[Tarantool-patches] [PATCH luajit v3] dbg: remove wrong GC state

[Mikhail Elhimov]

It was copied originally by mistake from the LuaVela fork of LuaJIT.
---
Changes in v3:
  * Added comment about reachability of all the GC states from the same lua script

Changes in v2:
  * Rebased onto https://lists.tarantool.org/pipermail/tarantool-patches/2026-October/030859.html to avoid conflicts

This patch is to be applied after https://lists.tarantool.org/pipermail/tarantool-patches/2026-October/030859.html

Branch: https://github.com/tarantool/luajit/tree/elhimov/gh-noticket-dbg-remove-wrong-gcstate

 src/luajit_dbg.py                             |   3 +-
 .../debug-extension-tests.py                  | 113 ++++++++++++++++--
 2 files changed, 104 insertions(+), 12 deletions(-)

diff --git a/src/luajit_dbg.py b/src/luajit_dbg.py
index a685ecfa..8f7249ab 100644
--- a/src/luajit_dbg.py
+++ b/src/luajit_dbg.py
@@ -1244,7 +1244,6 @@ def gc_state(g):
         3: 'SWEEPSTRING',
         4: 'SWEEP',
         5: 'FINALIZE',
-        6: 'LAST',
     }.get(int(g['gc']['state']), 'INVALID')
 
 
@@ -2850,7 +2849,7 @@ LJState.__doc__ = '''
 lj-state
 The command requires no args and dumps current VM and GC states:
 * VM state: <{}>
-* GC state: <PAUSE|PROPAGATE|ATOMIC|SWEEPSTRING|SWEEP|FINALIZE|LAST>
+* GC state: <PAUSE|PROPAGATE|ATOMIC|SWEEPSTRING|SWEEP|FINALIZE>
 * JIT state: <IDLE|ACTIVE|RECORD|START|END|ASM|ERR>
     '''.format('|'.join(VM_STATES))
 
diff --git a/test/tarantool-debugger-tests/debug-extension-tests.py b/test/tarantool-debugger-tests/debug-extension-tests.py
index bbd53dcc..9b6484ca 100644
--- a/test/tarantool-debugger-tests/debug-extension-tests.py
+++ b/test/tarantool-debugger-tests/debug-extension-tests.py
@@ -133,6 +133,23 @@ elif machine == 'arm64' or machine == 'aarch64':
 else:
     raise Exception('Unknown architecture in testing')
 
+GCS_PAUSE = 'PAUSE'
+GCS_PROPAGATE = 'PROPAGATE'
+GCS_ATOMIC = 'ATOMIC'
+GCS_SWEEPSTRING = 'SWEEPSTRING'
+GCS_SWEEP = 'SWEEP'
+GCS_FINALIZE = 'FINALIZE'
+
+RX_VALID_GC_STATES = r'({})'.format('|'.join([
+    GCS_PAUSE,
+    GCS_PROPAGATE,
+    GCS_ATOMIC,
+    GCS_SWEEPSTRING,
+    GCS_SWEEP,
+    GCS_FINALIZE,
+]))
+RX_INVALID_GC_STATE = r'INVALID'
+
 
 class TestCaseBase(unittest.TestCase):
     @classmethod
@@ -283,23 +300,24 @@ RX_VM_STATE = '({})'.format('|'.join([
 ]))
 
 
-class TestLJState(TestCaseBase):
-    extension_cmds = 'lj-state'
-    location = 'lj_cf_print'
-    lua_script = 'print(1)'
-    pattern = (
+def compose_lj_state_rx(rx_gc_state):
+    return (
         r'VM state: ' + RX_VM_STATE + '\n'
-        r'GC state: [A-Z]+\n'
+        r'GC state: ' + rx_gc_state + '\n'
         r'JIT state: [A-Z]+\n'
     )
 
 
-class TestLJGC(TestCaseBase):
-    extension_cmds = 'lj-gc'
+class TestLJState(TestCaseBase):
+    extension_cmds = 'lj-state'
     location = 'lj_cf_print'
     lua_script = 'print(1)'
-    pattern = (
-        r'GC stats: [A-Z]+\n'
+    pattern = compose_lj_state_rx(RX_VALID_GC_STATES)
+
+
+def compose_gc_stats_rx(rx_gc_state):
+    return (
+        r'GC stats: ' + rx_gc_state + '\n'
         r'\ttotal: \d+\n'
         r'\tthreshold: \d+\n'
         r'\tdebt: \d+\n'
@@ -315,6 +333,81 @@ class TestLJGC(TestCaseBase):
     )
 
 
+class TestLJGC(TestCaseBase):
+    extension_cmds = 'lj-gc'
+    location = 'lj_cf_print'
+    lua_script = 'print(1)'
+    pattern = compose_gc_stats_rx(RX_VALID_GC_STATES)
+
+
+class TestGCStateBase(TestCaseBase):
+    extension_cmds = (
+        'lj-state\n'
+        'lj-gc\n'
+    )
+
+    # The same lua code is used to check all the GC states because
+    # it allows to reach any of them:
+    # 1) `gc_mark_start()` (PAUSE) is always present in the full
+    #    GC cycle.
+    # 2) `propagatemark()` (PROPAGATE) is reachable because our
+    #    build has FFI and the finalizer table is marked as a weak
+    #    table, i.e. it is gray. Hence, the g->gc.gray list is not
+    #    empty.
+    # 3) `atomic()` (ATOMIC) is always present in the full GC
+    #    cycle.
+    # 4) `gc_sweep_str_chain()` (SWEEPSTRING) is always present in
+    #    the full GC cycle.
+    # 5) `gc_sweep()` (SWEEP) is always present in the full GC
+    #    cycle.
+    # 6) `gc_finalize()` (FINALIZE) is met due to the finalizer to
+    #    userdata in the script.
+    lua_script = (
+        'local obj = newproxy(true)\n'
+        'getmetatable(obj).__gc = function() end\n'
+        'obj = nil\n'
+        'collectgarbage()\n'
+    )
+
+    @classmethod
+    def setUpClass(cls):
+        cls.pattern = (
+            compose_lj_state_rx(cls.gc_state) +
+            compose_gc_stats_rx(cls.gc_state)
+        )
+        super(TestGCStateBase, cls).setUpClass()
+
+
+class TestGCStatePause(TestGCStateBase):
+    gc_state = GCS_PAUSE
+    location = 'gc_mark_start'
+
+
+class TestGCStatePropagate(TestGCStateBase):
+    gc_state = GCS_PROPAGATE
+    location = 'propagatemark'
+
+
+class TestGCStateAtomic(TestGCStateBase):
+    gc_state = GCS_ATOMIC
+    location = 'atomic'
+
+
+class TestGCStateSweepString(TestGCStateBase):
+    gc_state = GCS_SWEEPSTRING
+    location = 'gc_sweep_str_chain'
+
+
+class TestGCStateSweep(TestGCStateBase):
+    gc_state = GCS_SWEEP
+    location = 'gc_sweep'
+
+
+class TestGCStateFinalize(TestGCStateBase):
+    gc_state = GCS_FINALIZE
+    location = 'gc_finalize'
+
+
 STACK_RX = (
     r'-+ Red zone:\s+\d+ slots -+\n'
     r'(' + RX_ADDR + r'\s+' + RX_FRAME + r' VALUE: nil\n?)*\n'
-- 
2.43.0