[lldb] Add a return opcode to the formatter bytecode

[adrian-prantl]

In LLVM we love our early exists and this opcode allows for simpler code generation.

[llvmbot]

@llvm/pr-subscribers-lldb

Author: Adrian Prantl (adrian-prantl)

Changes

In LLVM we love our early exists and this opcode allows for simpler code generation.

---

Full diff: https://github.com/llvm/llvm-project/pull/121602.diff

5 Files Affected:

• (modified) lldb/docs/resources/formatterbytecode.rst (+1)
• (modified) lldb/examples/python/formatter_bytecode.py (+4)
• (modified) lldb/source/DataFormatters/FormatterBytecode.cpp (+3)
• (modified) lldb/source/DataFormatters/FormatterBytecode.def (+1)
• (modified) lldb/unittests/DataFormatter/FormatterBytecodeTest.cpp (+10)

diff --git a/lldb/docs/resources/formatterbytecode.rst b/lldb/docs/resources/formatterbytecode.rst
index 20e148363ef957..34fb0f7ee924c4 100644
--- a/lldb/docs/resources/formatterbytecode.rst
+++ b/lldb/docs/resources/formatterbytecode.rst
@@ -75,6 +75,7 @@ These manipulate the control stack and program counter. Both `if` and `ifelse` e
  0x12      `ifelse`    `(UInt -> )` pop two blocks from the control stack, if
                        the top of the data stack is nonzero, execute the first,
                        otherwise the second.
+ 0x13      `return`    pop the entire control stack and return
 ========  ==========  ============================================================
 
 Literals for basic types
diff --git a/lldb/examples/python/formatter_bytecode.py b/lldb/examples/python/formatter_bytecode.py
index ccd0c68a75483c..36a14be283f319 100644
--- a/lldb/examples/python/formatter_bytecode.py
+++ b/lldb/examples/python/formatter_bytecode.py
@@ -35,6 +35,7 @@ def define_opcode(n, mnemonic, name):
 define_opcode(0x10, "{", "begin")
 define_opcode(0x11, "if", "if")
 define_opcode(0x12, "ifelse", "ifelse")
+define_opcode(0x13, "return", "return")
 
 define_opcode(0x20, None, "lit_uint")
 define_opcode(0x21, None, "lit_int")
@@ -342,6 +343,9 @@ def next_byte():
             else:
                 frame.append(control.pop())
                 control.pop()
+        elif b == op_return:
+            control.clear()
+            return data[-1]
 
         # Literals.
         elif b == op_lit_uint:
diff --git a/lldb/source/DataFormatters/FormatterBytecode.cpp b/lldb/source/DataFormatters/FormatterBytecode.cpp
index e49c7506781875..7f3dbe0dba37d6 100644
--- a/lldb/source/DataFormatters/FormatterBytecode.cpp
+++ b/lldb/source/DataFormatters/FormatterBytecode.cpp
@@ -304,6 +304,9 @@ llvm::Error Interpret(std::vector<ControlStackElement> &control,
       control.pop_back();
       activate_block();
       continue;
+    case op_return:
+      control.clear();
+      return pc.takeError();
 
     // Literals.
     case op_lit_uint:
diff --git a/lldb/source/DataFormatters/FormatterBytecode.def b/lldb/source/DataFormatters/FormatterBytecode.def
index c6645631fa0065..29e0bee541c73c 100644
--- a/lldb/source/DataFormatters/FormatterBytecode.def
+++ b/lldb/source/DataFormatters/FormatterBytecode.def
@@ -27,6 +27,7 @@ DEFINE_OPCODE(0x06, "rot",  rot)
 DEFINE_OPCODE(0x10, "{", begin)
 DEFINE_OPCODE(0x11, "if", if)
 DEFINE_OPCODE(0x12, "ifelse", ifelse)
+DEFINE_OPCODE(0x13, "return", return)
 
 DEFINE_OPCODE(0x20, nullptr, lit_uint)
 DEFINE_OPCODE(0x21, nullptr, lit_int)
diff --git a/lldb/unittests/DataFormatter/FormatterBytecodeTest.cpp b/lldb/unittests/DataFormatter/FormatterBytecodeTest.cpp
index 7307db650c1629..5e980c3e1913cf 100644
--- a/lldb/unittests/DataFormatter/FormatterBytecodeTest.cpp
+++ b/lldb/unittests/DataFormatter/FormatterBytecodeTest.cpp
@@ -97,6 +97,16 @@ TEST_F(FormatterBytecodeTest, ControlOps) {
                           data));
     ASSERT_EQ(data.Pop<uint64_t>(), 42u);
   }
+  {
+    DataStack data;
+    ASSERT_TRUE(Interpret({op_lit_uint, 1, op_begin, 3, op_lit_uint, 42,
+                           op_return, op_if, op_lit_uint, 23},
+                          data));
+    ASSERT_EQ(data.Pop<uint64_t>(), 42u);
+  }
+}
+
+TEST_F(FormatterBytecodeTest, ConversionOps) {
   {
     DataStack data(lldb::ValueObjectSP{});
     ASSERT_TRUE(Interpret({op_is_null}, data));

[kastiglione]

Nested conditionals is where this will be helpful:

Consider this python:

if first == 1:
    if second == 2:
        return "thing"
return "other"

without a return op, the above code would have to be transformed to:

if first == 1:
    if second == 2:
        result = "thing"
if not (first == 1 and second == 2):
    result = "other"

the bytecode for this this would look something like:

… pick # get `first` from the stack
dup 1 ==
{
    … pick # get `second` from the stack
    dup 2 ==
    {
        "thing"
    } if
} if
& ~ # not (first == 1 and second == 2)
{
    "other"
} if

with a return op, it simplifies to:

… pick # get `first` from the stack
1 ==
{
    … pick # get `second` from the stack
    2 ==
    {
        "thing"
        return
    } if
} if
"other"