/* 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.

*/

#include "Python.h"

/* Set a key error with the specified argument, wrapping it in a

* tuple automatically so that tuple keys are not unpacked as the

* exception arguments. */

static void

set_key_error(PyObject *arg)

{

PyObject *tup;

tup = PyTuple_Pack(1, arg);

if (!tup)

return; /* caller will expect error to be set anyway */

PyErr_SetObject(PyExc_KeyError, tup);

Py_DECREF(tup);

}

/* Define this out if you don't want conversion statistics on exit. */

#undef SHOW_CONVERSION_COUNTS

/* See large comment block below. This must be >= 1. */

#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 strings and ints) are very regular in common

cases:

>>> map(hash, (0, 1, 2, 3))

[0, 1, 2, 3]

>>> map(hash, ("namea", "nameb", "namec", "named"))

[-1658398457, -1658398460, -1658398459, -1658398462]

>>>

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.

The same is approximately true when keys are "consecutive" strings. 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

[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:

j = (5*j) + 1 + perturb;

perturb >>= PERTURB_SHIFT;

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.

Theoretical Python 2.5 headache: hash codes are only C "long", but

sizeof(Py_ssize_t) > sizeof(long) may be possible. In that case, and if a

dict is genuinely huge, then only the slots directly reachable via indexing

by a C long can be the first slot in a probe sequence. The probe sequence

will still eventually reach every slot in the table, but the collision rate

on initial probes may be much higher than this scheme was designed for.

Getting a hash code as fat as Py_ssize_t is the only real cure. But in

practice, this probably won't make a lick of difference for many years (at

which point everyone will have terabytes of RAM on 64-bit boxes).

*/

/* Object used as dummy key to fill deleted entries */

static PyObject *dummy = NULL; /* Initialized by first call to newPyDictObject() */

#ifdef Py_REF_DEBUG

PyObject *

_PyDict_Dummy(void)

{

return dummy;

}

#endif

/* forward declarations */

static PyDictEntry *

lookdict_string(PyDictObject *mp, PyObject *key, long hash);

#ifdef SHOW_CONVERSION_COUNTS

static long created = 0L;

static long converted = 0L;

static void

show_counts(void)

{

fprintf(stderr, "created %ld string dicts\n", created);

fprintf(stderr, "converted %ld to normal dicts\n", converted);

fprintf(stderr, "%.2f%% conversion rate\n", (100.0*converted)/created);

}

#endif

/* Debug statistic to compare allocations with reuse through the free list */

#undef SHOW_ALLOC_COUNT

#ifdef SHOW_ALLOC_COUNT

static size_t count_alloc = 0;

static size_t count_reuse = 0;

static void

show_alloc(void)

{

fprintf(stderr, "Dict allocations: %" PY_FORMAT_SIZE_T "d\n",

count_alloc);

fprintf(stderr, "Dict reuse through freelist: %" PY_FORMAT_SIZE_T

"d\n", count_reuse);

fprintf(stderr, "%.2f%% reuse rate\n\n",

(100.0*count_reuse/(count_alloc+count_reuse)));

}

#endif

/* Debug statistic to count GC tracking of dicts */

#ifdef SHOW_TRACK_COUNT

static Py_ssize_t count_untracked = 0;

static Py_ssize_t count_tracked = 0;

static void

show_track(void)

{

fprintf(stderr, "Dicts created: %" PY_FORMAT_SIZE_T "d\n",

count_tracked + count_untracked);

fprintf(stderr, "Dicts tracked by the GC: %" PY_FORMAT_SIZE_T

"d\n", count_tracked);

fprintf(stderr, "%.2f%% dict tracking rate\n\n",

(100.0*count_tracked/(count_untracked+count_tracked)));

}

#endif

/* Initialization macros.

There are two ways to create a dict: PyDict_New() is the main C API

function, and the tp_new slot maps to dict_new(). In the latter case we

can save a little time over what PyDict_New does because it's guaranteed

that the PyDictObject struct is already zeroed out.

Everyone except dict_new() should use EMPTY_TO_MINSIZE (unless they have

an excellent reason not to).

*/

#define INIT_NONZERO_DICT_SLOTS(mp) do { \

(mp)->ma_table = (mp)->ma_smalltable; \

(mp)->ma_mask = PyDict_MINSIZE - 1; \

} while(0)

#define EMPTY_TO_MINSIZE(mp) do { \

memset((mp)->ma_smalltable, 0, sizeof((mp)->ma_smalltable)); \

(mp)->ma_used = (mp)->ma_fill = 0; \

INIT_NONZERO_DICT_SLOTS(mp); \

} while(0)

/* Dictionary reuse scheme to save calls to malloc, free, and memset */

#ifndef PyDict_MAXFREELIST

#define PyDict_MAXFREELIST 80

#endif

static PyDictObject *free_list[PyDict_MAXFREELIST];

static int numfree = 0;

void

PyDict_Fini(void)

{

PyDictObject *op;

while (numfree) {

op = free_list[--numfree];

assert(PyDict_CheckExact(op));

PyObject_GC_Del(op);

}

}

PyObject *

PyDict_New(void)

{

register PyDictObject *mp;

if (dummy == NULL) { /* Auto-initialize dummy */

dummy = PyString_FromString("<dummy key>");

if (dummy == NULL)

return NULL;

#ifdef SHOW_CONVERSION_COUNTS

Py_AtExit(show_counts);

#endif

#ifdef SHOW_ALLOC_COUNT

Py_AtExit(show_alloc);

#endif

#ifdef SHOW_TRACK_COUNT

Py_AtExit(show_track);

#endif

}

if (numfree) {

mp = free_list[--numfree];

assert (mp != NULL);

assert (Py_TYPE(mp) == &PyDict_Type);

_Py_NewReference((PyObject *)mp);

if (mp->ma_fill) {

EMPTY_TO_MINSIZE(mp);

} else {

/* At least set ma_table and ma_mask; these are wrong

if an empty but presized dict is added to freelist */

INIT_NONZERO_DICT_SLOTS(mp);

}

assert (mp->ma_used == 0);

assert (mp->ma_table == mp->ma_smalltable);

assert (mp->ma_mask == PyDict_MINSIZE - 1);

#ifdef SHOW_ALLOC_COUNT

count_reuse++;

#endif

} else {

mp = PyObject_GC_New(PyDictObject, &PyDict_Type);

if (mp == NULL)

return NULL;

EMPTY_TO_MINSIZE(mp);

#ifdef SHOW_ALLOC_COUNT

count_alloc++;

#endif

}

mp->ma_lookup = lookdict_string;

#ifdef SHOW_TRACK_COUNT

count_untracked++;

#endif

#ifdef SHOW_CONVERSION_COUNTS

++created;

#endif

return (PyObject *)mp;

}

/*

The basic lookup function used by all operations.

This is based on Algorithm D from Knuth Vol. 3, Sec. 6.4.

Open addressing is preferred over chaining since the link overhead for

chaining would be substantial (100% with typical malloc overhead).

The initial probe index is computed as hash mod the table size. Subsequent

probe indices are computed as explained earlier.

All arithmetic on hash should ignore overflow.

(The details in this version are due to Tim Peters, building on many past

contributions by Reimer Behrends, Jyrki Alakuijala, Vladimir Marangozov and

Christian Tismer).

lookdict() is general-purpose, and may return NULL if (and only if) a

comparison raises an exception (this was new in Python 2.5).

lookdict_string() below is specialized to string keys, comparison of which can

never raise an exception; that function can never return NULL. For both, when

the key isn't found a PyDictEntry* is returned for which the me_value field is

NULL; this is the slot in the dict at which the key would have been found, and

the caller can (if it wishes) add the <key, value> pair to the returned

PyDictEntry*.

*/

static PyDictEntry *

lookdict(PyDictObject *mp, PyObject *key, register long hash)

{

register size_t i;

register size_t perturb;

register PyDictEntry *freeslot;

register size_t mask = (size_t)mp->ma_mask;

PyDictEntry *ep0 = mp->ma_table;

register PyDictEntry *ep;

register int cmp;

PyObject *startkey;

i = (size_t)hash & mask;

ep = &ep0[i];

if (ep->me_key == NULL || ep->me_key == key)

return ep;

if (ep->me_key == dummy)

freeslot = ep;

else {

if (ep->me_hash == hash) {

startkey = ep->me_key;

Py_INCREF(startkey);

cmp = PyObject_RichCompareBool(startkey, key, Py_EQ);

Py_DECREF(startkey);

if (cmp < 0)

return NULL;

if (ep0 == mp->ma_table && ep->me_key == startkey) {

if (cmp > 0)

return ep;

}

else {

/* The compare did major nasty stuff to the

* dict: start over.

* XXX A clever adversary could prevent this

* XXX from terminating.

*/

return lookdict(mp, key, hash);

}

}

freeslot = NULL;

}

/* In the loop, me_key == dummy is by far (factor of 100s) the

least likely outcome, so test for that last. */

for (perturb = hash; ; perturb >>= PERTURB_SHIFT) {

i = (i << 2) + i + perturb + 1;

ep = &ep0[i & mask];

if (ep->me_key == NULL)

return freeslot == NULL ? ep : freeslot;

if (ep->me_key == key)

return ep;

if (ep->me_hash == hash && ep->me_key != dummy) {

startkey = ep->me_key;

Py_INCREF(startkey);

cmp = PyObject_RichCompareBool(startkey, key, Py_EQ);

Py_DECREF(startkey);

if (cmp < 0)

return NULL;

if (ep0 == mp->ma_table && ep->me_key == startkey) {

if (cmp > 0)

return ep;

}

else {

/* The compare did major nasty stuff to the

* dict: start over.

* XXX A clever adversary could prevent this

* XXX from terminating.

*/

return lookdict(mp, key, hash);

}

}

else if (ep->me_key == dummy && freeslot == NULL)

freeslot = ep;

}

assert(0); /* NOT REACHED */

return 0;

}

/*

* Hacked up version of lookdict which can assume keys are always strings;

* this assumption allows testing for errors during PyObject_RichCompareBool()

* to be dropped; string-string comparisons never raise exceptions. This also

* means we don't need to go through PyObject_RichCompareBool(); we can always

* use _PyString_Eq() directly.

*

* This is valuable because dicts with only string keys are very common.

*/

static PyDictEntry *

lookdict_string(PyDictObject *mp, PyObject *key, register long hash)

{

register size_t i;

register size_t perturb;

register PyDictEntry *freeslot;

register size_t mask = (size_t)mp->ma_mask;

PyDictEntry *ep0 = mp->ma_table;

register PyDictEntry *ep;

/* Make sure this function doesn't have to handle non-string keys,

including subclasses of str; e.g., one reason to subclass

strings is to override __eq__, and for speed we don't cater to

that here. */

if (!PyString_CheckExact(key)) {

#ifdef SHOW_CONVERSION_COUNTS

++converted;

#endif

mp->ma_lookup = lookdict;

return lookdict(mp, key, hash);

}

i = hash & mask;

ep = &ep0[i];

if (ep->me_key == NULL || ep->me_key == key)

return ep;

if (ep->me_key == dummy)

freeslot = ep;

else {

if (ep->me_hash == hash && _PyString_Eq(ep->me_key, key))

return ep;

freeslot = NULL;

}

/* In the loop, me_key == dummy is by far (factor of 100s) the

least likely outcome, so test for that last. */

for (perturb = hash; ; perturb >>= PERTURB_SHIFT) {

i = (i << 2) + i + perturb + 1;

ep = &ep0[i & mask];

if (ep->me_key == NULL)

return freeslot == NULL ? ep : freeslot;

if (ep->me_key == key

|| (ep->me_hash == hash

&& ep->me_key != dummy

&& _PyString_Eq(ep->me_key, key)))

return ep;

if (ep->me_key == dummy && freeslot == NULL)

freeslot = ep;

}

assert(0); /* NOT REACHED */

return 0;

}

#ifdef SHOW_TRACK_COUNT

#define INCREASE_TRACK_COUNT \

(count_tracked++, count_untracked--);

#define DECREASE_TRACK_COUNT \

(count_tracked--, count_untracked++);

#else

#define INCREASE_TRACK_COUNT

#define DECREASE_TRACK_COUNT

#endif

#define MAINTAIN_TRACKING(mp, key, value) \

do { \

if (!_PyObject_GC_IS_TRACKED(mp)) { \

if (_PyObject_GC_MAY_BE_TRACKED(key) || \

_PyObject_GC_MAY_BE_TRACKED(value)) { \

_PyObject_GC_TRACK(mp); \

INCREASE_TRACK_COUNT \

} \

} \

} while(0)

void

_PyDict_MaybeUntrack(PyObject *op)

{

PyDictObject *mp;

PyObject *value;

Py_ssize_t mask, i;

PyDictEntry *ep;

if (!PyDict_CheckExact(op) || !_PyObject_GC_IS_TRACKED(op))

return;

mp = (PyDictObject *) op;

ep = mp->ma_table;

mask = mp->ma_mask;

for (i = 0; i <= mask; i++) {

if ((value = ep[i].me_value) == NULL)

continue;

if (_PyObject_GC_MAY_BE_TRACKED(value) ||

_PyObject_GC_MAY_BE_TRACKED(ep[i].me_key))

return;

}

DECREASE_TRACK_COUNT

_PyObject_GC_UNTRACK(op);

}

/*

Internal routine to insert a new item into the table when you have entry object.

Used by insertdict.

*/

static int

insertdict_by_entry(register PyDictObject *mp, PyObject *key, long hash,

PyDictEntry *ep, PyObject *value)

{

PyObject *old_value;

MAINTAIN_TRACKING(mp, key, value);

if (ep->me_value != NULL) {

old_value = ep->me_value;

ep->me_value = value;

Py_DECREF(old_value); /* which **CAN** re-enter */

Py_DECREF(key);

}

else {

if (ep->me_key == NULL)

mp->ma_fill++;

else {

assert(ep->me_key == dummy);

Py_DECREF(dummy);

}

ep->me_key = key;

ep->me_hash = (Py_ssize_t)hash;

ep->me_value = value;

mp->ma_used++;

}

return 0;

}

/*

Internal routine to insert a new item into the table.

Used both by the internal resize routine and by the public insert routine.

Eats a reference to key and one to value.

Returns -1 if an error occurred, or 0 on success.

*/

static int

insertdict(register PyDictObject *mp, PyObject *key, long hash, PyObject *value)

{

register PyDictEntry *ep;

assert(mp->ma_lookup != NULL);

ep = mp->ma_lookup(mp, key, hash);

if (ep == NULL) {

Py_DECREF(key);

Py_DECREF(value);

return -1;

}

return insertdict_by_entry(mp, key, hash, ep, value);

}

/*

Internal routine used by dictresize() to insert an item which is

known to be absent from the dict. This routine also assumes that

the dict contains no deleted entries. Besides the performance benefit,

using insertdict() in dictresize() is dangerous (SF bug #1456209).

Note that no refcounts are changed by this routine; if needed, the caller

is responsible for incref'ing `key` and `value`.

*/

static void

insertdict_clean(register PyDictObject *mp, PyObject *key, long hash,

PyObject *value)

{

register size_t i;

register size_t perturb;

register size_t mask = (size_t)mp->ma_mask;

PyDictEntry *ep0 = mp->ma_table;

register PyDictEntry *ep;

MAINTAIN_TRACKING(mp, key, value);

i = hash & mask;

ep = &ep0[i];

for (perturb = hash; ep->me_key != NULL; perturb >>= PERTURB_SHIFT) {

i = (i << 2) + i + perturb + 1;

ep = &ep0[i & mask];

}

assert(ep->me_value == NULL);

mp->ma_fill++;

ep->me_key = key;

ep->me_hash = (Py_ssize_t)hash;

ep->me_value = value;

mp->ma_used++;

}

/*

Restructure the table by allocating a new table and reinserting all

items again. When entries have been deleted, the new table may

actually be smaller than the old one.

*/

static int

dictresize(PyDictObject *mp, Py_ssize_t minused)

{

Py_ssize_t newsize;

PyDictEntry *oldtable, *newtable, *ep;

Py_ssize_t i;

int is_oldtable_malloced;

PyDictEntry small_copy[PyDict_MINSIZE];

assert(minused >= 0);

/* Find the smallest table size > minused. */

for (newsize = PyDict_MINSIZE;

newsize <= minused && newsize > 0;

newsize <<= 1)

;

if (newsize <= 0) {

PyErr_NoMemory();

return -1;

}

/* Get space for a new table. */

oldtable = mp->ma_table;

assert(oldtable != NULL);

is_oldtable_malloced = oldtable != mp->ma_smalltable;

if (newsize == PyDict_MINSIZE) {

/* A large table is shrinking, or we can't get any smaller. */

newtable = mp->ma_smalltable;

if (newtable == oldtable) {

if (mp->ma_fill == mp->ma_used) {

/* No dummies, so no point doing anything. */

return 0;

}

/* We're not going to resize it, but rebuild the

table anyway to purge old dummy entries.

Subtle: This is *necessary* if fill==size,

as lookdict needs at least one virgin slot to

terminate failing searches. If fill < size, it's

merely desirable, as dummies slow searches. */

assert(mp->ma_fill > mp->ma_used);

memcpy(small_copy, oldtable, sizeof(small_copy));

oldtable = small_copy;

}

}

else {

newtable = PyMem_NEW(PyDictEntry, newsize);

if (newtable == NULL) {

PyErr_NoMemory();

return -1;

}

}

/* Make the dict empty, using the new table. */

assert(newtable != oldtable);

mp->ma_table = newtable;

mp->ma_mask = newsize - 1;

memset(newtable, 0, sizeof(PyDictEntry) * newsize);

mp->ma_used = 0;

i = mp->ma_fill;

mp->ma_fill = 0;

/* Copy the data over; this is refcount-neutral for active entries;

dummy entries aren't copied over, of course */

for (ep = oldtable; i > 0; ep++) {

if (ep->me_value != NULL) { /* active entry */

--i;

insertdict_clean(mp, ep->me_key, (long)ep->me_hash,

ep->me_value);

}

else if (ep->me_key != NULL) { /* dummy entry */

--i;

assert(ep->me_key == dummy);

Py_DECREF(ep->me_key);

}

/* else key == value == NULL: nothing to do */

}

if (is_oldtable_malloced)

PyMem_DEL(oldtable);

return 0;

}

/* Create a new dictionary pre-sized to hold an estimated number of elements.

Underestimates are okay because the dictionary will resize as necessary.

Overestimates just mean the dictionary will be more sparse than usual.

*/

PyObject *

_PyDict_NewPresized(Py_ssize_t minused)

{

PyObject *op = PyDict_New();

if (minused>5 && op != NULL && dictresize((PyDictObject *)op, minused) == -1) {

Py_DECREF(op);

return NULL;

}

return op;

}

/* Note that, for historical reasons, PyDict_GetItem() suppresses all errors

* that may occur (originally dicts supported only string keys, and exceptions

* weren't possible). So, while the original intent was that a NULL return

* meant the key wasn't present, in reality it can mean that, or that an error

* (suppressed) occurred while computing the key's hash, or that some error

* (suppressed) occurred when comparing keys in the dict's internal probe

* sequence. A nasty example of the latter is when a Python-coded comparison

* function hits a stack-depth error, which can cause this to return NULL

* even if the key is present.

*/

PyObject *

PyDict_GetItem(PyObject *op, PyObject *key)

{

long hash;

PyDictObject *mp = (PyDictObject *)op;

PyDictEntry *ep;

PyThreadState *tstate;

if (!PyDict_Check(op))

return NULL;

if (!PyString_CheckExact(key) ||

(hash = ((PyStringObject *) key)->ob_shash) == -1)

{

hash = PyObject_Hash(key);

if (hash == -1) {

PyErr_Clear();

return NULL;

}

}

/* We can arrive here with a NULL tstate during initialization: try

running "python -Wi" for an example related to string interning.

Let's just hope that no exception occurs then... This must be

_PyThreadState_Current and not PyThreadState_GET() because in debug

mode, the latter complains if tstate is NULL. */

tstate = _PyThreadState_Current;

if (tstate != NULL && tstate->curexc_type != NULL) {

/* preserve the existing exception */

PyObject *err_type, *err_value, *err_tb;

PyErr_Fetch(&err_type, &err_value, &err_tb);

ep = (mp->ma_lookup)(mp, key, hash);

/* ignore errors */

PyErr_Restore(err_type, err_value, err_tb);

if (ep == NULL)

return NULL;

}

else {

ep = (mp->ma_lookup)(mp, key, hash);

if (ep == NULL) {

PyErr_Clear();

return NULL;

}

}

return ep->me_value;

}

/* Variant of PyDict_GetItem() that doesn't suppress exceptions.

This returns NULL *with* an exception set if an exception occurred.

It returns NULL *without* an exception set if the key wasn't present.

*/

PyObject *

_PyDict_GetItemWithError(PyObject *op, PyObject *key)

{

long hash;

PyDictObject *mp = (PyDictObject *)op;

PyDictEntry *ep;

if (!PyDict_Check(op)) {

PyErr_BadInternalCall();

return NULL;

}

if (!PyString_CheckExact(key) ||

(hash = ((PyStringObject *) key)->ob_shash) == -1)

{

hash = PyObject_Hash(key);

if (hash == -1) {

return NULL;

}

}

ep = (mp->ma_lookup)(mp, key, hash);

if (ep == NULL) {

return NULL;

}

return ep->me_value;

}

static int

dict_set_item_by_hash_or_entry(register PyObject *op, PyObject *key,

long hash, PyDictEntry *ep, PyObject *value)

{

register PyDictObject *mp;

register Py_ssize_t n_used;

mp = (PyDictObject *)op;

assert(mp->ma_fill <= mp->ma_mask); /* at least one empty slot */

n_used = mp->ma_used;

Py_INCREF(value);

Py_INCREF(key);

if (ep == NULL) {

if (insertdict(mp, key, hash, value) != 0)

return -1;

}

else {

if (insertdict_by_entry(mp, key, hash, ep, value) != 0)

return -1;

}

/* If we added a key, we can safely resize. Otherwise just return!

* If fill >= 2/3 size, adjust size. Normally, this doubles or

* quaduples the size, but it's also possible for the dict to shrink

* (if ma_fill is much larger than ma_used, meaning a lot of dict

* keys have been * deleted).

*

* Quadrupling the size improves average dictionary sparseness

* (reducing collisions) at the cost of some memory and iteration

* speed (which loops over every possible entry). It also halves

* the number of expensive resize operations in a growing dictionary.

*

* Very large dictionaries (over 50K items) use doubling instead.

* This may help applications with severe memory constraints.

*/

if (!(mp->ma_used > n_used && mp->ma_fill*3 >= (mp->ma_mask+1)*2))

return 0;

return dictresize(mp, (mp->ma_used > 50000 ? 2 : 4) * mp->ma_used);

}

/* CAUTION: PyDict_SetItem() must guarantee that it won't resize the

* dictionary if it's merely replacing the value for an existing key.

* This means that it's safe to loop over a dictionary with PyDict_Next()

* and occasionally replace a value -- but you can't insert new keys or

* remove them.

*/

int

PyDict_SetItem(register PyObject *op, PyObject *key, PyObject *value)

{

register long hash;

if (!PyDict_Check(op)) {

PyErr_BadInternalCall();

return -1;

}

assert(key);

assert(value);

if (PyString_CheckExact(key)) {

hash = ((PyStringObject *)key)->ob_shash;

if (hash == -1)

hash = PyObject_Hash(key);

}

else {

hash = PyObject_Hash(key);

if (hash == -1)

return -1;

}

return dict_set_item_by_hash_or_entry(op, key, hash, NULL, value);

}

static int

delitem_common(PyDictObject *mp, PyDictEntry *ep)

{

PyObject *old_value, *old_key;

old_key = ep->me_key;

Py_INCREF(dummy);

ep->me_key = dummy;

old_value = ep->me_value;

ep->me_value = NULL;

mp->ma_used--;

Py_DECREF(old_value);

Py_DECREF(old_key);

return 0;

}

int

PyDict_DelItem(PyObject *op, PyObject *key)

{

register PyDictObject *mp;

register long hash;

register PyDictEntry *ep;

if (!PyDict_Check(op)) {

PyErr_BadInternalCall();

return -1;

}

assert(key);

if (!PyString_CheckExact(key) ||

(hash = ((PyStringObject *) key)->ob_shash) == -1) {

hash = PyObject_Hash(key);

if (hash == -1)

return -1;

}

mp = (PyDictObject *)op;

ep = (mp->ma_lookup)(mp, key, hash);

if (ep == NULL)

return -1;

if (ep->me_value == NULL) {

set_key_error(key);

return -1;

}

return delitem_common(mp, ep);

}

int

_PyDict_DelItemIf(PyObject *op, PyObject *key,

int (*predicate)(PyObject *value))

{

register PyDictObject *mp;

register long hash;

register PyDictEntry *ep;

int res;

if (!PyDict_Check(op)) {

PyErr_BadInternalCall();

return -1;

}

assert(key);

if (!PyString_CheckExact(key) ||

(hash = ((PyStringObject *) key)->ob_shash) == -1) {

hash = PyObject_Hash(key);

if (hash == -1)

return -1;

}

mp = (PyDictObject *)op;

ep = (mp->ma_lookup)(mp, key, hash);

if (ep == NULL)

return -1;

if (ep->me_value == NULL) {

set_key_error(key);

return -1;

}

res = predicate(ep->me_value);

if (res == -1)

return -1;

if (res > 0)

return delitem_common(mp, ep);

else

return 0;

}

void

PyDict_Clear(PyObject *op)

{

PyDictObject *mp;

PyDictEntry *ep, *table;

int table_is_malloced;

Py_ssize_t fill;

PyDictEntry small_copy[PyDict_MINSIZE];

#ifdef Py_DEBUG

Py_ssize_t i, n;

#endif

if (!PyDict_Check(op))

return;

mp = (PyDictObject *)op;

#ifdef Py_DEBUG

n = mp->ma_mask + 1;

i = 0;

#endif

table = mp->ma_table;

assert(table != NULL);

table_is_malloced = table != mp->ma_smalltable;

/* This is delicate. During the process of clearing the dict,

* decrefs can cause the dict to mutate. To avoid fatal confusion

* (voice of experience), we have to make the dict empty before

* clearing the slots, and never refer to anything via mp->xxx while

* clearing.

*/

fill = mp->ma_fill;

if (table_is_malloced)

EMPTY_TO_MINSIZE(mp);

else if (fill > 0) {

/* It's a small table with something that needs to be cleared.

* Afraid the only safe way is to copy the dict entries into

* another small table first.

*/

memcpy(small_copy, table, sizeof(small_copy));

table = small_copy;

EMPTY_TO_MINSIZE(mp);

}

/* else it's a small table that's already empty */

/* Now we can finally clear things. If C had refcounts, we could

* assert that the refcount on table is 1 now, i.e. that this function

* has unique access to it, so decref side-effects can't alter it.

*/

for (ep = table; fill > 0; ++ep) {

#ifdef Py_DEBUG

assert(i < n);

++i;

#endif

if (ep->me_key) {

--fill;

Py_DECREF(ep->me_key);

Py_XDECREF(ep->me_value);

}

#ifdef Py_DEBUG

else

assert(ep->me_value == NULL);

#endif

}

if (table_is_malloced)

PyMem_DEL(table);