Switch to lief for binary parsing, add OSX support.

.gitignore

@@ -17,3 +17,7 @@ tests/*/*.log
 
 # GDB
 .gdb_history
+.idea
+.venv
+pyproject.toml
+uv.lock

.python-version

@@ -0,0 +1 @@
+3.12

README.md

@@ -147,7 +147,6 @@ The tool does not transparently support all features of POSIX shared libraries.
 The tool also lacks the following important features:
 * symbol versions are not handled at all
 * keep fast paths of shims together to reduce I$ pressure
-* support for macOS
 
 Finally, there are some minor TODOs in code.
 

arch/osx/aarch64/config.ini

@@ -0,0 +1,3 @@
+[Arch]
+PointerSize = 8
+SymbolReloc = R_AARCH64_ABS64, SYMBOLIC, RELATIVE

arch/osx/aarch64/table.S.tpl

@@ -0,0 +1,67 @@
+/*
+ * Copyright 2018-2025 Yury Gribov
+ *
+ * The MIT License (MIT)
+ *
+ * Use of this source code is governed by MIT license that can be
+ * found in the LICENSE.txt file.
+ */
+
+#define lr x30
+#define ip0 x16
+
+.data
+  .globl __${lib_suffix}_tramp_table
+  .private_extern __${lib_suffix}_tramp_table
+  .p2align 3
+__${lib_suffix}_tramp_table:
+  .zero $table_size
+
+  .text
+
+  .globl __${lib_suffix}_tramp_resolve
+  .private_extern __${lib_suffix}_tramp_resolve
+
+  .globl __${lib_suffix}_save_regs_and_resolve
+  .private_extern __${lib_suffix}_save_regs_and_resolve
+__${lib_suffix}_save_regs_and_resolve:
+  .cfi_startproc
+
+#define PUSH_PAIR(reg1, reg2) stp reg1, reg2, [sp, #-16]!; .cfi_adjust_cfa_offset 16; .cfi_rel_offset reg1, 0; .cfi_rel_offset reg2, 8
+#define POP_PAIR(reg1, reg2) ldp reg1, reg2, [sp], #16; .cfi_adjust_cfa_offset -16; .cfi_restore reg2; .cfi_restore reg1
+
+#define PUSH_WIDE_PAIR(reg1, reg2) stp reg1, reg2, [sp, #-32]!; .cfi_adjust_cfa_offset 32; .cfi_rel_offset reg1, 0; .cfi_rel_offset reg2, 16
+#define POP_WIDE_PAIR(reg1, reg2) ldp reg1, reg2, [sp], #32; .cfi_adjust_cfa_offset -32; .cfi_restore reg2; .cfi_restore reg1
+
+  // Save only arguments (and lr)
+  PUSH_PAIR(x0, x1)
+  PUSH_PAIR(x2, x3)
+  PUSH_PAIR(x4, x5)
+  PUSH_PAIR(x6, x7)
+  PUSH_PAIR(x8, lr)
+
+  ldr x0, [sp, #80]  // 16*5
+
+  PUSH_WIDE_PAIR(q0, q1)
+  PUSH_WIDE_PAIR(q2, q3)
+  PUSH_WIDE_PAIR(q4, q5)
+  PUSH_WIDE_PAIR(q6, q7)
+
+  // Stack is aligned at 16 bytes
+  bl __${lib_suffix}_tramp_resolve
+  mov ip0, x0
+
+  POP_WIDE_PAIR(q6, q7)
+  POP_WIDE_PAIR(q4, q5)
+  POP_WIDE_PAIR(q2, q3)
+  POP_WIDE_PAIR(q0, q1)
+
+  POP_PAIR(x8, lr)
+  POP_PAIR(x6, x7)
+  POP_PAIR(x4, x5)
+  POP_PAIR(x2, x3)
+  POP_PAIR(x0, x1)
+
+  br lr
+
+  .cfi_endproc

arch/osx/aarch64/trampoline.S.tpl

@@ -0,0 +1,42 @@
+/*
+ * Copyright 2018-2025 Yury Gribov
+ *
+ * The MIT License (MIT)
+ *
+ * Use of this source code is governed by MIT license that can be
+ * found in the LICENSE.txt file.
+ */
+
+  .globl _$sym
+  .p2align 4
+#ifndef IMPLIB_EXPORT_SHIMS
+  .private_extern _$sym
+#endif
+_$sym:
+  .cfi_startproc
+
+1:
+  // Load address via Mach-O PAGE/PAGEOFF directives
+  adrp ip0, (__${lib_suffix}_tramp_table + $offset)@PAGE
+  ldr ip0, [ip0, (__${lib_suffix}_tramp_table + $offset)@PAGEOFF]
+
+  cbz ip0, 2f
+
+  // Fast path
+  br ip0
+
+2:
+  // Slow path
+  mov ip0, #($number & 0xffff)
+#if $number > 0xffff
+  movk ip0, #($number >> 16), lsl #16
+#endif
+  stp ip0, lr, [sp, #-16]!
+  .cfi_adjust_cfa_offset 16
+  .cfi_rel_offset lr, 8
+  bl __${lib_suffix}_save_regs_and_resolve
+  ldp xzr, lr, [sp], #16
+  .cfi_adjust_cfa_offset -16
+  .cfi_restore lr
+  br ip0
+  .cfi_endproc