dictobject.c

/* Dictionary object implementation using a hash table */

/* The distribution includes a separate file, Objects/dictnotes.txt,
   describing explorations into dictionary design and optimization.
   It covers typical dictionary use patterns, the parameters for
   tuning dictionaries, and several ideas for possible optimizations.
*/

/* PyDictKeysObject

This implements the dictionary's hashtable.

As of Python 3.6, this is compact and ordered. Basic idea is described here:
* https://mail.python.org/pipermail/python-dev/2012-December/123028.html
* https://morepypy.blogspot.com/2015/01/faster-more-memory-efficient-and-more.html

layout:

+---------------------+
| dk_refcnt           |
| dk_log2_size        |
| dk_log2_index_bytes |
| dk_kind             |
| dk_version          |
| dk_usable           |
| dk_nentries         |
+---------------------+
| dk_indices[]        |
|                     |
+---------------------+
| dk_entries[]        |
|                     |
+---------------------+

dk_indices is actual hashtable.  It holds index in entries, or DKIX_EMPTY(-1)
or DKIX_DUMMY(-2).
Size of indices is dk_size.  Type of each index in indices varies with dk_size:

* int8  for          dk_size <= 128
* int16 for 256   <= dk_size <= 2**15
* int32 for 2**16 <= dk_size <= 2**31
* int64 for 2**32 <= dk_size

dk_entries is array of PyDictKeyEntry when dk_kind == DICT_KEYS_GENERAL or
PyDictUnicodeEntry otherwise. Its length is USABLE_FRACTION(dk_size).

NOTE: Since negative value is used for DKIX_EMPTY and DKIX_DUMMY, type of
dk_indices entry is signed integer and int16 is used for table which
dk_size == 256.
*/


/*
The DictObject can be in one of two forms.

Either:
  A combined table:
    ma_values == NULL, dk_refcnt == 1.
    Values are stored in the me_value field of the PyDictKeyEntry.
Or:
  A split table:
    ma_values != NULL, dk_refcnt >= 1
    Values are stored in the ma_values array.
    Only string (unicode) keys are allowed.

There are four kinds of slots in the table (slot is index, and
DK_ENTRIES(keys)[index] if index >= 0):

1. Unused.  index == DKIX_EMPTY
   Does not hold an active (key, value) pair now and never did.  Unused can
   transition to Active upon key insertion.  This is each slot's initial state.

2. Active.  index >= 0, me_key != NULL and me_value != NULL
   Holds an active (key, value) pair.  Active can transition to Dummy or
   Pending upon key deletion (for combined and split tables respectively).
   This is the only case in which me_value != NULL.

3. Dummy.  index == DKIX_DUMMY  (combined only)
   Previously held an active (key, value) pair, but that was deleted and an
   active pair has not yet overwritten the slot.  Dummy can transition to
   Active upon key insertion.  Dummy slots cannot be made Unused again
   else the probe sequence in case of collision would have no way to know
   they were once active.
   In free-threaded builds dummy slots are not re-used to allow lock-free
   lookups to proceed safely.

4. Pending. index >= 0, key != NULL, and value == NULL  (split only)
   Not yet inserted in split-table.
*/

/*
Preserving insertion order

It's simple for combined table.  Since dk_entries is mostly append only, we can
get insertion order by just iterating dk_entries.

One exception is .popitem().  It removes last item in dk_entries and decrement
dk_nentries to achieve amortized O(1).  Since there are DKIX_DUMMY remains in
dk_indices, we can't increment dk_usable even though dk_nentries is
decremented.

To preserve the order in a split table, a bit vector is used  to record the
insertion order. When a key is inserted the bit vector is shifted up by 4 bits
and the index of the key is stored in the low 4 bits.
As a consequence of this, split keys have a maximum size of 16.
*/

/* PyDict_MINSIZE is the starting size for any new dict.
 * 8 allows dicts with no more than 5 active entries; experiments suggested
 * this suffices for the majority of dicts (consisting mostly of usually-small
 * dicts created to pass keyword arguments).
 * Making this 8, rather than 4 reduces the number of resizes for most
 * dictionaries, without any significant extra memory use.
 */
#define PyDict_LOG_MINSIZE 3
#define PyDict_MINSIZE 8

#include "Python.h"
#include "pycore_bitutils.h"      // _Py_bit_length
#include "pycore_call.h"          // _PyObject_CallNoArgs()
#include "pycore_ceval.h"         // _PyEval_GetBuiltin()
#include "pycore_code.h"          // stats
#include "pycore_critical_section.h" // Py_BEGIN_CRITICAL_SECTION, Py_END_CRITICAL_SECTION
#include "pycore_dict.h"          // export _PyDict_SizeOf()
#include "pycore_freelist.h"      // _PyFreeListState_GET()
#include "pycore_gc.h"            // _PyObject_GC_IS_TRACKED()
#include "pycore_object.h"        // _PyObject_GC_TRACK(), _PyDebugAllocatorStats()
#include "pycore_pyatomic_ft_wrappers.h" // FT_ATOMIC_LOAD_SSIZE_RELAXED
#include "pycore_pyerrors.h"      // _PyErr_GetRaisedException()
#include "pycore_pystate.h"       // _PyThreadState_GET()
#include "pycore_setobject.h"     // _PySet_NextEntry()
#include "pycore_tuple.h"         // _PyTuple_Recycle()
#include "pycore_unicodeobject.h" // _PyUnicode_InternImmortal()

#include "stringlib/eq.h"                // unicode_eq()
#include <stdbool.h>


/*[clinic input]
class dict "PyDictObject *" "&PyDict_Type"
[clinic start generated code]*/
/*[clinic end generated code: output=da39a3ee5e6b4b0d input=f157a5a0ce9589d6]*/


/*
To ensure the lookup algorithm terminates, there must be at least one Unused
slot (NULL key) in the table.
To avoid slowing down lookups on a near-full table, we resize the table when
it's USABLE_FRACTION (currently two-thirds) full.
*/

#ifdef Py_GIL_DISABLED

static inline void
ASSERT_DICT_LOCKED(PyObject *op)
{
    _Py_CRITICAL_SECTION_ASSERT_OBJECT_LOCKED(op);
}
#define ASSERT_DICT_LOCKED(op) ASSERT_DICT_LOCKED(_Py_CAST(PyObject*, op))
#define ASSERT_WORLD_STOPPED_OR_DICT_LOCKED(op)                         \
    if (!_PyInterpreterState_GET()->stoptheworld.world_stopped) {       \
        ASSERT_DICT_LOCKED(op);                                         \
    }
#define ASSERT_WORLD_STOPPED_OR_OBJ_LOCKED(op)                         \
    if (!_PyInterpreterState_GET()->stoptheworld.world_stopped) {      \
        _Py_CRITICAL_SECTION_ASSERT_OBJECT_LOCKED(op);                 \
    }

#define IS_DICT_SHARED(mp) _PyObject_GC_IS_SHARED(mp)
#define SET_DICT_SHARED(mp) _PyObject_GC_SET_SHARED(mp)
#define LOAD_INDEX(keys, size, idx) _Py_atomic_load_int##size##_relaxed(&((const int##size##_t*)keys->dk_indices)[idx]);
#define STORE_INDEX(keys, size, idx, value) _Py_atomic_store_int##size##_relaxed(&((int##size##_t*)keys->dk_indices)[idx], (int##size##_t)value);
#define ASSERT_OWNED_OR_SHARED(mp) \
    assert(_Py_IsOwnedByCurrentThread((PyObject *)mp) || IS_DICT_SHARED(mp));

#define LOCK_KEYS_IF_SPLIT(keys, kind) \
        if (kind == DICT_KEYS_SPLIT) { \
            LOCK_KEYS(keys);           \
        }

#define UNLOCK_KEYS_IF_SPLIT(keys, kind) \
        if (kind == DICT_KEYS_SPLIT) {   \
            UNLOCK_KEYS(keys);           \
        }

static inline Py_ssize_t
load_keys_nentries(PyDictObject *mp)
{
    PyDictKeysObject *keys = _Py_atomic_load_ptr(&mp->ma_keys);
    return _Py_atomic_load_ssize(&keys->dk_nentries);
}

static inline void
set_keys(PyDictObject *mp, PyDictKeysObject *keys)
{
    ASSERT_OWNED_OR_SHARED(mp);
    _Py_atomic_store_ptr_release(&mp->ma_keys, keys);
}

static inline void
set_values(PyDictObject *mp, PyDictValues *values)
{
    ASSERT_OWNED_OR_SHARED(mp);
    _Py_atomic_store_ptr_release(&mp->ma_values, values);
}

#define LOCK_KEYS(keys) PyMutex_LockFlags(&keys->dk_mutex, _Py_LOCK_DONT_DETACH)
#define UNLOCK_KEYS(keys) PyMutex_Unlock(&keys->dk_mutex)

#define ASSERT_KEYS_LOCKED(keys) assert(PyMutex_IsLocked(&keys->dk_mutex))
#define LOAD_SHARED_KEY(key) _Py_atomic_load_ptr_acquire(&key)
#define STORE_SHARED_KEY(key, value) _Py_atomic_store_ptr_release(&key, value)
// Inc refs the keys object, giving the previous value
#define INCREF_KEYS(dk)  _Py_atomic_add_ssize(&dk->dk_refcnt, 1)
// Dec refs the keys object, giving the previous value
#define DECREF_KEYS(dk)  _Py_atomic_add_ssize(&dk->dk_refcnt, -1)
#define LOAD_KEYS_NENTRIES(keys) _Py_atomic_load_ssize_relaxed(&keys->dk_nentries)

#define INCREF_KEYS_FT(dk) dictkeys_incref(dk)
#define DECREF_KEYS_FT(dk, shared) dictkeys_decref(_PyInterpreterState_GET(), dk, shared)

static inline void split_keys_entry_added(PyDictKeysObject *keys)
{
    ASSERT_KEYS_LOCKED(keys);

    // We increase before we decrease so we never get too small of a value
    // when we're racing with reads
    _Py_atomic_store_ssize_relaxed(&keys->dk_nentries, keys->dk_nentries + 1);
    _Py_atomic_store_ssize_release(&keys->dk_usable, keys->dk_usable - 1);
}

#else /* Py_GIL_DISABLED */

#define ASSERT_DICT_LOCKED(op)
#define ASSERT_WORLD_STOPPED_OR_DICT_LOCKED(op)
#define ASSERT_WORLD_STOPPED_OR_OBJ_LOCKED(op)
#define LOCK_KEYS(keys)
#define UNLOCK_KEYS(keys)
#define ASSERT_KEYS_LOCKED(keys)
#define LOAD_SHARED_KEY(key) key
#define STORE_SHARED_KEY(key, value) key = value
#define INCREF_KEYS(dk)  dk->dk_refcnt++
#define DECREF_KEYS(dk)  dk->dk_refcnt--
#define LOAD_KEYS_NENTRIES(keys) keys->dk_nentries
#define INCREF_KEYS_FT(dk)
#define DECREF_KEYS_FT(dk, shared)
#define LOCK_KEYS_IF_SPLIT(keys, kind)
#define UNLOCK_KEYS_IF_SPLIT(keys, kind)
#define IS_DICT_SHARED(mp) (false)
#define SET_DICT_SHARED(mp)
#define LOAD_INDEX(keys, size, idx) ((const int##size##_t*)(keys->dk_indices))[idx]
#define STORE_INDEX(keys, size, idx, value) ((int##size##_t*)(keys->dk_indices))[idx] = (int##size##_t)value

static inline void split_keys_entry_added(PyDictKeysObject *keys)
{
    keys->dk_usable--;
    keys->dk_nentries++;
}

static inline void
set_keys(PyDictObject *mp, PyDictKeysObject *keys)
{
    mp->ma_keys = keys;
}

static inline void
set_values(PyDictObject *mp, PyDictValues *values)
{
    mp->ma_values = values;
}

static inline Py_ssize_t
load_keys_nentries(PyDictObject *mp)
{
    return mp->ma_keys->dk_nentries;
}


#endif

#define STORE_KEY(ep, key) FT_ATOMIC_STORE_PTR_RELEASE((ep)->me_key, key)
#define STORE_VALUE(ep, value) FT_ATOMIC_STORE_PTR_RELEASE((ep)->me_value, value)
#define STORE_SPLIT_VALUE(mp, idx, value) FT_ATOMIC_STORE_PTR_RELEASE(mp->ma_values->values[idx], value)
#define STORE_HASH(ep, hash) FT_ATOMIC_STORE_SSIZE_RELAXED((ep)->me_hash, hash)
#define STORE_KEYS_USABLE(keys, usable) FT_ATOMIC_STORE_SSIZE_RELAXED(keys->dk_usable, usable)
#define STORE_KEYS_NENTRIES(keys, nentries) FT_ATOMIC_STORE_SSIZE_RELAXED(keys->dk_nentries, nentries)
#define STORE_USED(mp, used) FT_ATOMIC_STORE_SSIZE_RELAXED(mp->ma_used, used)

#define PERTURB_SHIFT 5

/*
Major subtleties ahead:  Most hash schemes depend on having a "good" hash
function, in the sense of simulating randomness.  Python doesn't:  its most
important hash functions (for ints) are very regular in common
cases:

  >>>[hash(i) for i in range(4)]
  [0, 1, 2, 3]

This isn't necessarily bad!  To the contrary, in a table of size 2**i, taking
the low-order i bits as the initial table index is extremely fast, and there
are no collisions at all for dicts indexed by a contiguous range of ints. So
this gives better-than-random behavior in common cases, and that's very
desirable.

OTOH, when collisions occur, the tendency to fill contiguous slices of the
hash table makes a good collision resolution strategy crucial.  Taking only
the last i bits of the hash code is also vulnerable:  for example, consider
the list [i << 16 for i in range(20000)] as a set of keys.  Since ints are
their own hash codes, and this fits in a dict of size 2**15, the last 15 bits
 of every hash code are all 0:  they *all* map to the same table index.

But catering to unusual cases should not slow the usual ones, so we just take
the last i bits anyway.  It's up to collision resolution to do the rest.  If
we *usually* find the key we're looking for on the first try (and, it turns
out, we usually do -- the table load factor is kept under 2/3, so the odds
are solidly in our favor), then it makes best sense to keep the initial index
computation dirt cheap.

The first half of collision resolution is to visit table indices via this
recurrence:

    j = ((5*j) + 1) mod 2**i

For any initial j in range(2**i), repeating that 2**i times generates each
int in range(2**i) exactly once (see any text on random-number generation for
proof).  By itself, this doesn't help much:  like linear probing (setting
j += 1, or j -= 1, on each loop trip), it scans the table entries in a fixed
order.  This would be bad, except that's not the only thing we do, and it's
actually *good* in the common cases where hash keys are consecutive.  In an
example that's really too small to make this entirely clear, for a table of
size 2**3 the order of indices is:

    0 -> 1 -> 6 -> 7 -> 4 -> 5 -> 2 -> 3 -> 0 [and here it's repeating]

If two things come in at index 5, the first place we look after is index 2,
not 6, so if another comes in at index 6 the collision at 5 didn't hurt it.
Linear probing is deadly in this case because there the fixed probe order
is the *same* as the order consecutive keys are likely to arrive.  But it's
extremely unlikely hash codes will follow a 5*j+1 recurrence by accident,
and certain that consecutive hash codes do not.

The other half of the strategy is to get the other bits of the hash code
into play.  This is done by initializing a (unsigned) vrbl "perturb" to the
full hash code, and changing the recurrence to:

    perturb >>= PERTURB_SHIFT;
    j = (5*j) + 1 + perturb;
    use j % 2**i as the next table index;

Now the probe sequence depends (eventually) on every bit in the hash code,
and the pseudo-scrambling property of recurring on 5*j+1 is more valuable,
because it quickly magnifies small differences in the bits that didn't affect
the initial index.  Note that because perturb is unsigned, if the recurrence
is executed often enough perturb eventually becomes and remains 0.  At that
point (very rarely reached) the recurrence is on (just) 5*j+1 again, and
that's certain to find an empty slot eventually (since it generates every int
in range(2**i), and we make sure there's always at least one empty slot).

Selecting a good value for PERTURB_SHIFT is a balancing act.  You want it
small so that the high bits of the hash code continue to affect the probe
sequence across iterations; but you want it large so that in really bad cases
the high-order hash bits have an effect on early iterations.  5 was "the
best" in minimizing total collisions across experiments Tim Peters ran (on
both normal and pathological cases), but 4 and 6 weren't significantly worse.

Historical: Reimer Behrends contributed the idea of using a polynomial-based
approach, using repeated multiplication by x in GF(2**n) where an irreducible
polynomial for each table size was chosen such that x was a primitive root.
Christian Tismer later extended that to use division by x instead, as an
efficient way to get the high bits of the hash code into play.  This scheme
also gave excellent collision statistics, but was more expensive:  two
if-tests were required inside the loop; computing "the next" index took about
the same number of operations but without as much potential parallelism
(e.g., computing 5*j can go on at the same time as computing 1+perturb in the
above, and then shifting perturb can be done while the table index is being
masked); and the PyDictObject struct required a member to hold the table's
polynomial.  In Tim's experiments the current scheme ran faster, produced
equally good collision statistics, needed less code & used less memory.

*/

static int dictresize(PyInterpreterState *interp, PyDictObject *mp,
                      uint8_t log_newsize, int unicode);

static PyObject* dict_iter(PyObject *dict);

static int
setitem_lock_held(PyDictObject *mp, PyObject *key, PyObject *value);
static int
dict_setdefault_ref_lock_held(PyObject *d, PyObject *key, PyObject *default_value,
                    PyObject **result, int incref_result);

#ifndef NDEBUG
static int _PyObject_InlineValuesConsistencyCheck(PyObject *obj);
#endif

#include "clinic/dictobject.c.h"


static inline Py_hash_t
unicode_get_hash(PyObject *o)
{
    assert(PyUnicode_CheckExact(o));
    return FT_ATOMIC_LOAD_SSIZE_RELAXED(_PyASCIIObject_CAST(o)->hash);
}

/* Print summary info about the state of the optimized allocator */
void
_PyDict_DebugMallocStats(FILE *out)
{
    _PyDebugAllocatorStats(out, "free PyDictObject",
                           _Py_FREELIST_SIZE(dicts),
                           sizeof(PyDictObject));
    _PyDebugAllocatorStats(out, "free PyDictKeysObject",
                           _Py_FREELIST_SIZE(dictkeys),
                           sizeof(PyDictKeysObject));
}

#define DK_MASK(dk) (DK_SIZE(dk)-1)

#define _Py_DICT_IMMORTAL_INITIAL_REFCNT PY_SSIZE_T_MIN

static void free_keys_object(PyDictKeysObject *keys, bool use_qsbr);

/* PyDictKeysObject has refcounts like PyObject does, so we have the
   following two functions to mirror what Py_INCREF() and Py_DECREF() do.
   (Keep in mind that PyDictKeysObject isn't actually a PyObject.)
   Likewise a PyDictKeysObject can be immortal (e.g. Py_EMPTY_KEYS),
   so we apply a naive version of what Py_INCREF() and Py_DECREF() do
   for immortal objects. */

static inline void
dictkeys_incref(PyDictKeysObject *dk)
{
    if (FT_ATOMIC_LOAD_SSIZE_RELAXED(dk->dk_refcnt) < 0) {
        assert(FT_ATOMIC_LOAD_SSIZE_RELAXED(dk->dk_refcnt) == _Py_DICT_IMMORTAL_INITIAL_REFCNT);
        return;
    }
#ifdef Py_REF_DEBUG
    _Py_IncRefTotal(_PyThreadState_GET());
#endif
    INCREF_KEYS(dk);
}

static inline void
dictkeys_decref(PyInterpreterState *interp, PyDictKeysObject *dk, bool use_qsbr)
{
    if (FT_ATOMIC_LOAD_SSIZE_RELAXED(dk->dk_refcnt) < 0) {
        assert(FT_ATOMIC_LOAD_SSIZE_RELAXED(dk->dk_refcnt) == _Py_DICT_IMMORTAL_INITIAL_REFCNT);
        return;
    }
    assert(FT_ATOMIC_LOAD_SSIZE(dk->dk_refcnt) > 0);
#ifdef Py_REF_DEBUG
    _Py_DecRefTotal(_PyThreadState_GET());
#endif
    if (DECREF_KEYS(dk) == 1) {
        if (DK_IS_UNICODE(dk)) {
            PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(dk);
            Py_ssize_t i, n;
            for (i = 0, n = dk->dk_nentries; i < n; i++) {
                Py_XDECREF(entries[i].me_key);
                Py_XDECREF(entries[i].me_value);
            }
        }
        else {
            PyDictKeyEntry *entries = DK_ENTRIES(dk);
            Py_ssize_t i, n;
            for (i = 0, n = dk->dk_nentries; i < n; i++) {
                Py_XDECREF(entries[i].me_key);
                Py_XDECREF(entries[i].me_value);
            }
        }
        free_keys_object(dk, use_qsbr);
    }
}

/* lookup indices.  returns DKIX_EMPTY, DKIX_DUMMY, or ix >=0 */
static inline Py_ssize_t
dictkeys_get_index(const PyDictKeysObject *keys, Py_ssize_t i)
{
    int log2size = DK_LOG_SIZE(keys);
    Py_ssize_t ix;

    if (log2size < 8) {
        ix = LOAD_INDEX(keys, 8, i);
    }
    else if (log2size < 16) {
        ix = LOAD_INDEX(keys, 16, i);
    }
#if SIZEOF_VOID_P > 4
    else if (log2size >= 32) {
        ix = LOAD_INDEX(keys, 64, i);
    }
#endif
    else {
        ix = LOAD_INDEX(keys, 32, i);
    }
    assert(ix >= DKIX_DUMMY);
    return ix;
}

/* write to indices. */
static inline void
dictkeys_set_index(PyDictKeysObject *keys, Py_ssize_t i, Py_ssize_t ix)
{
    int log2size = DK_LOG_SIZE(keys);

    assert(ix >= DKIX_DUMMY);
    assert(keys->dk_version == 0);

    if (log2size < 8) {
        assert(ix <= 0x7f);
        STORE_INDEX(keys, 8, i, ix);
    }
    else if (log2size < 16) {
        assert(ix <= 0x7fff);
        STORE_INDEX(keys, 16, i, ix);
    }
#if SIZEOF_VOID_P > 4
    else if (log2size >= 32) {
        STORE_INDEX(keys, 64, i, ix);
    }
#endif
    else {
        assert(ix <= 0x7fffffff);
        STORE_INDEX(keys, 32, i, ix);
    }
}


/* USABLE_FRACTION is the maximum dictionary load.
 * Increasing this ratio makes dictionaries more dense resulting in more
 * collisions.  Decreasing it improves sparseness at the expense of spreading
 * indices over more cache lines and at the cost of total memory consumed.
 *
 * USABLE_FRACTION must obey the following:
 *     (0 < USABLE_FRACTION(n) < n) for all n >= 2
 *
 * USABLE_FRACTION should be quick to calculate.
 * Fractions around 1/2 to 2/3 seem to work well in practice.
 */
#define USABLE_FRACTION(n) (((n) << 1)/3)

/* Find the smallest dk_size >= minsize. */
static inline uint8_t
calculate_log2_keysize(Py_ssize_t minsize)
{
#if SIZEOF_LONG == SIZEOF_SIZE_T
    minsize = (minsize | PyDict_MINSIZE) - 1;
    return _Py_bit_length(minsize | (PyDict_MINSIZE-1));
#elif defined(_MSC_VER)
    // On 64bit Windows, sizeof(long) == 4.
    minsize = (minsize | PyDict_MINSIZE) - 1;
    unsigned long msb;
    _BitScanReverse64(&msb, (uint64_t)minsize);
    return (uint8_t)(msb + 1);
#else
    uint8_t log2_size;
    for (log2_size = PyDict_LOG_MINSIZE;
            (((Py_ssize_t)1) << log2_size) < minsize;
            log2_size++)
        ;
    return log2_size;
#endif
}

/* estimate_keysize is reverse function of USABLE_FRACTION.
 *
 * This can be used to reserve enough size to insert n entries without
 * resizing.
 */
static inline uint8_t
estimate_log2_keysize(Py_ssize_t n)
{
    return calculate_log2_keysize((n*3 + 1) / 2);
}


/* GROWTH_RATE. Growth rate upon hitting maximum load.
 * Currently set to used*3.
 * This means that dicts double in size when growing without deletions,
 * but have more head room when the number of deletions is on a par with the
 * number of insertions.  See also bpo-17563 and bpo-33205.
 *
 * GROWTH_RATE was set to used*4 up to version 3.2.
 * GROWTH_RATE was set to used*2 in version 3.3.0
 * GROWTH_RATE was set to used*2 + capacity/2 in 3.4.0-3.6.0.
 */
#define GROWTH_RATE(d) ((d)->ma_used*3)

/* This immutable, empty PyDictKeysObject is used for PyDict_Clear()
 * (which cannot fail and thus can do no allocation).
 *
 * See https://github.com/python/cpython/pull/127568#discussion_r1868070614
 * for the rationale of using dk_log2_index_bytes=3 instead of 0.
 */
static PyDictKeysObject empty_keys_struct = {
        _Py_DICT_IMMORTAL_INITIAL_REFCNT, /* dk_refcnt */
        0, /* dk_log2_size */
        3, /* dk_log2_index_bytes */
        DICT_KEYS_UNICODE, /* dk_kind */
#ifdef Py_GIL_DISABLED
        {0}, /* dk_mutex */
#endif
        1, /* dk_version */
        0, /* dk_usable (immutable) */
        0, /* dk_nentries */
        {DKIX_EMPTY, DKIX_EMPTY, DKIX_EMPTY, DKIX_EMPTY,
         DKIX_EMPTY, DKIX_EMPTY, DKIX_EMPTY, DKIX_EMPTY}, /* dk_indices */
};

#define Py_EMPTY_KEYS &empty_keys_struct

/* Uncomment to check the dict content in _PyDict_CheckConsistency() */
// #define DEBUG_PYDICT

#ifdef DEBUG_PYDICT
#  define ASSERT_CONSISTENT(op) assert(_PyDict_CheckConsistency((PyObject *)(op), 1))
#else
#  define ASSERT_CONSISTENT(op) assert(_PyDict_CheckConsistency((PyObject *)(op), 0))
#endif

static inline int
get_index_from_order(PyDictObject *mp, Py_ssize_t i)
{
    assert(mp->ma_used <= SHARED_KEYS_MAX_SIZE);
    assert(i < mp->ma_values->size);
    uint8_t *array = get_insertion_order_array(mp->ma_values);
    return array[i];
}

#ifdef DEBUG_PYDICT
static void
dump_entries(PyDictKeysObject *dk)
{
    for (Py_ssize_t i = 0; i < dk->dk_nentries; i++) {
        if (DK_IS_UNICODE(dk)) {
            PyDictUnicodeEntry *ep = &DK_UNICODE_ENTRIES(dk)[i];
            printf("key=%p value=%p\n", ep->me_key, ep->me_value);
        }
        else {
            PyDictKeyEntry *ep = &DK_ENTRIES(dk)[i];
            printf("key=%p hash=%lx value=%p\n", ep->me_key, ep->me_hash, ep->me_value);
        }
    }
}
#endif

int
_PyDict_CheckConsistency(PyObject *op, int check_content)
{
    ASSERT_WORLD_STOPPED_OR_DICT_LOCKED(op);

#define CHECK(expr) \
    do { if (!(expr)) { _PyObject_ASSERT_FAILED_MSG(op, Py_STRINGIFY(expr)); } } while (0)

    assert(op != NULL);
    CHECK(PyDict_Check(op));
    PyDictObject *mp = (PyDictObject *)op;

    PyDictKeysObject *keys = mp->ma_keys;
    int splitted = _PyDict_HasSplitTable(mp);
    Py_ssize_t usable = USABLE_FRACTION(DK_SIZE(keys));

    // In the free-threaded build, shared keys may be concurrently modified,
    // so use atomic loads.
    Py_ssize_t dk_usable = FT_ATOMIC_LOAD_SSIZE_ACQUIRE(keys->dk_usable);
    Py_ssize_t dk_nentries = FT_ATOMIC_LOAD_SSIZE_ACQUIRE(keys->dk_nentries);

    CHECK(0 <= mp->ma_used && mp->ma_used <= usable);
    CHECK(0 <= dk_usable && dk_usable <= usable);
    CHECK(0 <= dk_nentries && dk_nentries <= usable);
    CHECK(dk_usable + dk_nentries <= usable);

    if (!splitted) {
        /* combined table */
        CHECK(keys->dk_kind != DICT_KEYS_SPLIT);
        CHECK(keys->dk_refcnt == 1 || keys == Py_EMPTY_KEYS);
    }
    else {
        CHECK(keys->dk_kind == DICT_KEYS_SPLIT);
        CHECK(mp->ma_used <= SHARED_KEYS_MAX_SIZE);
        if (mp->ma_values->embedded) {
            CHECK(mp->ma_values->embedded == 1);
            CHECK(mp->ma_values->valid == 1);
        }
    }

    if (check_content) {
        LOCK_KEYS_IF_SPLIT(keys, keys->dk_kind);
        for (Py_ssize_t i=0; i < DK_SIZE(keys); i++) {
            Py_ssize_t ix = dictkeys_get_index(keys, i);
            CHECK(DKIX_DUMMY <= ix && ix <= usable);
        }

        if (keys->dk_kind == DICT_KEYS_GENERAL) {
            PyDictKeyEntry *entries = DK_ENTRIES(keys);
            for (Py_ssize_t i=0; i < usable; i++) {
                PyDictKeyEntry *entry = &entries[i];
                PyObject *key = entry->me_key;

                if (key != NULL) {
                    /* test_dict fails if PyObject_Hash() is called again */
                    CHECK(entry->me_hash != -1);
                    CHECK(entry->me_value != NULL);

                    if (PyUnicode_CheckExact(key)) {
                        Py_hash_t hash = unicode_get_hash(key);
                        CHECK(entry->me_hash == hash);
                    }
                }
            }
        }
        else {
            PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(keys);
            for (Py_ssize_t i=0; i < usable; i++) {
                PyDictUnicodeEntry *entry = &entries[i];
                PyObject *key = entry->me_key;

                if (key != NULL) {
                    CHECK(PyUnicode_CheckExact(key));
                    Py_hash_t hash = unicode_get_hash(key);
                    CHECK(hash != -1);
                    if (!splitted) {
                        CHECK(entry->me_value != NULL);
                    }
                }

                if (splitted) {
                    CHECK(entry->me_value == NULL);
                }
            }
        }

        if (splitted) {
            CHECK(mp->ma_used <= SHARED_KEYS_MAX_SIZE);
            /* splitted table */
            int duplicate_check = 0;
            for (Py_ssize_t i=0; i < mp->ma_used; i++) {
                int index = get_index_from_order(mp, i);
                CHECK((duplicate_check & (1<<index)) == 0);
                duplicate_check |= (1<<index);
                CHECK(mp->ma_values->values[index] != NULL);
            }
        }
        UNLOCK_KEYS_IF_SPLIT(keys, keys->dk_kind);
    }
    return 1;

#undef CHECK
}


static PyDictKeysObject*
new_keys_object(PyInterpreterState *interp, uint8_t log2_size, bool unicode)
{
    Py_ssize_t usable;
    int log2_bytes;
    size_t entry_size = unicode ? sizeof(PyDictUnicodeEntry) : sizeof(PyDictKeyEntry);

    assert(log2_size >= PyDict_LOG_MINSIZE);

    usable = USABLE_FRACTION((size_t)1<<log2_size);
    if (log2_size < 8) {
        log2_bytes = log2_size;
    }
    else if (log2_size < 16) {
        log2_bytes = log2_size + 1;
    }
#if SIZEOF_VOID_P > 4
    else if (log2_size >= 32) {
        log2_bytes = log2_size + 3;
    }
#endif
    else {
        log2_bytes = log2_size + 2;
    }

    PyDictKeysObject *dk = NULL;
    if (log2_size == PyDict_LOG_MINSIZE && unicode) {
        dk = _Py_FREELIST_POP_MEM(dictkeys);
    }
    if (dk == NULL) {
        dk = PyMem_Malloc(sizeof(PyDictKeysObject)
                          + ((size_t)1 << log2_bytes)
                          + entry_size * usable);
        if (dk == NULL) {
            PyErr_NoMemory();
            return NULL;
        }
    }
#ifdef Py_REF_DEBUG
    _Py_IncRefTotal(_PyThreadState_GET());
#endif
    dk->dk_refcnt = 1;
    dk->dk_log2_size = log2_size;
    dk->dk_log2_index_bytes = log2_bytes;
    dk->dk_kind = unicode ? DICT_KEYS_UNICODE : DICT_KEYS_GENERAL;
#ifdef Py_GIL_DISABLED
    dk->dk_mutex = (PyMutex){0};
#endif
    dk->dk_nentries = 0;
    dk->dk_usable = usable;
    dk->dk_version = 0;
    memset(&dk->dk_indices[0], 0xff, ((size_t)1 << log2_bytes));
    memset(&dk->dk_indices[(size_t)1 << log2_bytes], 0, entry_size * usable);
    return dk;
}

static void
free_keys_object(PyDictKeysObject *keys, bool use_qsbr)
{
#ifdef Py_GIL_DISABLED
    if (use_qsbr) {
        _PyMem_FreeDelayed(keys);
        return;
    }
#endif
    if (DK_LOG_SIZE(keys) == PyDict_LOG_MINSIZE && keys->dk_kind == DICT_KEYS_UNICODE) {
        _Py_FREELIST_FREE(dictkeys, keys, PyMem_Free);
    }
    else {
        PyMem_Free(keys);
    }
}

static size_t
values_size_from_count(size_t count)
{
    assert(count >= 1);
    size_t suffix_size = _Py_SIZE_ROUND_UP(count, sizeof(PyObject *));
    assert(suffix_size < 128);
    assert(suffix_size % sizeof(PyObject *) == 0);
    return (count + 1) * sizeof(PyObject *) + suffix_size;
}

#define CACHED_KEYS(tp) (((PyHeapTypeObject*)tp)->ht_cached_keys)

static inline PyDictValues*
new_values(size_t size)
{
    size_t n = values_size_from_count(size);
    PyDictValues *res = (PyDictValues *)PyMem_Malloc(n);
    if (res == NULL) {
        return NULL;
    }
    res->embedded = 0;
    res->size = 0;
    assert(size < 256);
    res->capacity = (uint8_t)size;
    return res;
}

static inline void
free_values(PyDictValues *values, bool use_qsbr)
{
    assert(values->embedded == 0);
#ifdef Py_GIL_DISABLED
    if (use_qsbr) {
        _PyMem_FreeDelayed(values);
        return;
    }
#endif
    PyMem_Free(values);
}

/* Consumes a reference to the keys object */
static PyObject *
new_dict(PyInterpreterState *interp,
         PyDictKeysObject *keys, PyDictValues *values,
         Py_ssize_t used, int free_values_on_failure)
{
    assert(keys != NULL);
    PyDictObject *mp = _Py_FREELIST_POP(PyDictObject, dicts);
    if (mp == NULL) {
        mp = PyObject_GC_New(PyDictObject, &PyDict_Type);
        if (mp == NULL) {
            dictkeys_decref(interp, keys, false);
            if (free_values_on_failure) {
                free_values(values, false);
            }
            return NULL;
        }
    }
    assert(Py_IS_TYPE(mp, &PyDict_Type));
    mp->ma_keys = keys;
    mp->ma_values = values;
    mp->ma_used = used;
    mp->_ma_watcher_tag = 0;
    ASSERT_CONSISTENT(mp);
    _PyObject_GC_TRACK(mp);
    return (PyObject *)mp;
}

static PyObject *
new_dict_with_shared_keys(PyInterpreterState *interp, PyDictKeysObject *keys)
{
    size_t size = shared_keys_usable_size(keys);
    PyDictValues *values = new_values(size);
    if (values == NULL) {
        return PyErr_NoMemory();
    }
    dictkeys_incref(keys);
    for (size_t i = 0; i < size; i++) {
        values->values[i] = NULL;
    }
    return new_dict(interp, keys, values, 0, 1);
}


static PyDictKeysObject *
clone_combined_dict_keys(PyDictObject *orig)
{
    assert(PyDict_Check(orig));
    assert(Py_TYPE(orig)->tp_iter == dict_iter);
    assert(orig->ma_values == NULL);
    assert(orig->ma_keys != Py_EMPTY_KEYS);
    assert(orig->ma_keys->dk_refcnt == 1);

    ASSERT_DICT_LOCKED(orig);

    size_t keys_size = _PyDict_KeysSize(orig->ma_keys);
    PyDictKeysObject *keys = PyMem_Malloc(keys_size);
    if (keys == NULL) {
        PyErr_NoMemory();
        return NULL;
    }

    memcpy(keys, orig->ma_keys, keys_size);

    /* After copying key/value pairs, we need to incref all
       keys and values and they are about to be co-owned by a
       new dict object. */
    PyObject **pkey, **pvalue;
    size_t offs;
    if (DK_IS_UNICODE(orig->ma_keys)) {
        PyDictUnicodeEntry *ep0 = DK_UNICODE_ENTRIES(keys);
        pkey = &ep0->me_key;
        pvalue = &ep0->me_value;
        offs = sizeof(PyDictUnicodeEntry) / sizeof(PyObject*);
    }
    else {
        PyDictKeyEntry *ep0 = DK_ENTRIES(keys);
        pkey = &ep0->me_key;
        pvalue = &ep0->me_value;
        offs = sizeof(PyDictKeyEntry) / sizeof(PyObject*);
    }

    Py_ssize_t n = keys->dk_nentries;
    for (Py_ssize_t i = 0; i < n; i++) {
        PyObject *value = *pvalue;
        if (value != NULL) {
            Py_INCREF(value);
            Py_INCREF(*pkey);
        }
        pvalue += offs;
        pkey += offs;
    }

    /* Since we copied the keys table we now have an extra reference
       in the system.  Manually call increment _Py_RefTotal to signal that
       we have it now; calling dictkeys_incref would be an error as
       keys->dk_refcnt is already set to 1 (after memcpy). */
#ifdef Py_REF_DEBUG
    _Py_IncRefTotal(_PyThreadState_GET());
#endif
    return keys;
}

PyObject *
PyDict_New(void)
{
    PyInterpreterState *interp = _PyInterpreterState_GET();
    /* We don't incref Py_EMPTY_KEYS here because it is immortal. */
    return new_dict(interp, Py_EMPTY_KEYS, NULL, 0, 0);
}

/* Search index of hash table from offset of entry table */
static Py_ssize_t
lookdict_index(PyDictKeysObject *k, Py_hash_t hash, Py_ssize_t index)
{
    size_t mask = DK_MASK(k);
    size_t perturb = (size_t)hash;
    size_t i = (size_t)hash & mask;

    for (;;) {
        Py_ssize_t ix = dictkeys_get_index(k, i);
        if (ix == index) {
            return i;
        }
        if (ix == DKIX_EMPTY) {
            return DKIX_EMPTY;
        }
        perturb >>= PERTURB_SHIFT;
        i = mask & (i*5 + perturb + 1);
    }
    Py_UNREACHABLE();
}