined &&
-        o !== null &&
-        typeof o === "object" &&
-        (o as any).type === "constant",
-    );
-
-    if (!hasContant) {
-      newBytecode.push(instr);
-      continue;
-    }
-
-    const newOperands = [];
-    for (const operand of operands) {
-      if ((operand as any)?.type === "constant") {
-        const tsOperand = operand as any;
-        newOperands.push(operand);
-        newOperands.push({
-          type: "constant",
-          value: tsOperand.value,
-          key: true,
-        });
-      } else {
-        newOperands.push(operand);
-      }
-    }
-
-    instr.length = 0;
-    instr.push(op, ...newOperands);
-
-    newBytecode.push(instr);
-  }
-
-  return {
-    bytecode: newBytecode,
-  };
-}