chiark / gitweb /
hashmap: rewrite the implementation
authorMichal Schmidt <mschmidt@redhat.com>
Tue, 14 Oct 2014 23:27:16 +0000 (01:27 +0200)
committerMichal Schmidt <mschmidt@redhat.com>
Thu, 30 Oct 2014 18:50:51 +0000 (19:50 +0100)
This is a rewrite of the hashmap implementation. Its advantage is lower
memory usage.

It uses open addressing (entries are stored in an array, as opposed to
linked lists). Hash collisions are resolved with linear probing and
Robin Hood displacement policy. See the references in hashmap.c.

Some fun empirical findings about hashmap usage in systemd on my laptop:
  - 98 % of allocated hashmaps are Sets.
  - Sets contain 78 % of all entries, plain Hashmaps 17 %, and
    OrderedHashmaps 5 %.
  - 60 % of allocated hashmaps contain only 1 entry.
  - 90 % of allocated hashmaps contain 5 or fewer entries.
  - 75 % of all entries are in hashmaps that use trivial_hash_ops.

Clearly it makes sense to:
  - store entries in distinct entry types. Especially for Sets - their
    entries are the most numerous and they require the least information
    to store an entry.
  - have a way to store small numbers of entries directly in the hashmap
    structs, and only allocate the usual entry arrays when the direct
    storage is full.

The implementation has an optional debugging feature (enabled by
defining the ENABLE_HASHMAP_DEBUG macro), where it:
  - tracks all allocated hashmaps in a linked list so that one can
    easily find them in gdb,
  - tracks which function/line allocated a given hashmap, and
  - checks for invalid mixing of hashmap iteration and modification.

Since entries are not allocated one-by-one anymore, mempools are not
used for entries. Originally I meant to drop mempools entirely, but it's
still worth it to use them for the hashmap structs. My testing indicates
that it makes loading of units about 5 % faster (a test with 10000 units
where more than 200000 hashmaps are allocated - pure malloc: 449±4 ms,
mempools: 427±7 ms).

Here are some memory usage numbers, taken on my laptop with a more or
less normal Fedora setup after booting with SELinux disabled (SELinux
increases systemd's memory usage significantly):

systemd (PID 1)                            Original   New    Change
dirty memory (from pmap -x 1) [KiB]            2152  1264     -41 %
total heap allocations (from gdb-heap) [KiB]   1623   756     -53 %

Makefile.am
src/shared/hashmap.c
src/shared/hashmap.h
src/shared/set.c [deleted file]
src/shared/set.h

index 69d598b..946e46f 100644 (file)
@@ -774,7 +774,6 @@ libsystemd_shared_la_SOURCES = \
        src/shared/hashmap.h \
        src/shared/siphash24.c \
        src/shared/siphash24.h \
-       src/shared/set.c \
        src/shared/set.h \
        src/shared/fdset.c \
        src/shared/fdset.h \
index 6f5f820..2bc3b38 100644 (file)
@@ -4,6 +4,7 @@
   This file is part of systemd.
 
   Copyright 2010 Lennart Poettering
+  Copyright 2014 Michal Schmidt
 
   systemd is free software; you can redistribute it and/or modify it
   under the terms of the GNU Lesser General Public License as published by
 
 #include "util.h"
 #include "hashmap.h"
+#include "set.h"
 #include "macro.h"
 #include "siphash24.h"
+#include "strv.h"
+#include "list.h"
 #include "mempool.h"
 
-#define INITIAL_N_BUCKETS 31
-
-struct hashmap_entry {
+/*
+ * Implementation of hashmaps.
+ * Addressing: open
+ *   - uses less RAM compared to closed addressing (chaining), because
+ *     our entries are small (especially in Sets, which tend to contain
+ *     the majority of entries in systemd).
+ * Collision resolution: Robin Hood
+ *   - tends to equalize displacement of entries from their optimal buckets.
+ * Probe sequence: linear
+ *   - though theoretically worse than random probing/uniform hashing/double
+ *     hashing, it is good for cache locality.
+ *
+ * References:
+ * Celis, P. 1986. Robin Hood Hashing.
+ * Ph.D. Dissertation. University of Waterloo, Waterloo, Ont., Canada, Canada.
+ * https://cs.uwaterloo.ca/research/tr/1986/CS-86-14.pdf
+ * - The results are derived for random probing. Suggests deletion with
+ *   tombstones and two mean-centered search methods. None of that works
+ *   well for linear probing.
+ *
+ * Janson, S. 2005. Individual displacements for linear probing hashing with different insertion policies.
+ * ACM Trans. Algorithms 1, 2 (October 2005), 177-213.
+ * DOI=10.1145/1103963.1103964 http://doi.acm.org/10.1145/1103963.1103964
+ * http://www.math.uu.se/~svante/papers/sj157.pdf
+ * - Applies to Robin Hood with linear probing. Contains remarks on
+ *   the unsuitability of mean-centered search with linear probing.
+ *
+ * Viola, A. 2005. Exact distribution of individual displacements in linear probing hashing.
+ * ACM Trans. Algorithms 1, 2 (October 2005), 214-242.
+ * DOI=10.1145/1103963.1103965 http://doi.acm.org/10.1145/1103963.1103965
+ * - Similar to Janson. Note that Viola writes about C_{m,n} (number of probes
+ *   in a successful search), and Janson writes about displacement. C = d + 1.
+ *
+ * Goossaert, E. 2013. Robin Hood hashing: backward shift deletion.
+ * http://codecapsule.com/2013/11/17/robin-hood-hashing-backward-shift-deletion/
+ * - Explanation of backward shift deletion with pictures.
+ *
+ * Khuong, P. 2013. The Other Robin Hood Hashing.
+ * http://www.pvk.ca/Blog/2013/11/26/the-other-robin-hood-hashing/
+ * - Short summary of random vs. linear probing, and tombstones vs. backward shift.
+ */
+
+/*
+ * XXX Ideas for improvement:
+ * For unordered hashmaps, randomize iteration order, similarly to Perl:
+ * http://blog.booking.com/hardening-perls-hash-function.html
+ */
+
+/* INV_KEEP_FREE = 1 / (1 - max_load_factor)
+ * e.g. 1 / (1 - 0.8) = 5 ... keep one fifth of the buckets free. */
+#define INV_KEEP_FREE            5U
+
+/* Fields common to entries of all hashmap/set types */
+struct hashmap_base_entry {
         const void *key;
+};
+
+/* Entry types for specific hashmap/set types
+ * hashmap_base_entry must be at the beginning of each entry struct. */
+
+struct plain_hashmap_entry {
+        struct hashmap_base_entry b;
         void *value;
-        struct hashmap_entry *bucket_next, *bucket_previous;
-        struct hashmap_entry *iterate_next, *iterate_previous;
 };
 
-struct Hashmap {
-        const struct hash_ops *hash_ops;
+struct ordered_hashmap_entry {
+        struct plain_hashmap_entry p;
+        unsigned iterate_next, iterate_previous;
+};
 
-        struct hashmap_entry *iterate_list_head, *iterate_list_tail;
+struct set_entry {
+        struct hashmap_base_entry b;
+};
 
-        struct hashmap_entry ** buckets;
-        unsigned n_buckets, n_entries;
+/* In several functions it is advantageous to have the hash table extended
+ * virtually by a couple of additional buckets. We reserve special index values
+ * for these "swap" buckets. */
+#define _IDX_SWAP_BEGIN     (UINT_MAX - 3)
+#define IDX_PUT             (_IDX_SWAP_BEGIN + 0)
+#define IDX_TMP             (_IDX_SWAP_BEGIN + 1)
+#define _IDX_SWAP_END       (_IDX_SWAP_BEGIN + 2)
 
-        uint8_t hash_key[HASH_KEY_SIZE];
-        bool from_pool:1;
+#define IDX_FIRST           (UINT_MAX - 1) /* special index for freshly initialized iterators */
+#define IDX_NIL             UINT_MAX       /* special index value meaning "none" or "end" */
+
+assert_cc(IDX_FIRST == _IDX_SWAP_END);
+assert_cc(IDX_FIRST == _IDX_ITERATOR_FIRST);
+
+/* Storage space for the "swap" buckets.
+ * All entry types can fit into a ordered_hashmap_entry. */
+struct swap_entries {
+        struct ordered_hashmap_entry e[_IDX_SWAP_END - _IDX_SWAP_BEGIN];
 };
 
-struct hashmap_tile {
-        Hashmap h;
-        struct hashmap_entry *initial_buckets[INITIAL_N_BUCKETS];
+/* Distance from Initial Bucket */
+typedef uint8_t dib_raw_t;
+#define DIB_RAW_OVERFLOW ((dib_raw_t)0xfdU) /* indicates DIB value is greater than representable */
+#define DIB_RAW_REHASH   ((dib_raw_t)0xfeU) /* entry yet to be rehashed during in-place resize */
+#define DIB_RAW_FREE     ((dib_raw_t)0xffU) /* a free bucket */
+#define DIB_RAW_INIT   ((char)DIB_RAW_FREE) /* a byte to memset a DIB store with when initializing */
+
+#define DIB_FREE UINT_MAX
+
+#ifdef ENABLE_HASHMAP_DEBUG
+struct hashmap_debug_info {
+        LIST_FIELDS(struct hashmap_debug_info, debug_list);
+        unsigned max_entries;  /* high watermark of n_entries */
+
+        /* who allocated this hashmap */
+        int line;
+        const char *file;
+        const char *func;
+
+        /* fields to detect modification while iterating */
+        unsigned put_count;    /* counts puts into the hashmap */
+        unsigned rem_count;    /* counts removals from hashmap */
+        unsigned last_rem_idx; /* remembers last removal index */
 };
 
-static DEFINE_MEMPOOL(hashmap_pool, struct hashmap_tile, 8);
-static DEFINE_MEMPOOL(hashmap_entry_pool, struct hashmap_entry, 64);
+/* Tracks all existing hashmaps. Get at it from gdb. See sd_dump_hashmaps.py */
+static LIST_HEAD(struct hashmap_debug_info, hashmap_debug_list);
 
-#ifdef VALGRIND
+#define HASHMAP_DEBUG_FIELDS struct hashmap_debug_info debug;
 
-__attribute__((destructor)) static void cleanup_pools(void) {
-        /* Be nice to valgrind */
+#else /* !ENABLE_HASHMAP_DEBUG */
+#define HASHMAP_DEBUG_FIELDS
+#endif /* ENABLE_HASHMAP_DEBUG */
 
-        mempool_drop(&hashmap_entry_pool);
-        mempool_drop(&hashmap_pool);
-}
+enum HashmapType {
+        HASHMAP_TYPE_PLAIN,
+        HASHMAP_TYPE_ORDERED,
+        HASHMAP_TYPE_SET,
+        _HASHMAP_TYPE_MAX
+};
 
-#endif
+struct _packed_ indirect_storage {
+        char    *storage;                  /* where buckets and DIBs are stored */
+        uint8_t  hash_key[HASH_KEY_SIZE];  /* hash key; changes during resize */
+
+        unsigned n_entries;                /* number of stored entries */
+        unsigned n_buckets;                /* number of buckets */
+
+        unsigned idx_lowest_entry;         /* Index below which all buckets are free.
+                                              Makes "while(hashmap_steal_first())" loops
+                                              O(n) instead of O(n^2) for unordered hashmaps. */
+        uint8_t  _pad[3];                  /* padding for the whole HashmapBase */
+        /* The bitfields in HashmapBase complete the alignment of the whole thing. */
+};
+
+struct direct_storage {
+        /* This gives us 39 bytes on 64bit, or 35 bytes on 32bit.
+         * That's room for 4 set_entries + 4 DIB bytes + 3 unused bytes on 64bit,
+         *              or 7 set_entries + 7 DIB bytes + 0 unused bytes on 32bit. */
+        char storage[sizeof(struct indirect_storage)];
+};
+
+#define DIRECT_BUCKETS(entry_t) \
+        (sizeof(struct direct_storage) / (sizeof(entry_t) + sizeof(dib_raw_t)))
+
+/* We should be able to store at least one entry directly. */
+assert_cc(DIRECT_BUCKETS(struct ordered_hashmap_entry) >= 1);
+
+/* We have 3 bits for n_direct_entries. */
+assert_cc(DIRECT_BUCKETS(struct set_entry) < (1 << 3));
+
+/* Hashmaps with directly stored entries all use this shared hash key.
+ * It's no big deal if the key is guessed, because there can be only
+ * a handful of directly stored entries in a hashmap. When a hashmap
+ * outgrows direct storage, it gets its own key for indirect storage. */
+static uint8_t shared_hash_key[HASH_KEY_SIZE];
+static bool shared_hash_key_initialized;
+
+/* Fields that all hashmap/set types must have */
+struct HashmapBase {
+        const struct hash_ops *hash_ops;  /* hash and compare ops to use */
+
+        union _packed_ {
+                struct indirect_storage indirect; /* if  has_indirect */
+                struct direct_storage direct;     /* if !has_indirect */
+        };
+
+        enum HashmapType type:2;     /* HASHMAP_TYPE_* */
+        bool has_indirect:1;         /* whether indirect storage is used */
+        unsigned n_direct_entries:3; /* Number of entries in direct storage.
+                                      * Only valid if !has_indirect. */
+        bool from_pool:1;            /* whether was allocated from mempool */
+        HASHMAP_DEBUG_FIELDS         /* optional hashmap_debug_info */
+};
+
+/* Specific hash types
+ * HashmapBase must be at the beginning of each hashmap struct. */
+
+struct Hashmap {
+        struct HashmapBase b;
+};
+
+struct OrderedHashmap {
+        struct HashmapBase b;
+        unsigned iterate_list_head, iterate_list_tail;
+};
+
+struct Set {
+        struct HashmapBase b;
+};
+
+DEFINE_MEMPOOL(hashmap_pool,         Hashmap,        8);
+DEFINE_MEMPOOL(ordered_hashmap_pool, OrderedHashmap, 8);
+/* No need for a separate Set pool */
+assert_cc(sizeof(Hashmap) == sizeof(Set));
+
+struct hashmap_type_info {
+        size_t head_size;
+        size_t entry_size;
+        struct mempool *mempool;
+        unsigned n_direct_buckets;
+};
+
+static const struct hashmap_type_info hashmap_type_info[_HASHMAP_TYPE_MAX] = {
+        [HASHMAP_TYPE_PLAIN] = {
+                .head_size        = sizeof(Hashmap),
+                .entry_size       = sizeof(struct plain_hashmap_entry),
+                .mempool          = &hashmap_pool,
+                .n_direct_buckets = DIRECT_BUCKETS(struct plain_hashmap_entry),
+        },
+        [HASHMAP_TYPE_ORDERED] = {
+                .head_size        = sizeof(OrderedHashmap),
+                .entry_size       = sizeof(struct ordered_hashmap_entry),
+                .mempool          = &ordered_hashmap_pool,
+                .n_direct_buckets = DIRECT_BUCKETS(struct ordered_hashmap_entry),
+        },
+        [HASHMAP_TYPE_SET] = {
+                .head_size        = sizeof(Set),
+                .entry_size       = sizeof(struct set_entry),
+                .mempool          = &hashmap_pool,
+                .n_direct_buckets = DIRECT_BUCKETS(struct set_entry),
+        },
+};
 
 unsigned long string_hash_func(const void *p, const uint8_t hash_key[HASH_KEY_SIZE]) {
         uint64_t u;
@@ -138,9 +340,44 @@ const struct hash_ops devt_hash_ops = {
 };
 #endif
 
-static unsigned bucket_hash(Hashmap *h, const void *p) {
-        return (unsigned) (h->hash_ops->hash(p, h->hash_key) % h->n_buckets);
+static unsigned n_buckets(HashmapBase *h) {
+        return h->has_indirect ? h->indirect.n_buckets
+                               : hashmap_type_info[h->type].n_direct_buckets;
+}
+
+static unsigned n_entries(HashmapBase *h) {
+        return h->has_indirect ? h->indirect.n_entries
+                               : h->n_direct_entries;
+}
+
+static void n_entries_inc(HashmapBase *h) {
+        if (h->has_indirect)
+                h->indirect.n_entries++;
+        else
+                h->n_direct_entries++;
+}
+
+static void n_entries_dec(HashmapBase *h) {
+        if (h->has_indirect)
+                h->indirect.n_entries--;
+        else
+                h->n_direct_entries--;
+}
+
+static char *storage_ptr(HashmapBase *h) {
+        return h->has_indirect ? h->indirect.storage
+                               : h->direct.storage;
+}
+
+static uint8_t *hash_key(HashmapBase *h) {
+        return h->has_indirect ? h->indirect.hash_key
+                               : shared_hash_key;
+}
+
+static unsigned base_bucket_hash(HashmapBase *h, const void *p) {
+        return (unsigned) (h->hash_ops->hash(p, hash_key(h)) % n_buckets(h));
 }
+#define bucket_hash(h, p) base_bucket_hash(HASHMAP_BASE(h), p)
 
 static void get_hash_key(uint8_t hash_key[HASH_KEY_SIZE], bool reuse_is_ok) {
         static uint8_t current[HASH_KEY_SIZE];
@@ -161,147 +398,484 @@ static void get_hash_key(uint8_t hash_key[HASH_KEY_SIZE], bool reuse_is_ok) {
         memcpy(hash_key, current, sizeof(current));
 }
 
-Hashmap *hashmap_new(const struct hash_ops *hash_ops) {
-        bool b;
-        struct hashmap_tile *ht;
-        Hashmap *h;
+static struct hashmap_base_entry *bucket_at(HashmapBase *h, unsigned idx) {
+        return (struct hashmap_base_entry*)
+                (storage_ptr(h) + idx * hashmap_type_info[h->type].entry_size);
+}
+
+static struct plain_hashmap_entry *plain_bucket_at(Hashmap *h, unsigned idx) {
+        return (struct plain_hashmap_entry*) bucket_at(HASHMAP_BASE(h), idx);
+}
+
+static struct ordered_hashmap_entry *ordered_bucket_at(OrderedHashmap *h, unsigned idx) {
+        return (struct ordered_hashmap_entry*) bucket_at(HASHMAP_BASE(h), idx);
+}
 
-        b = is_main_thread();
+static struct set_entry *set_bucket_at(Set *h, unsigned idx) {
+        return (struct set_entry*) bucket_at(HASHMAP_BASE(h), idx);
+}
 
-        if (b) {
-                ht = mempool_alloc_tile(&hashmap_pool);
-                if (!ht)
-                        return NULL;
+static struct ordered_hashmap_entry *bucket_at_swap(struct swap_entries *swap, unsigned idx) {
+        return &swap->e[idx - _IDX_SWAP_BEGIN];
+}
 
-                memzero(ht, sizeof(struct hashmap_tile));
-        } else {
-                ht = malloc0(sizeof(struct hashmap_tile));
+/* Returns a pointer to the bucket at index idx.
+ * Understands real indexes and swap indexes, hence "_virtual". */
+static struct hashmap_base_entry *bucket_at_virtual(HashmapBase *h, struct swap_entries *swap,
+                                                    unsigned idx) {
+        if (idx < _IDX_SWAP_BEGIN)
+                return bucket_at(h, idx);
+
+        if (idx < _IDX_SWAP_END)
+                return &bucket_at_swap(swap, idx)->p.b;
+
+        assert_not_reached("Invalid index");
+}
+
+static dib_raw_t *dib_raw_ptr(HashmapBase *h) {
+        return (dib_raw_t*)
+                (storage_ptr(h) + hashmap_type_info[h->type].entry_size * n_buckets(h));
+}
+
+static unsigned bucket_distance(HashmapBase *h, unsigned idx, unsigned from) {
+        return idx >= from ? idx - from
+                           : n_buckets(h) + idx - from;
+}
+
+static unsigned bucket_calculate_dib(HashmapBase *h, unsigned idx, dib_raw_t raw_dib) {
+        unsigned initial_bucket;
+
+        if (raw_dib == DIB_RAW_FREE)
+                return DIB_FREE;
+
+        if (_likely_(raw_dib < DIB_RAW_OVERFLOW))
+                return raw_dib;
+
+        /*
+         * Having an overflow DIB value is very unlikely. The hash function
+         * would have to be bad. For example, in a table of size 2^24 filled
+         * to load factor 0.9 the maximum observed DIB is only about 60.
+         * In theory (assuming I used Maxima correctly), for an infinite size
+         * hash table with load factor 0.8 the probability of a given entry
+         * having DIB > 40 is 1.9e-8.
+         * This returns the correct DIB value by recomputing the hash value in
+         * the unlikely case. XXX Hitting this case could be a hint to rehash.
+         */
+        initial_bucket = bucket_hash(h, bucket_at(h, idx)->key);
+        return bucket_distance(h, idx, initial_bucket);
+}
+
+static void bucket_set_dib(HashmapBase *h, unsigned idx, unsigned dib) {
+        dib_raw_ptr(h)[idx] = dib != DIB_FREE ? MIN(dib, DIB_RAW_OVERFLOW) : DIB_RAW_FREE;
+}
+
+static unsigned skip_free_buckets(HashmapBase *h, unsigned idx) {
+        dib_raw_t *dibs;
+
+        dibs = dib_raw_ptr(h);
+
+        for ( ; idx < n_buckets(h); idx++)
+                if (dibs[idx] != DIB_RAW_FREE)
+                        return idx;
+
+        return IDX_NIL;
+}
+
+static void bucket_mark_free(HashmapBase *h, unsigned idx) {
+        memset(bucket_at(h, idx), 0, hashmap_type_info[h->type].entry_size);
+        bucket_set_dib(h, idx, DIB_FREE);
+}
+
+static void bucket_move_entry(HashmapBase *h, struct swap_entries *swap,
+                              unsigned from, unsigned to) {
+        struct hashmap_base_entry *e_from, *e_to;
+
+        assert(from != to);
 
-                if (!ht)
-                        return NULL;
+        e_from = bucket_at_virtual(h, swap, from);
+        e_to   = bucket_at_virtual(h, swap, to);
+
+        memcpy(e_to, e_from, hashmap_type_info[h->type].entry_size);
+
+        if (h->type == HASHMAP_TYPE_ORDERED) {
+                OrderedHashmap *lh = (OrderedHashmap*) h;
+                struct ordered_hashmap_entry *le, *le_to;
+
+                le_to = (struct ordered_hashmap_entry*) e_to;
+
+                if (le_to->iterate_next != IDX_NIL) {
+                        le = (struct ordered_hashmap_entry*)
+                             bucket_at_virtual(h, swap, le_to->iterate_next);
+                        le->iterate_previous = to;
+                }
+
+                if (le_to->iterate_previous != IDX_NIL) {
+                        le = (struct ordered_hashmap_entry*)
+                             bucket_at_virtual(h, swap, le_to->iterate_previous);
+                        le->iterate_next = to;
+                }
+
+                if (lh->iterate_list_head == from)
+                        lh->iterate_list_head = to;
+                if (lh->iterate_list_tail == from)
+                        lh->iterate_list_tail = to;
         }
+}
 
-        h = &ht->h;
-        h->hash_ops = hash_ops ? hash_ops : &trivial_hash_ops;
+static unsigned next_idx(HashmapBase *h, unsigned idx) {
+        return (idx + 1U) % n_buckets(h);
+}
 
-        h->n_buckets = INITIAL_N_BUCKETS;
-        h->n_entries = 0;
-        h->iterate_list_head = h->iterate_list_tail = NULL;
+static unsigned prev_idx(HashmapBase *h, unsigned idx) {
+        return (n_buckets(h) + idx - 1U) % n_buckets(h);
+}
 
-        h->buckets = ht->initial_buckets;
+static void *entry_value(HashmapBase *h, struct hashmap_base_entry *e) {
+        switch (h->type) {
 
-        h->from_pool = b;
+        case HASHMAP_TYPE_PLAIN:
+        case HASHMAP_TYPE_ORDERED:
+                return ((struct plain_hashmap_entry*)e)->value;
 
-        get_hash_key(h->hash_key, true);
+        case HASHMAP_TYPE_SET:
+                return (void*) e->key;
 
-        return h;
+        default:
+                assert_not_reached("Unknown hashmap type");
+        }
 }
 
-int hashmap_ensure_allocated(Hashmap **h, const struct hash_ops *hash_ops) {
-        Hashmap *q;
+static void base_remove_entry(HashmapBase *h, unsigned idx) {
+        unsigned left, right, prev, dib;
+        dib_raw_t raw_dib, *dibs;
 
-        assert(h);
+        dibs = dib_raw_ptr(h);
+        assert(dibs[idx] != DIB_RAW_FREE);
 
-        if (*h)
-                return 0;
+#ifdef ENABLE_HASHMAP_DEBUG
+        h->debug.rem_count++;
+        h->debug.last_rem_idx = idx;
+#endif
 
-        q = hashmap_new(hash_ops);
-        if (!q)
-                return -ENOMEM;
+        left = idx;
+        /* Find the stop bucket ("right"). It is either free or has DIB == 0. */
+        for (right = next_idx(h, left); ; right = next_idx(h, right)) {
+                raw_dib = dibs[right];
+                if (raw_dib == 0 || raw_dib == DIB_RAW_FREE)
+                        break;
+
+                /* The buckets are not supposed to be all occupied and with DIB > 0.
+                 * That would mean we could make everyone better off by shifting them
+                 * backward. This scenario is impossible. */
+                assert(left != right);
+        }
 
-        *h = q;
-        return 0;
+        if (h->type == HASHMAP_TYPE_ORDERED) {
+                OrderedHashmap *lh = (OrderedHashmap*) h;
+                struct ordered_hashmap_entry *le = ordered_bucket_at(lh, idx);
+
+                if (le->iterate_next != IDX_NIL)
+                        ordered_bucket_at(lh, le->iterate_next)->iterate_previous = le->iterate_previous;
+                else
+                        lh->iterate_list_tail = le->iterate_previous;
+
+                if (le->iterate_previous != IDX_NIL)
+                        ordered_bucket_at(lh, le->iterate_previous)->iterate_next = le->iterate_next;
+                else
+                        lh->iterate_list_head = le->iterate_next;
+        }
+
+        /* Now shift all buckets in the interval (left, right) one step backwards */
+        for (prev = left, left = next_idx(h, left); left != right;
+             prev = left, left = next_idx(h, left)) {
+                dib = bucket_calculate_dib(h, left, dibs[left]);
+                assert(dib != 0);
+                bucket_move_entry(h, NULL, left, prev);
+                bucket_set_dib(h, prev, dib - 1);
+        }
+
+        bucket_mark_free(h, prev);
+        n_entries_dec(h);
 }
+#define remove_entry(h, idx) base_remove_entry(HASHMAP_BASE(h), idx)
+
+static unsigned hashmap_iterate_in_insertion_order(OrderedHashmap *h, Iterator *i) {
+        struct ordered_hashmap_entry *e;
+        unsigned idx;
 
-static void link_entry(Hashmap *h, struct hashmap_entry *e, unsigned hash) {
         assert(h);
-        assert(e);
-
-        /* Insert into hash table */
-        e->bucket_next = h->buckets[hash];
-        e->bucket_previous = NULL;
-        if (h->buckets[hash])
-                h->buckets[hash]->bucket_previous = e;
-        h->buckets[hash] = e;
-
-        /* Insert into iteration list */
-        e->iterate_previous = h->iterate_list_tail;
-        e->iterate_next = NULL;
-        if (h->iterate_list_tail) {
-                assert(h->iterate_list_head);
-                h->iterate_list_tail->iterate_next = e;
+        assert(i);
+
+        if (i->idx == IDX_NIL)
+                goto at_end;
+
+        if (i->idx == IDX_FIRST && h->iterate_list_head == IDX_NIL)
+                goto at_end;
+
+        if (i->idx == IDX_FIRST) {
+                idx = h->iterate_list_head;
+                e = ordered_bucket_at(h, idx);
         } else {
-                assert(!h->iterate_list_head);
-                h->iterate_list_head = e;
+                idx = i->idx;
+                e = ordered_bucket_at(h, idx);
+                /*
+                 * We allow removing the current entry while iterating, but removal may cause
+                 * a backward shift. The next entry may thus move one bucket to the left.
+                 * To detect when it happens, we remember the key pointer of the entry we were
+                 * going to iterate next. If it does not match, there was a backward shift.
+                 */
+                if (e->p.b.key != i->next_key) {
+                        idx = prev_idx(HASHMAP_BASE(h), idx);
+                        e = ordered_bucket_at(h, idx);
+                }
+                assert(e->p.b.key == i->next_key);
         }
-        h->iterate_list_tail = e;
 
-        h->n_entries++;
-        assert(h->n_entries >= 1);
+#ifdef ENABLE_HASHMAP_DEBUG
+        i->prev_idx = idx;
+#endif
+
+        if (e->iterate_next != IDX_NIL) {
+                struct ordered_hashmap_entry *n;
+                i->idx = e->iterate_next;
+                n = ordered_bucket_at(h, i->idx);
+                i->next_key = n->p.b.key;
+        } else
+                i->idx = IDX_NIL;
+
+        return idx;
+
+at_end:
+        i->idx = IDX_NIL;
+        return IDX_NIL;
 }
 
-static void unlink_entry(Hashmap *h, struct hashmap_entry *e, unsigned hash) {
+static unsigned hashmap_iterate_in_internal_order(HashmapBase *h, Iterator *i) {
+        unsigned idx;
+
         assert(h);
-        assert(e);
+        assert(i);
 
-        /* Remove from iteration list */
-        if (e->iterate_next)
-                e->iterate_next->iterate_previous = e->iterate_previous;
-        else
-                h->iterate_list_tail = e->iterate_previous;
+        if (i->idx == IDX_NIL)
+                goto at_end;
 
-        if (e->iterate_previous)
-                e->iterate_previous->iterate_next = e->iterate_next;
-        else
-                h->iterate_list_head = e->iterate_next;
+        if (i->idx == IDX_FIRST) {
+                /* fast forward to the first occupied bucket */
+                if (h->has_indirect) {
+                        i->idx = skip_free_buckets(h, h->indirect.idx_lowest_entry);
+                        h->indirect.idx_lowest_entry = i->idx;
+                } else
+                        i->idx = skip_free_buckets(h, 0);
+
+                if (i->idx == IDX_NIL)
+                        goto at_end;
+        } else {
+                struct hashmap_base_entry *e;
+
+                assert(i->idx > 0);
 
-        /* Remove from hash table bucket list */
-        if (e->bucket_next)
-                e->bucket_next->bucket_previous = e->bucket_previous;
+                e = bucket_at(h, i->idx);
+                /*
+                 * We allow removing the current entry while iterating, but removal may cause
+                 * a backward shift. The next entry may thus move one bucket to the left.
+                 * To detect when it happens, we remember the key pointer of the entry we were
+                 * going to iterate next. If it does not match, there was a backward shift.
+                 */
+                if (e->key != i->next_key)
+                        e = bucket_at(h, --i->idx);
 
-        if (e->bucket_previous)
-                e->bucket_previous->bucket_next = e->bucket_next;
+                assert(e->key == i->next_key);
+        }
+
+        idx = i->idx;
+#ifdef ENABLE_HASHMAP_DEBUG
+        i->prev_idx = idx;
+#endif
+
+        i->idx = skip_free_buckets(h, i->idx + 1);
+        if (i->idx != IDX_NIL)
+                i->next_key = bucket_at(h, i->idx)->key;
         else
-                h->buckets[hash] = e->bucket_next;
+                i->idx = IDX_NIL;
+
+        return idx;
 
-        assert(h->n_entries >= 1);
-        h->n_entries--;
+at_end:
+        i->idx = IDX_NIL;
+        return IDX_NIL;
 }
 
-static void remove_entry(Hashmap *h, struct hashmap_entry *e) {
-        unsigned hash;
+static unsigned hashmap_iterate_entry(HashmapBase *h, Iterator *i) {
+        if (!h) {
+                i->idx = IDX_NIL;
+                return IDX_NIL;
+        }
 
-        assert(h);
-        assert(e);
+#ifdef ENABLE_HASHMAP_DEBUG
+        if (i->idx == IDX_FIRST) {
+                i->put_count = h->debug.put_count;
+                i->rem_count = h->debug.rem_count;
+        } else {
+                /* While iterating, must not add any new entries */
+                assert(i->put_count == h->debug.put_count);
+                /* ... or remove entries other than the current one */
+                assert(i->rem_count == h->debug.rem_count ||
+                       (i->rem_count == h->debug.rem_count - 1 &&
+                        i->prev_idx == h->debug.last_rem_idx));
+                /* Reset our removals counter */
+                i->rem_count = h->debug.rem_count;
+        }
+#endif
 
-        hash = bucket_hash(h, e->key);
-        unlink_entry(h, e, hash);
+        return h->type == HASHMAP_TYPE_ORDERED ? hashmap_iterate_in_insertion_order((OrderedHashmap*) h, i)
+                                               : hashmap_iterate_in_internal_order(h, i);
+}
 
-        if (h->from_pool)
-                mempool_free_tile(&hashmap_entry_pool, e);
-        else
-                free(e);
+void *internal_hashmap_iterate(HashmapBase *h, Iterator *i, const void **key) {
+        struct hashmap_base_entry *e;
+        void *data;
+        unsigned idx;
+
+        idx = hashmap_iterate_entry(h, i);
+        if (idx == IDX_NIL) {
+                if (key)
+                        *key = NULL;
+
+                return NULL;
+        }
+
+        e = bucket_at(h, idx);
+        data = entry_value(h, e);
+        if (key)
+                *key = e->key;
+
+        return data;
 }
 
-void hashmap_free(Hashmap*h) {
+void *set_iterate(Set *s, Iterator *i) {
+        return internal_hashmap_iterate(HASHMAP_BASE(s), i, NULL);
+}
 
-        /* Free the hashmap, but nothing in it */
+#define HASHMAP_FOREACH_IDX(idx, h, i) \
+        for ((i) = ITERATOR_FIRST, (idx) = hashmap_iterate_entry((h), &(i)); \
+             (idx != IDX_NIL); \
+             (idx) = hashmap_iterate_entry((h), &(i)))
+
+static void reset_direct_storage(HashmapBase *h) {
+        const struct hashmap_type_info *hi = &hashmap_type_info[h->type];
+        void *p;
+
+        assert(!h->has_indirect);
+
+        p = mempset(h->direct.storage, 0, hi->entry_size * hi->n_direct_buckets);
+        memset(p, DIB_RAW_INIT, sizeof(dib_raw_t) * hi->n_direct_buckets);
+}
+
+static struct HashmapBase *hashmap_base_new(const struct hash_ops *hash_ops, enum HashmapType type  HASHMAP_DEBUG_PARAMS) {
+        HashmapBase *h;
+        const struct hashmap_type_info *hi = &hashmap_type_info[type];
+        bool use_pool;
+
+        use_pool = is_main_thread();
+
+        h = use_pool ? mempool_alloc0_tile(hi->mempool) : malloc0(hi->head_size);
 
         if (!h)
-                return;
+                return NULL;
+
+        h->type = type;
+        h->from_pool = use_pool;
+        h->hash_ops = hash_ops ? hash_ops : &trivial_hash_ops;
+
+        if (type == HASHMAP_TYPE_ORDERED) {
+                OrderedHashmap *lh = (OrderedHashmap*)h;
+                lh->iterate_list_head = lh->iterate_list_tail = IDX_NIL;
+        }
+
+        reset_direct_storage(h);
 
-        hashmap_clear(h);
+        if (!shared_hash_key_initialized) {
+                random_bytes(shared_hash_key, sizeof(shared_hash_key));
+                shared_hash_key_initialized= true;
+        }
+
+#ifdef ENABLE_HASHMAP_DEBUG
+        LIST_PREPEND(debug_list, hashmap_debug_list, &h->debug);
+        h->debug.func = func;
+        h->debug.file = file;
+        h->debug.line = line;
+#endif
+
+        return h;
+}
 
-        if (h->buckets != (struct hashmap_entry**) ((uint8_t*) h + ALIGN(sizeof(Hashmap))))
-                free(h->buckets);
+Hashmap *internal_hashmap_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
+        return (Hashmap*)        hashmap_base_new(hash_ops, HASHMAP_TYPE_PLAIN    HASHMAP_DEBUG_PASS_ARGS);
+}
+
+OrderedHashmap *internal_ordered_hashmap_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
+        return (OrderedHashmap*) hashmap_base_new(hash_ops, HASHMAP_TYPE_ORDERED  HASHMAP_DEBUG_PASS_ARGS);
+}
+
+Set *internal_set_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
+        return (Set*)            hashmap_base_new(hash_ops, HASHMAP_TYPE_SET      HASHMAP_DEBUG_PASS_ARGS);
+}
+
+static int hashmap_base_ensure_allocated(HashmapBase **h, const struct hash_ops *hash_ops,
+                                         enum HashmapType type  HASHMAP_DEBUG_PARAMS) {
+        HashmapBase *q;
+
+        assert(h);
+
+        if (*h)
+                return 0;
+
+        q = hashmap_base_new(hash_ops, type  HASHMAP_DEBUG_PASS_ARGS);
+        if (!q)
+                return -ENOMEM;
+
+        *h = q;
+        return 0;
+}
+
+int internal_hashmap_ensure_allocated(Hashmap **h, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
+        return hashmap_base_ensure_allocated((HashmapBase**)h, hash_ops, HASHMAP_TYPE_PLAIN    HASHMAP_DEBUG_PASS_ARGS);
+}
+
+int internal_ordered_hashmap_ensure_allocated(OrderedHashmap **h, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
+        return hashmap_base_ensure_allocated((HashmapBase**)h, hash_ops, HASHMAP_TYPE_ORDERED  HASHMAP_DEBUG_PASS_ARGS);
+}
+
+int internal_set_ensure_allocated(Set **s, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS) {
+        return hashmap_base_ensure_allocated((HashmapBase**)s, hash_ops, HASHMAP_TYPE_SET      HASHMAP_DEBUG_PASS_ARGS);
+}
+
+static void hashmap_free_no_clear(HashmapBase *h) {
+        assert(!h->has_indirect);
+        assert(!h->n_direct_entries);
+
+#ifdef ENABLE_HASHMAP_DEBUG
+        LIST_REMOVE(debug_list, hashmap_debug_list, &h->debug);
+#endif
 
         if (h->from_pool)
-                mempool_free_tile(&hashmap_pool, container_of(h, struct hashmap_tile, h));
+                mempool_free_tile(hashmap_type_info[h->type].mempool, h);
         else
                 free(h);
 }
 
-void hashmap_free_free(Hashmap *h) {
+void internal_hashmap_free(HashmapBase *h) {
+
+        /* Free the hashmap, but nothing in it */
+
+        if (!h)
+                return;
+
+        internal_hashmap_clear(h);
+        hashmap_free_no_clear(h);
+}
+
+void internal_hashmap_free_free(HashmapBase *h) {
 
         /* Free the hashmap and all data objects in it, but not the
          * keys */
@@ -309,8 +883,8 @@ void hashmap_free_free(Hashmap *h) {
         if (!h)
                 return;
 
-        hashmap_clear_free(h);
-        hashmap_free(h);
+        internal_hashmap_clear_free(h);
+        hashmap_free_no_clear(h);
 }
 
 void hashmap_free_free_free(Hashmap *h) {
@@ -321,258 +895,499 @@ void hashmap_free_free_free(Hashmap *h) {
                 return;
 
         hashmap_clear_free_free(h);
-        hashmap_free(h);
+        hashmap_free_no_clear(HASHMAP_BASE(h));
 }
 
-void hashmap_clear(Hashmap *h) {
+void internal_hashmap_clear(HashmapBase *h) {
         if (!h)
                 return;
 
-        while (h->iterate_list_head)
-                remove_entry(h, h->iterate_list_head);
+        if (h->has_indirect) {
+                free(h->indirect.storage);
+                h->has_indirect = false;
+        }
+
+        h->n_direct_entries = 0;
+        reset_direct_storage(h);
+
+        if (h->type == HASHMAP_TYPE_ORDERED) {
+                OrderedHashmap *lh = (OrderedHashmap*) h;
+                lh->iterate_list_head = lh->iterate_list_tail = IDX_NIL;
+        }
 }
 
-void hashmap_clear_free(Hashmap *h) {
-        void *p;
+void internal_hashmap_clear_free(HashmapBase *h) {
+        unsigned idx;
 
         if (!h)
                 return;
 
-        while ((p = hashmap_steal_first(h)))
-                free(p);
+        for (idx = skip_free_buckets(h, 0); idx != IDX_NIL;
+             idx = skip_free_buckets(h, idx + 1))
+                free(entry_value(h, bucket_at(h, idx)));
+
+        internal_hashmap_clear(h);
 }
 
 void hashmap_clear_free_free(Hashmap *h) {
+        unsigned idx;
+
         if (!h)
                 return;
 
-        while (h->iterate_list_head) {
-                void *a, *b;
-
-                a = h->iterate_list_head->value;
-                b = (void*) h->iterate_list_head->key;
-                remove_entry(h, h->iterate_list_head);
-                free(a);
-                free(b);
+        for (idx = skip_free_buckets(HASHMAP_BASE(h), 0); idx != IDX_NIL;
+             idx = skip_free_buckets(HASHMAP_BASE(h), idx + 1)) {
+                struct plain_hashmap_entry *e = plain_bucket_at(h, idx);
+                free((void*)e->b.key);
+                free(e->value);
         }
+
+        internal_hashmap_clear(HASHMAP_BASE(h));
 }
 
-static struct hashmap_entry *hash_scan(Hashmap *h, unsigned hash, const void *key) {
-        struct hashmap_entry *e;
-        assert(h);
-        assert(hash < h->n_buckets);
+static int resize_buckets(HashmapBase *h, unsigned entries_add);
+
+/*
+ * Finds an empty bucket to put an entry into, starting the scan at 'idx'.
+ * Performs Robin Hood swaps as it goes. The entry to put must be placed
+ * by the caller into swap slot IDX_PUT.
+ * If used for in-place resizing, may leave a displaced entry in swap slot
+ * IDX_PUT. Caller must rehash it next.
+ * Returns: true if it left a displaced entry to rehash next in IDX_PUT,
+ *          false otherwise.
+ */
+static bool hashmap_put_robin_hood(HashmapBase *h, unsigned idx,
+                                   struct swap_entries *swap) {
+        dib_raw_t raw_dib, *dibs;
+        unsigned dib, distance;
+
+#ifdef ENABLE_HASHMAP_DEBUG
+        h->debug.put_count++;
+#endif
+
+        dibs = dib_raw_ptr(h);
+
+        for (distance = 0; ; distance++) {
+                raw_dib = dibs[idx];
+                if (raw_dib == DIB_RAW_FREE || raw_dib == DIB_RAW_REHASH) {
+                        if (raw_dib == DIB_RAW_REHASH)
+                                bucket_move_entry(h, swap, idx, IDX_TMP);
+
+                        if (h->has_indirect && h->indirect.idx_lowest_entry > idx)
+                                h->indirect.idx_lowest_entry = idx;
 
-        for (e = h->buckets[hash]; e; e = e->bucket_next)
-                if (h->hash_ops->compare(e->key, key) == 0)
-                        return e;
+                        bucket_set_dib(h, idx, distance);
+                        bucket_move_entry(h, swap, IDX_PUT, idx);
+                        if (raw_dib == DIB_RAW_REHASH) {
+                                bucket_move_entry(h, swap, IDX_TMP, IDX_PUT);
+                                return true;
+                        }
 
-        return NULL;
+                        return false;
+                }
+
+                dib = bucket_calculate_dib(h, idx, raw_dib);
+
+                if (dib < distance) {
+                        /* Found a wealthier entry. Go Robin Hood! */
+
+                        bucket_set_dib(h, idx, distance);
+
+                        /* swap the entries */
+                        bucket_move_entry(h, swap, idx, IDX_TMP);
+                        bucket_move_entry(h, swap, IDX_PUT, idx);
+                        bucket_move_entry(h, swap, IDX_TMP, IDX_PUT);
+
+                        distance = dib;
+                }
+
+                idx = next_idx(h, idx);
+        }
 }
 
-static int resize_buckets(Hashmap *h, unsigned entries_add) {
-        struct hashmap_entry **n, *i;
-        unsigned m, new_n_entries, new_n_buckets;
-        uint8_t nkey[HASH_KEY_SIZE];
+/*
+ * Puts an entry into a hashmap, boldly - no check whether key already exists.
+ * The caller must place the entry (only its key and value, not link indexes)
+ * in swap slot IDX_PUT.
+ * Caller must ensure: the key does not exist yet in the hashmap.
+ *                     that resize is not needed if !may_resize.
+ * Returns: 1 if entry was put successfully.
+ *          -ENOMEM if may_resize==true and resize failed with -ENOMEM.
+ *          Cannot return -ENOMEM if !may_resize.
+ */
+static int hashmap_base_put_boldly(HashmapBase *h, unsigned idx,
+                                   struct swap_entries *swap, bool may_resize) {
+        struct ordered_hashmap_entry *new_entry;
+        int r;
+
+        assert(idx < n_buckets(h));
+
+        new_entry = bucket_at_swap(swap, IDX_PUT);
+
+        if (may_resize) {
+                r = resize_buckets(h, 1);
+                if (r < 0)
+                        return r;
+                if (r > 0)
+                        idx = bucket_hash(h, new_entry->p.b.key);
+        }
+        assert(n_entries(h) < n_buckets(h));
+
+        if (h->type == HASHMAP_TYPE_ORDERED) {
+                OrderedHashmap *lh = (OrderedHashmap*) h;
+
+                new_entry->iterate_next = IDX_NIL;
+                new_entry->iterate_previous = lh->iterate_list_tail;
+
+                if (lh->iterate_list_tail != IDX_NIL) {
+                        struct ordered_hashmap_entry *old_tail;
+
+                        old_tail = ordered_bucket_at(lh, lh->iterate_list_tail);
+                        assert(old_tail->iterate_next == IDX_NIL);
+                        old_tail->iterate_next = IDX_PUT;
+                }
+
+                lh->iterate_list_tail = IDX_PUT;
+                if (lh->iterate_list_head == IDX_NIL)
+                        lh->iterate_list_head = IDX_PUT;
+        }
+
+        assert_se(hashmap_put_robin_hood(h, idx, swap) == false);
+
+        n_entries_inc(h);
+#ifdef ENABLE_HASHMAP_DEBUG
+        h->debug.max_entries = MAX(h->debug.max_entries, n_entries(h));
+#endif
+
+        return 1;
+}
+#define hashmap_put_boldly(h, idx, swap, may_resize) \
+        hashmap_base_put_boldly(HASHMAP_BASE(h), idx, swap, may_resize)
+
+/*
+ * Returns 0 if resize is not needed.
+ *         1 if succesfully resized.
+ *         -ENOMEM on allocation failure.
+ */
+static int resize_buckets(HashmapBase *h, unsigned entries_add) {
+        struct swap_entries swap;
+        char *new_storage;
+        dib_raw_t *old_dibs, *new_dibs;
+        const struct hashmap_type_info *hi;
+        unsigned idx, optimal_idx;
+        unsigned old_n_buckets, new_n_buckets, n_rehashed, new_n_entries;
+        uint8_t new_shift;
+        bool rehash_next;
 
         assert(h);
 
-        new_n_entries = h->n_entries + entries_add;
+        hi = &hashmap_type_info[h->type];
+        new_n_entries = n_entries(h) + entries_add;
 
         /* overflow? */
-        if (_unlikely_(new_n_entries < entries_add || new_n_entries > UINT_MAX / 4))
+        if (_unlikely_(new_n_entries < entries_add))
                 return -ENOMEM;
 
-        new_n_buckets = new_n_entries * 4 / 3;
-
-        if (_likely_(new_n_buckets <= h->n_buckets))
+        /* For direct storage we allow 100% load, because it's tiny. */
+        if (!h->has_indirect && new_n_entries <= hi->n_direct_buckets)
                 return 0;
 
-        /* Increase by four at least */
-        m = MAX((h->n_entries+1)*4-1, new_n_buckets);
-
-        /* If we hit OOM we simply risk packed hashmaps... */
-        n = new0(struct hashmap_entry*, m);
-        if (!n)
+        /*
+         * Load factor = n/m = 1 - (1/INV_KEEP_FREE).
+         * From it follows: m = n + n/(INV_KEEP_FREE - 1)
+         */
+        new_n_buckets = new_n_entries + new_n_entries / (INV_KEEP_FREE - 1);
+        /* overflow? */
+        if (_unlikely_(new_n_buckets < new_n_entries))
                 return -ENOMEM;
 
-        /* Let's use a different randomized hash key for the
-         * extension, so that people cannot guess what we are using
-         * here forever */
-        get_hash_key(nkey, false);
+        if (_unlikely_(new_n_buckets > UINT_MAX / (hi->entry_size + sizeof(dib_raw_t))))
+                return -ENOMEM;
 
-        for (i = h->iterate_list_head; i; i = i->iterate_next) {
-                unsigned long old_bucket, new_bucket;
+        old_n_buckets = n_buckets(h);
 
-                old_bucket = h->hash_ops->hash(i->key, h->hash_key) % h->n_buckets;
+        if (_likely_(new_n_buckets <= old_n_buckets))
+                return 0;
 
-                /* First, drop from old bucket table */
-                if (i->bucket_next)
-                        i->bucket_next->bucket_previous = i->bucket_previous;
+        new_shift = log2u_round_up(MAX(
+                        new_n_buckets * (hi->entry_size + sizeof(dib_raw_t)),
+                        2 * sizeof(struct direct_storage)));
 
-                if (i->bucket_previous)
-                        i->bucket_previous->bucket_next = i->bucket_next;
-                else
-                        h->buckets[old_bucket] = i->bucket_next;
+        /* Realloc storage (buckets and DIB array). */
+        new_storage = realloc(h->has_indirect ? h->indirect.storage : NULL,
+                              1U << new_shift);
+        if (!new_storage)
+                return -ENOMEM;
 
-                /* Then, add to new backet table */
-                new_bucket = h->hash_ops->hash(i->key, nkey) % m;
+        /* Must upgrade direct to indirect storage. */
+        if (!h->has_indirect) {
+                memcpy(new_storage, h->direct.storage,
+                       old_n_buckets * (hi->entry_size + sizeof(dib_raw_t)));
+                h->indirect.n_entries = h->n_direct_entries;
+                h->indirect.idx_lowest_entry = 0;
+                h->n_direct_entries = 0;
+        }
 
-                i->bucket_next = n[new_bucket];
-                i->bucket_previous = NULL;
-                if (n[new_bucket])
-                        n[new_bucket]->bucket_previous = i;
-                n[new_bucket] = i;
+        /* Get a new hash key. If we've just upgraded to indirect storage,
+         * allow reusing a previously generated key. It's still a different key
+         * from the shared one that we used for direct storage. */
+        get_hash_key(h->indirect.hash_key, !h->has_indirect);
+
+        h->has_indirect = true;
+        h->indirect.storage = new_storage;
+        h->indirect.n_buckets = (1U << new_shift) /
+                                (hi->entry_size + sizeof(dib_raw_t));
+
+        old_dibs = (dib_raw_t*)(new_storage + hi->entry_size * old_n_buckets);
+        new_dibs = dib_raw_ptr(h);
+
+        /*
+         * Move the DIB array to the new place, replacing valid DIB values with
+         * DIB_RAW_REHASH to indicate all of the used buckets need rehashing.
+         * Note: Overlap is not possible, because we have at least doubled the
+         * number of buckets and dib_raw_t is smaller than any entry type.
+         */
+        for (idx = 0; idx < old_n_buckets; idx++) {
+                assert(old_dibs[idx] != DIB_RAW_REHASH);
+                new_dibs[idx] = old_dibs[idx] == DIB_RAW_FREE ? DIB_RAW_FREE
+                                                              : DIB_RAW_REHASH;
         }
 
-        if (h->buckets != (struct hashmap_entry**) ((uint8_t*) h + ALIGN(sizeof(Hashmap))))
-                free(h->buckets);
+        /* Zero the area of newly added entries (including the old DIB area) */
+        memset(bucket_at(h, old_n_buckets), 0,
+               (n_buckets(h) - old_n_buckets) * hi->entry_size);
 
-        h->buckets = n;
-        h->n_buckets = m;
+        /* The upper half of the new DIB array needs initialization */
+        memset(&new_dibs[old_n_buckets], DIB_RAW_INIT,
+               (n_buckets(h) - old_n_buckets) * sizeof(dib_raw_t));
 
-        memcpy(h->hash_key, nkey, HASH_KEY_SIZE);
+        /* Rehash entries that need it */
+        n_rehashed = 0;
+        for (idx = 0; idx < old_n_buckets; idx++) {
+                if (new_dibs[idx] != DIB_RAW_REHASH)
+                        continue;
 
-        return 1;
-}
+                optimal_idx = bucket_hash(h, bucket_at(h, idx)->key);
 
-static int __hashmap_put(Hashmap *h, const void *key, void *value, unsigned hash) {
-        /* For when we know no such entry exists yet */
+                /*
+                 * Not much to do if by luck the entry hashes to its current
+                 * location. Just set its DIB.
+                 */
+                if (optimal_idx == idx) {
+                        new_dibs[idx] = 0;
+                        n_rehashed++;
+                        continue;
+                }
+
+                new_dibs[idx] = DIB_RAW_FREE;
+                bucket_move_entry(h, &swap, idx, IDX_PUT);
+                /* bucket_move_entry does not clear the source */
+                memset(bucket_at(h, idx), 0, hi->entry_size);
+
+                do {
+                        /*
+                         * Find the new bucket for the current entry. This may make
+                         * another entry homeless and load it into IDX_PUT.
+                         */
+                        rehash_next = hashmap_put_robin_hood(h, optimal_idx, &swap);
+                        n_rehashed++;
+
+                        /* Did the current entry displace another one? */
+                        if (rehash_next)
+                                optimal_idx = bucket_hash(h, bucket_at_swap(&swap, IDX_PUT)->p.b.key);
+                } while (rehash_next);
+        }
 
-        struct hashmap_entry *e;
+        assert(n_rehashed == n_entries(h));
 
-        if (resize_buckets(h, 1) > 0)
-                hash = bucket_hash(h, key);
+        return 1;
+}
 
-        if (h->from_pool)
-                e = mempool_alloc_tile(&hashmap_entry_pool);
-        else
-                e = new(struct hashmap_entry, 1);
+/*
+ * Finds an entry with a matching key
+ * Returns: index of the found entry, or IDX_NIL if not found.
+ */
+static unsigned base_bucket_scan(HashmapBase *h, unsigned idx, const void *key) {
+        struct hashmap_base_entry *e;
+        unsigned dib, distance;
+        dib_raw_t *dibs = dib_raw_ptr(h);
 
-        if (!e)
-                return -ENOMEM;
+        assert(idx < n_buckets(h));
 
-        e->key = key;
-        e->value = value;
+        for (distance = 0; ; distance++) {
+                if (dibs[idx] == DIB_RAW_FREE)
+                        return IDX_NIL;
 
-        link_entry(h, e, hash);
+                dib = bucket_calculate_dib(h, idx, dibs[idx]);
 
-        return 1;
+                if (dib < distance)
+                        return IDX_NIL;
+                if (dib == distance) {
+                        e = bucket_at(h, idx);
+                        if (h->hash_ops->compare(e->key, key) == 0)
+                                return idx;
+                }
+
+                idx = next_idx(h, idx);
+        }
 }
+#define bucket_scan(h, idx, key) base_bucket_scan(HASHMAP_BASE(h), idx, key)
 
 int hashmap_put(Hashmap *h, const void *key, void *value) {
-        struct hashmap_entry *e;
-        unsigned hash;
+        struct swap_entries swap;
+        struct plain_hashmap_entry *e;
+        unsigned hash, idx;
 
         assert(h);
 
         hash = bucket_hash(h, key);
-        e = hash_scan(h, hash, key);
-        if (e) {
+        idx = bucket_scan(h, hash, key);
+        if (idx != IDX_NIL) {
+                e = plain_bucket_at(h, idx);
                 if (e->value == value)
                         return 0;
                 return -EEXIST;
         }
 
-        return __hashmap_put(h, key, value, hash);
+        e = &bucket_at_swap(&swap, IDX_PUT)->p;
+        e->b.key = key;
+        e->value = value;
+        return hashmap_put_boldly(h, hash, &swap, true);
+}
+
+int set_put(Set *s, const void *key) {
+        struct swap_entries swap;
+        struct hashmap_base_entry *e;
+        unsigned hash, idx;
+
+        assert(s);
+
+        hash = bucket_hash(s, key);
+        idx = bucket_scan(s, hash, key);
+        if (idx != IDX_NIL)
+                return 0;
+
+        e = &bucket_at_swap(&swap, IDX_PUT)->p.b;
+        e->key = key;
+        return hashmap_put_boldly(s, hash, &swap, true);
 }
 
 int hashmap_replace(Hashmap *h, const void *key, void *value) {
-        struct hashmap_entry *e;
-        unsigned hash;
+        struct swap_entries swap;
+        struct plain_hashmap_entry *e;
+        unsigned hash, idx;
 
         assert(h);
 
         hash = bucket_hash(h, key);
-        e = hash_scan(h, hash, key);
-        if (e) {
-                e->key = key;
+        idx = bucket_scan(h, hash, key);
+        if (idx != IDX_NIL) {
+                e = plain_bucket_at(h, idx);
+#ifdef ENABLE_HASHMAP_DEBUG
+                /* Although the key is equal, the key pointer may have changed,
+                 * and this would break our assumption for iterating. So count
+                 * this operation as incompatible with iteration. */
+                if (e->b.key != key) {
+                        h->b.debug.put_count++;
+                        h->b.debug.rem_count++;
+                        h->b.debug.last_rem_idx = idx;
+                }
+#endif
+                e->b.key = key;
                 e->value = value;
                 return 0;
         }
 
-        return __hashmap_put(h, key, value, hash);
+        e = &bucket_at_swap(&swap, IDX_PUT)->p;
+        e->b.key = key;
+        e->value = value;
+        return hashmap_put_boldly(h, hash, &swap, true);
 }
 
 int hashmap_update(Hashmap *h, const void *key, void *value) {
-        struct hashmap_entry *e;
-        unsigned hash;
+        struct plain_hashmap_entry *e;
+        unsigned hash, idx;
 
         assert(h);
 
         hash = bucket_hash(h, key);
-        e = hash_scan(h, hash, key);
-        if (!e)
+        idx = bucket_scan(h, hash, key);
+        if (idx == IDX_NIL)
                 return -ENOENT;
 
+        e = plain_bucket_at(h, idx);
         e->value = value;
         return 0;
 }
 
-void* hashmap_get(Hashmap *h, const void *key) {
-        unsigned hash;
-        struct hashmap_entry *e;
+void *internal_hashmap_get(HashmapBase *h, const void *key) {
+        struct hashmap_base_entry *e;
+        unsigned hash, idx;
 
         if (!h)
                 return NULL;
 
         hash = bucket_hash(h, key);
-        e = hash_scan(h, hash, key);
-        if (!e)
+        idx = bucket_scan(h, hash, key);
+        if (idx == IDX_NIL)
                 return NULL;
 
-        return e->value;
+        e = bucket_at(h, idx);
+        return entry_value(h, e);
 }
 
-voidhashmap_get2(Hashmap *h, const void *key, void **key2) {
-        unsigned hash;
-        struct hashmap_entry *e;
+void *hashmap_get2(Hashmap *h, const void *key, void **key2) {
+        struct plain_hashmap_entry *e;
+        unsigned hash, idx;
 
         if (!h)
                 return NULL;
 
         hash = bucket_hash(h, key);
-        e = hash_scan(h, hash, key);
-        if (!e)
+        idx = bucket_scan(h, hash, key);
+        if (idx == IDX_NIL)
                 return NULL;
 
+        e = plain_bucket_at(h, idx);
         if (key2)
-                *key2 = (void*) e->key;
+                *key2 = (void*) e->b.key;
 
         return e->value;
 }
 
-bool hashmap_contains(Hashmap *h, const void *key) {
+bool internal_hashmap_contains(HashmapBase *h, const void *key) {
         unsigned hash;
 
         if (!h)
                 return false;
 
         hash = bucket_hash(h, key);
-        return !!hash_scan(h, hash, key);
+        return bucket_scan(h, hash, key) != IDX_NIL;
 }
 
-void* hashmap_remove(Hashmap *h, const void *key) {
-        struct hashmap_entry *e;
-        unsigned hash;
+void *internal_hashmap_remove(HashmapBase *h, const void *key) {
+        struct hashmap_base_entry *e;
+        unsigned hash, idx;
         void *data;
 
         if (!h)
                 return NULL;
 
         hash = bucket_hash(h, key);
-        e = hash_scan(h, hash, key);
-        if (!e)
+        idx = bucket_scan(h, hash, key);
+        if (idx == IDX_NIL)
                 return NULL;
 
-        data = e->value;
-        remove_entry(h, e);
+        e = bucket_at(h, idx);
+        data = entry_value(h, e);
+        remove_entry(h, idx);
 
         return data;
 }
 
-voidhashmap_remove2(Hashmap *h, const void *key, void **rkey) {
-        struct hashmap_entry *e;
-        unsigned hash;
+void *hashmap_remove2(Hashmap *h, const void *key, void **rkey) {
+        struct plain_hashmap_entry *e;
+        unsigned hash, idx;
         void *data;
 
         if (!h) {
@@ -582,228 +1397,249 @@ void* hashmap_remove2(Hashmap *h, const void *key, void **rkey) {
         }
 
         hash = bucket_hash(h, key);
-        e = hash_scan(h, hash, key);
-        if (!e) {
+        idx = bucket_scan(h, hash, key);
+        if (idx == IDX_NIL) {
                 if (rkey)
                         *rkey = NULL;
                 return NULL;
         }
 
+        e = plain_bucket_at(h, idx);
         data = e->value;
         if (rkey)
-                *rkey = (void*) e->key;
+                *rkey = (void*) e->b.key;
 
-        remove_entry(h, e);
+        remove_entry(h, idx);
 
         return data;
 }
 
 int hashmap_remove_and_put(Hashmap *h, const void *old_key, const void *new_key, void *value) {
-        struct hashmap_entry *e;
-        unsigned old_hash, new_hash;
+        struct swap_entries swap;
+        struct plain_hashmap_entry *e;
+        unsigned old_hash, new_hash, idx;
 
         if (!h)
                 return -ENOENT;
 
         old_hash = bucket_hash(h, old_key);
-        e = hash_scan(h, old_hash, old_key);
-        if (!e)
+        idx = bucket_scan(h, old_hash, old_key);
+        if (idx == IDX_NIL)
                 return -ENOENT;
 
         new_hash = bucket_hash(h, new_key);
-        if (hash_scan(h, new_hash, new_key))
+        if (bucket_scan(h, new_hash, new_key) != IDX_NIL)
                 return -EEXIST;
 
-        unlink_entry(h, e, old_hash);
+        remove_entry(h, idx);
 
-        e->key = new_key;
+        e = &bucket_at_swap(&swap, IDX_PUT)->p;
+        e->b.key = new_key;
         e->value = value;
+        assert_se(hashmap_put_boldly(h, new_hash, &swap, false) == 1);
+
+        return 0;
+}
+
+int set_remove_and_put(Set *s, const void *old_key, const void *new_key) {
+        struct swap_entries swap;
+        struct hashmap_base_entry *e;
+        unsigned old_hash, new_hash, idx;
 
-        link_entry(h, e, new_hash);
+        if (!s)
+                return -ENOENT;
+
+        old_hash = bucket_hash(s, old_key);
+        idx = bucket_scan(s, old_hash, old_key);
+        if (idx == IDX_NIL)
+                return -ENOENT;
+
+        new_hash = bucket_hash(s, new_key);
+        if (bucket_scan(s, new_hash, new_key) != IDX_NIL)
+                return -EEXIST;
+
+        remove_entry(s, idx);
+
+        e = &bucket_at_swap(&swap, IDX_PUT)->p.b;
+        e->key = new_key;
+        assert_se(hashmap_put_boldly(s, new_hash, &swap, false) == 1);
 
         return 0;
 }
 
 int hashmap_remove_and_replace(Hashmap *h, const void *old_key, const void *new_key, void *value) {
-        struct hashmap_entry *e, *k;
-        unsigned old_hash, new_hash;
+        struct swap_entries swap;
+        struct plain_hashmap_entry *e;
+        unsigned old_hash, new_hash, idx_old, idx_new;
 
         if (!h)
                 return -ENOENT;
 
         old_hash = bucket_hash(h, old_key);
-        e = hash_scan(h, old_hash, old_key);
-        if (!e)
+        idx_old = bucket_scan(h, old_hash, old_key);
+        if (idx_old == IDX_NIL)
                 return -ENOENT;
 
-        new_hash = bucket_hash(h, new_key);
-        k = hash_scan(h, new_hash, new_key);
-        if (k)
-                if (e != k)
-                        remove_entry(h, k);
-
-        unlink_entry(h, e, old_hash);
+        old_key = bucket_at(HASHMAP_BASE(h), idx_old)->key;
 
-        e->key = new_key;
+        new_hash = bucket_hash(h, new_key);
+        idx_new = bucket_scan(h, new_hash, new_key);
+        if (idx_new != IDX_NIL)
+                if (idx_old != idx_new) {
+                        remove_entry(h, idx_new);
+                        /* Compensate for a possible backward shift. */
+                        if (old_key != bucket_at(HASHMAP_BASE(h), idx_old)->key)
+                                idx_old = prev_idx(HASHMAP_BASE(h), idx_old);
+                        assert(old_key == bucket_at(HASHMAP_BASE(h), idx_old)->key);
+                }
+
+        remove_entry(h, idx_old);
+
+        e = &bucket_at_swap(&swap, IDX_PUT)->p;
+        e->b.key = new_key;
         e->value = value;
-
-        link_entry(h, e, new_hash);
+        assert_se(hashmap_put_boldly(h, new_hash, &swap, false) == 1);
 
         return 0;
 }
 
-voidhashmap_remove_value(Hashmap *h, const void *key, void *value) {
-        struct hashmap_entry *e;
-        unsigned hash;
+void *hashmap_remove_value(Hashmap *h, const void *key, void *value) {
+        struct plain_hashmap_entry *e;
+        unsigned hash, idx;
 
         if (!h)
                 return NULL;
 
         hash = bucket_hash(h, key);
-
-        e = hash_scan(h, hash, key);
-        if (!e)
+        idx = bucket_scan(h, hash, key);
+        if (idx == IDX_NIL)
                 return NULL;
 
+        e = plain_bucket_at(h, idx);
         if (e->value != value)
                 return NULL;
 
-        remove_entry(h, e);
+        remove_entry(h, idx);
 
         return value;
 }
 
-void *hashmap_iterate(Hashmap *h, Iterator *i, const void **key) {
-        struct hashmap_entry *e;
-
-        assert(i);
-
-        if (!h)
-                goto at_end;
-
-        if (*i == ITERATOR_LAST)
-                goto at_end;
-
-        if (*i == ITERATOR_FIRST && !h->iterate_list_head)
-                goto at_end;
-
-        e = *i == ITERATOR_FIRST ? h->iterate_list_head : (struct hashmap_entry*) *i;
+static unsigned find_first_entry(HashmapBase *h) {
+        Iterator i = ITERATOR_FIRST;
 
-        if (e->iterate_next)
-                *i = (Iterator) e->iterate_next;
-        else
-                *i = ITERATOR_LAST;
+        if (!h || !n_entries(h))
+                return IDX_NIL;
 
-        if (key)
-                *key = e->key;
-
-        return e->value;
-
-at_end:
-        *i = ITERATOR_LAST;
-
-        if (key)
-                *key = NULL;
-
-        return NULL;
+        return hashmap_iterate_entry(h, &i);
 }
 
-void* hashmap_first(Hashmap *h) {
-
-        if (!h)
-                return NULL;
+void *internal_hashmap_first(HashmapBase *h) {
+        unsigned idx;
 
-        if (!h->iterate_list_head)
+        idx = find_first_entry(h);
+        if (idx == IDX_NIL)
                 return NULL;
 
-        return h->iterate_list_head->value;
+        return entry_value(h, bucket_at(h, idx));
 }
 
-void* hashmap_first_key(Hashmap *h) {
+void *internal_hashmap_first_key(HashmapBase *h) {
+        struct hashmap_base_entry *e;
+        unsigned idx;
 
-        if (!h)
-                return NULL;
-
-        if (!h->iterate_list_head)
+        idx = find_first_entry(h);
+        if (idx == IDX_NIL)
                 return NULL;
 
-        return (void*) h->iterate_list_head->key;
+        e = bucket_at(h, idx);
+        return (void*) e->key;
 }
 
-void* hashmap_steal_first(Hashmap *h) {
+void *internal_hashmap_steal_first(HashmapBase *h) {
+        struct hashmap_base_entry *e;
         void *data;
+        unsigned idx;
 
-        if (!h)
-                return NULL;
-
-        if (!h->iterate_list_head)
+        idx = find_first_entry(h);
+        if (idx == IDX_NIL)
                 return NULL;
 
-        data = h->iterate_list_head->value;
-        remove_entry(h, h->iterate_list_head);
+        e = bucket_at(h, idx);
+        data = entry_value(h, e);
+        remove_entry(h, idx);
 
         return data;
 }
 
-void* hashmap_steal_first_key(Hashmap *h) {
+void *internal_hashmap_steal_first_key(HashmapBase *h) {
+        struct hashmap_base_entry *e;
         void *key;
+        unsigned idx;
 
-        if (!h)
-                return NULL;
-
-        if (!h->iterate_list_head)
+        idx = find_first_entry(h);
+        if (idx == IDX_NIL)
                 return NULL;
 
-        key = (void*) h->iterate_list_head->key;
-        remove_entry(h, h->iterate_list_head);
+        e = bucket_at(h, idx);
+        key = (void*) e->key;
+        remove_entry(h, idx);
 
         return key;
 }
 
-unsigned hashmap_size(Hashmap *h) {
+unsigned internal_hashmap_size(HashmapBase *h) {
 
         if (!h)
                 return 0;
 
-        return h->n_entries;
+        return n_entries(h);
 }
 
-unsigned hashmap_buckets(Hashmap *h) {
+unsigned internal_hashmap_buckets(HashmapBase *h) {
 
         if (!h)
                 return 0;
 
-        return h->n_buckets;
+        return n_buckets(h);
 }
 
-bool hashmap_isempty(Hashmap *h) {
+int internal_hashmap_merge(Hashmap *h, Hashmap *other) {
+        Iterator i;
+        unsigned idx;
 
-        if (!h)
-                return true;
+        assert(h);
 
-        return h->n_entries == 0;
-}
+        HASHMAP_FOREACH_IDX(idx, HASHMAP_BASE(other), i) {
+                struct plain_hashmap_entry *pe = plain_bucket_at(other, idx);
+                int r;
 
-int hashmap_merge(Hashmap *h, Hashmap *other) {
-        struct hashmap_entry *e;
+                r = hashmap_put(h, pe->b.key, pe->value);
+                if (r < 0 && r != -EEXIST)
+                        return r;
+        }
 
-        assert(h);
+        return 0;
+}
 
-        if (!other)
-                return 0;
+int set_merge(Set *s, Set *other) {
+        Iterator i;
+        unsigned idx;
 
-        for (e = other->iterate_list_head; e; e = e->iterate_next) {
+        assert(s);
+
+        HASHMAP_FOREACH_IDX(idx, HASHMAP_BASE(other), i) {
+                struct set_entry *se = set_bucket_at(other, idx);
                 int r;
 
-                r = hashmap_put(h, e->key, e->value);
-                if (r < 0 && r != -EEXIST)
+                r = set_put(s, se->b.key);
+                if (r < 0)
                         return r;
         }
 
         return 0;
 }
 
-int hashmap_reserve(Hashmap *h, unsigned entries_add) {
+int internal_hashmap_reserve(HashmapBase *h, unsigned entries_add) {
         int r;
 
         assert(h);
@@ -815,96 +1651,144 @@ int hashmap_reserve(Hashmap *h, unsigned entries_add) {
         return 0;
 }
 
-int hashmap_move(Hashmap *h, Hashmap *other) {
-        struct hashmap_entry *e, *n;
+/*
+ * The same as hashmap_merge(), but every new item from other is moved to h.
+ * Keys already in h are skipped and stay in other.
+ * Returns: 0 on success.
+ *          -ENOMEM on alloc failure, in which case no move has been done.
+ */
+int internal_hashmap_move(HashmapBase *h, HashmapBase *other) {
+        struct swap_entries swap;
+        struct hashmap_base_entry *e, *n;
+        Iterator i;
+        unsigned idx;
+        int r;
 
         assert(h);
 
-        /* The same as hashmap_merge(), but every new item from other
-         * is moved to h. */
-
         if (!other)
                 return 0;
 
-        for (e = other->iterate_list_head; e; e = n) {
-                unsigned h_hash, other_hash;
+        assert(other->type == h->type);
+
+        /*
+         * This reserves buckets for the worst case, where none of other's
+         * entries are yet present in h. This is preferable to risking
+         * an allocation failure in the middle of the moving and having to
+         * rollback or return a partial result.
+         */
+        r = resize_buckets(h, n_entries(other));
+        if (r < 0)
+                return r;
 
-                n = e->iterate_next;
+        HASHMAP_FOREACH_IDX(idx, other, i) {
+                unsigned h_hash;
 
+                e = bucket_at(other, idx);
                 h_hash = bucket_hash(h, e->key);
-                if (hash_scan(h, h_hash, e->key))
+                if (bucket_scan(h, h_hash, e->key) != IDX_NIL)
                         continue;
 
-                other_hash = bucket_hash(other, e->key);
-                unlink_entry(other, e, other_hash);
-                link_entry(h, e, h_hash);
+                n = &bucket_at_swap(&swap, IDX_PUT)->p.b;
+                n->key = e->key;
+                if (h->type != HASHMAP_TYPE_SET)
+                        ((struct plain_hashmap_entry*) n)->value =
+                                ((struct plain_hashmap_entry*) e)->value;
+                assert_se(hashmap_put_boldly(h, h_hash, &swap, false) == 1);
+
+                remove_entry(other, idx);
         }
 
         return 0;
 }
 
-int hashmap_move_one(Hashmap *h, Hashmap *other, const void *key) {
-        unsigned h_hash, other_hash;
-        struct hashmap_entry *e;
+int internal_hashmap_move_one(HashmapBase *h, HashmapBase *other, const void *key) {
+        struct swap_entries swap;
+        unsigned h_hash, other_hash, idx;
+        struct hashmap_base_entry *e, *n;
+        int r;
 
         assert(h);
 
         h_hash = bucket_hash(h, key);
-        if (hash_scan(h, h_hash, key))
+        if (bucket_scan(h, h_hash, key) != IDX_NIL)
                 return -EEXIST;
 
         if (!other)
                 return -ENOENT;
 
+        assert(other->type == h->type);
+
         other_hash = bucket_hash(other, key);
-        e = hash_scan(other, other_hash, key);
-        if (!e)
+        idx = bucket_scan(other, other_hash, key);
+        if (idx == IDX_NIL)
                 return -ENOENT;
 
-        unlink_entry(other, e, other_hash);
-        link_entry(h, e, h_hash);
+        e = bucket_at(other, idx);
+
+        n = &bucket_at_swap(&swap, IDX_PUT)->p.b;
+        n->key = e->key;
+        if (h->type != HASHMAP_TYPE_SET)
+                ((struct plain_hashmap_entry*) n)->value =
+                        ((struct plain_hashmap_entry*) e)->value;
+        r = hashmap_put_boldly(h, h_hash, &swap, true);
+        if (r < 0)
+                return r;
 
+        remove_entry(other, idx);
         return 0;
 }
 
-Hashmap *hashmap_copy(Hashmap *h) {
-        Hashmap *copy;
+HashmapBase *internal_hashmap_copy(HashmapBase *h) {
+        HashmapBase *copy;
+        int r;
 
         assert(h);
 
-        copy = hashmap_new(h->hash_ops);
+        copy = hashmap_base_new(h->hash_ops, h->type  HASHMAP_DEBUG_SRC_ARGS);
         if (!copy)
                 return NULL;
 
-        if (hashmap_merge(copy, h) < 0) {
-                hashmap_free(copy);
+        switch (h->type) {
+        case HASHMAP_TYPE_PLAIN:
+        case HASHMAP_TYPE_ORDERED:
+                r = hashmap_merge((Hashmap*)copy, (Hashmap*)h);
+                break;
+        case HASHMAP_TYPE_SET:
+                r = set_merge((Set*)copy, (Set*)h);
+                break;
+        default:
+                assert_not_reached("Unknown hashmap type");
+        }
+
+        if (r < 0) {
+                internal_hashmap_free(copy);
                 return NULL;
         }
 
         return copy;
 }
 
-char **hashmap_get_strv(Hashmap *h) {
+char **internal_hashmap_get_strv(HashmapBase *h) {
         char **sv;
-        Iterator it;
-        char *item;
-        int n;
+        Iterator i;
+        unsigned idx, n;
 
-        sv = new(char*, h->n_entries+1);
+        sv = new(char*, n_entries(h)+1);
         if (!sv)
                 return NULL;
 
         n = 0;
-        HASHMAP_FOREACH(item, h, it)
-                sv[n++] = item;
+        HASHMAP_FOREACH_IDX(idx, h, i)
+                sv[n++] = entry_value(h, bucket_at(h, idx));
         sv[n] = NULL;
 
         return sv;
 }
 
-void *hashmap_next(Hashmap *h, const void *key) {
-        unsigned hash;
-        struct hashmap_entry *e;
+void *ordered_hashmap_next(OrderedHashmap *h, const void *key) {
+        struct ordered_hashmap_entry *e;
+        unsigned hash, idx;
 
         assert(key);
 
@@ -912,13 +1796,55 @@ void *hashmap_next(Hashmap *h, const void *key) {
                 return NULL;
 
         hash = bucket_hash(h, key);
-        e = hash_scan(h, hash, key);
-        if (!e)
+        idx = bucket_scan(h, hash, key);
+        if (idx == IDX_NIL)
                 return NULL;
 
-        e = e->iterate_next;
-        if (!e)
+        e = ordered_bucket_at(h, idx);
+        if (e->iterate_next == IDX_NIL)
                 return NULL;
+        return ordered_bucket_at(h, e->iterate_next)->p.value;
+}
 
-        return e->value;
+int set_consume(Set *s, void *value) {
+        int r;
+
+        r = set_put(s, value);
+        if (r < 0)
+                free(value);
+
+        return r;
+}
+
+int set_put_strdup(Set *s, const char *p) {
+        char *c;
+        int r;
+
+        assert(s);
+        assert(p);
+
+        c = strdup(p);
+        if (!c)
+                return -ENOMEM;
+
+        r = set_consume(s, c);
+        if (r == -EEXIST)
+                return 0;
+
+        return r;
+}
+
+int set_put_strdupv(Set *s, char **l) {
+        int n = 0, r;
+        char **i;
+
+        STRV_FOREACH(i, l) {
+                r = set_put_strdup(s, *i);
+                if (r < 0)
+                        return r;
+
+                n += r;
+        }
+
+        return n;
 }
index 65fb3c0..9c6e0ca 100644 (file)
@@ -6,6 +6,7 @@
   This file is part of systemd.
 
   Copyright 2010 Lennart Poettering
+  Copyright 2014 Michal Schmidt
 
   systemd is free software; you can redistribute it and/or modify it
   under the terms of the GNU Lesser General Public License as published by
 #include "macro.h"
 #include "util.h"
 
-/* Pretty straightforward hash table implementation. As a minor
- * optimization a NULL hashmap object will be treated as empty hashmap
- * for all read operations. That way it is not necessary to
- * instantiate an object for each Hashmap use. */
+/*
+ * A hash table implementation. As a minor optimization a NULL hashmap object
+ * will be treated as empty hashmap for all read operations. That way it is not
+ * necessary to instantiate an object for each Hashmap use.
+ *
+ * If ENABLE_HASHMAP_DEBUG is defined (by configuring with --enable-hashmap-debug),
+ * the implemention will:
+ * - store extra data for debugging and statistics (see tools/gdb-sd_dump_hashmaps.py)
+ * - perform extra checks for invalid use of iterators
+ */
 
 #define HASH_KEY_SIZE 16
 
-typedef struct Hashmap Hashmap;
-typedef struct OrderedHashmap OrderedHashmap;
-typedef struct _IteratorStruct _IteratorStruct;
-typedef _IteratorStruct* Iterator;
+/* The base type for all hashmap and set types. Many functions in the
+ * implementation take (HashmapBase*) parameters and are run-time polymorphic,
+ * though the API is not meant to be polymorphic (do not call functions
+ * prefixed with two underscores directly). */
+typedef struct HashmapBase HashmapBase;
+
+/* Specific hashmap/set types */
+typedef struct Hashmap Hashmap;               /* Maps keys to values */
+typedef struct OrderedHashmap OrderedHashmap; /* Like Hashmap, but also remembers entry insertion order */
+typedef struct Set Set;                       /* Stores just keys */
+
+/* Ideally the Iterator would be an opaque struct, but it is instantiated
+ * by hashmap users, so the definition has to be here. Do not use its fields
+ * directly. */
+typedef struct {
+        unsigned idx;         /* index of an entry to be iterated next */
+        const void *next_key; /* expected value of that entry's key pointer */
+#ifdef ENABLE_HASHMAP_DEBUG
+        unsigned put_count;   /* hashmap's put_count recorded at start of iteration */
+        unsigned rem_count;   /* hashmap's rem_count in previous iteration */
+        unsigned prev_idx;    /* idx in previous iteration */
+#endif
+} Iterator;
 
-#define ITERATOR_FIRST ((Iterator) 0)
-#define ITERATOR_LAST ((Iterator) -1)
+#define _IDX_ITERATOR_FIRST (UINT_MAX - 1)
+#define ITERATOR_FIRST ((Iterator) { .idx = _IDX_ITERATOR_FIRST, .next_key = NULL })
 
 typedef unsigned long (*hash_func_t)(const void *p, const uint8_t hash_key[HASH_KEY_SIZE]);
 typedef int (*compare_func_t)(const void *a, const void *b);
@@ -82,153 +108,287 @@ extern const struct hash_ops devt_hash_ops = {
 #define devt_hash_ops uint64_hash_ops
 #endif
 
-Hashmap *hashmap_new(const struct hash_ops *hash_ops);
-static inline OrderedHashmap *ordered_hashmap_new(const struct hash_ops *hash_ops) {
-        return (OrderedHashmap*) hashmap_new(hash_ops);
+/* Macros for type checking */
+#define PTR_COMPATIBLE_WITH_HASHMAP_BASE(h) \
+        (__builtin_types_compatible_p(typeof(h), HashmapBase*) || \
+         __builtin_types_compatible_p(typeof(h), Hashmap*) || \
+         __builtin_types_compatible_p(typeof(h), OrderedHashmap*) || \
+         __builtin_types_compatible_p(typeof(h), Set*))
+
+#define PTR_COMPATIBLE_WITH_PLAIN_HASHMAP(h) \
+        (__builtin_types_compatible_p(typeof(h), Hashmap*) || \
+         __builtin_types_compatible_p(typeof(h), OrderedHashmap*)) \
+
+#define HASHMAP_BASE(h) \
+        __builtin_choose_expr(PTR_COMPATIBLE_WITH_HASHMAP_BASE(h), \
+                (HashmapBase*)(h), \
+                (void)0)
+
+#define PLAIN_HASHMAP(h) \
+        __builtin_choose_expr(PTR_COMPATIBLE_WITH_PLAIN_HASHMAP(h), \
+                (Hashmap*)(h), \
+                (void)0)
+
+#ifdef ENABLE_HASHMAP_DEBUG
+# define HASHMAP_DEBUG_PARAMS , const char *func, const char *file, int line
+# define HASHMAP_DEBUG_SRC_ARGS   , __func__, __FILE__, __LINE__
+# define HASHMAP_DEBUG_PASS_ARGS   , func, file, line
+#else
+# define HASHMAP_DEBUG_PARAMS
+# define HASHMAP_DEBUG_SRC_ARGS
+# define HASHMAP_DEBUG_PASS_ARGS
+#endif
+
+Hashmap *internal_hashmap_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS);
+OrderedHashmap *internal_ordered_hashmap_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS);
+#define hashmap_new(ops) internal_hashmap_new(ops  HASHMAP_DEBUG_SRC_ARGS)
+#define ordered_hashmap_new(ops) internal_ordered_hashmap_new(ops  HASHMAP_DEBUG_SRC_ARGS)
+
+void internal_hashmap_free(HashmapBase *h);
+static inline void hashmap_free(Hashmap *h) {
+        internal_hashmap_free(HASHMAP_BASE(h));
 }
-void hashmap_free(Hashmap *h);
 static inline void ordered_hashmap_free(OrderedHashmap *h) {
-        hashmap_free((Hashmap*) h);
+        internal_hashmap_free(HASHMAP_BASE(h));
+}
+
+void internal_hashmap_free_free(HashmapBase *h);
+static inline void hashmap_free_free(Hashmap *h) {
+        internal_hashmap_free_free(HASHMAP_BASE(h));
 }
-void hashmap_free_free(Hashmap *h);
 static inline void ordered_hashmap_free_free(OrderedHashmap *h) {
-        hashmap_free_free((Hashmap*) h);
+        internal_hashmap_free_free(HASHMAP_BASE(h));
 }
+
 void hashmap_free_free_free(Hashmap *h);
 static inline void ordered_hashmap_free_free_free(OrderedHashmap *h) {
-        hashmap_free_free_free((Hashmap*) h);
+        hashmap_free_free_free(PLAIN_HASHMAP(h));
 }
-Hashmap *hashmap_copy(Hashmap *h);
-static inline OrderedHashmap *ordered_hashmap_copy(OrderedHashmap *h) {
-        return (OrderedHashmap*) hashmap_copy((Hashmap*) h);
+
+HashmapBase *internal_hashmap_copy(HashmapBase *h);
+static inline Hashmap *hashmap_copy(Hashmap *h) {
+        return (Hashmap*) internal_hashmap_copy(HASHMAP_BASE(h));
 }
-int hashmap_ensure_allocated(Hashmap **h, const struct hash_ops *hash_ops);
-static inline int ordered_hashmap_ensure_allocated(OrderedHashmap **h, const struct hash_ops *hash_ops) {
-        return hashmap_ensure_allocated((Hashmap**) h, hash_ops);
+static inline OrderedHashmap *ordered_hashmap_copy(OrderedHashmap *h) {
+        return (OrderedHashmap*) internal_hashmap_copy(HASHMAP_BASE(h));
 }
 
+int internal_hashmap_ensure_allocated(Hashmap **h, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS);
+int internal_ordered_hashmap_ensure_allocated(OrderedHashmap **h, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS);
+#define hashmap_ensure_allocated(h, ops) internal_hashmap_ensure_allocated(h, ops  HASHMAP_DEBUG_SRC_ARGS)
+#define ordered_hashmap_ensure_allocated(h, ops) internal_ordered_hashmap_ensure_allocated(h, ops  HASHMAP_DEBUG_SRC_ARGS)
+
 int hashmap_put(Hashmap *h, const void *key, void *value);
 static inline int ordered_hashmap_put(OrderedHashmap *h, const void *key, void *value) {
-        return hashmap_put((Hashmap*) h, key, value);
+        return hashmap_put(PLAIN_HASHMAP(h), key, value);
 }
+
 int hashmap_update(Hashmap *h, const void *key, void *value);
 static inline int ordered_hashmap_update(OrderedHashmap *h, const void *key, void *value) {
-        return hashmap_update((Hashmap*) h, key, value);
+        return hashmap_update(PLAIN_HASHMAP(h), key, value);
 }
+
 int hashmap_replace(Hashmap *h, const void *key, void *value);
 static inline int ordered_hashmap_replace(OrderedHashmap *h, const void *key, void *value) {
-        return hashmap_replace((Hashmap*) h, key, value);
+        return hashmap_replace(PLAIN_HASHMAP(h), key, value);
+}
+
+void *internal_hashmap_get(HashmapBase *h, const void *key);
+static inline void *hashmap_get(Hashmap *h, const void *key) {
+        return internal_hashmap_get(HASHMAP_BASE(h), key);
 }
-void *hashmap_get(Hashmap *h, const void *key);
 static inline void *ordered_hashmap_get(OrderedHashmap *h, const void *key) {
-        return hashmap_get((Hashmap*) h, key);
+        return internal_hashmap_get(HASHMAP_BASE(h), key);
 }
+
 void *hashmap_get2(Hashmap *h, const void *key, void **rkey);
 static inline void *ordered_hashmap_get2(OrderedHashmap *h, const void *key, void **rkey) {
-        return hashmap_get2((Hashmap*) h, key, rkey);
+        return hashmap_get2(PLAIN_HASHMAP(h), key, rkey);
+}
+
+bool internal_hashmap_contains(HashmapBase *h, const void *key);
+static inline bool hashmap_contains(Hashmap *h, const void *key) {
+        return internal_hashmap_contains(HASHMAP_BASE(h), key);
 }
-bool hashmap_contains(Hashmap *h, const void *key);
 static inline bool ordered_hashmap_contains(OrderedHashmap *h, const void *key) {
-        return hashmap_contains((Hashmap*) h, key);
+        return internal_hashmap_contains(HASHMAP_BASE(h), key);
+}
+
+void *internal_hashmap_remove(HashmapBase *h, const void *key);
+static inline void *hashmap_remove(Hashmap *h, const void *key) {
+        return internal_hashmap_remove(HASHMAP_BASE(h), key);
 }
-void *hashmap_remove(Hashmap *h, const void *key);
 static inline void *ordered_hashmap_remove(OrderedHashmap *h, const void *key) {
-        return hashmap_remove((Hashmap*) h, key);
+        return internal_hashmap_remove(HASHMAP_BASE(h), key);
 }
+
 void *hashmap_remove2(Hashmap *h, const void *key, void **rkey);
 static inline void *ordered_hashmap_remove2(OrderedHashmap *h, const void *key, void **rkey) {
-        return hashmap_remove2((Hashmap*) h, key, rkey);
+        return hashmap_remove2(PLAIN_HASHMAP(h), key, rkey);
 }
+
 void *hashmap_remove_value(Hashmap *h, const void *key, void *value);
 static inline void *ordered_hashmap_remove_value(OrderedHashmap *h, const void *key, void *value) {
-        return hashmap_remove_value((Hashmap*) h, key, value);
+        return hashmap_remove_value(PLAIN_HASHMAP(h), key, value);
 }
+
 int hashmap_remove_and_put(Hashmap *h, const void *old_key, const void *new_key, void *value);
 static inline int ordered_hashmap_remove_and_put(OrderedHashmap *h, const void *old_key, const void *new_key, void *value) {
-        return hashmap_remove_and_put((Hashmap*) h, old_key, new_key, value);
+        return hashmap_remove_and_put(PLAIN_HASHMAP(h), old_key, new_key, value);
 }
+
 int hashmap_remove_and_replace(Hashmap *h, const void *old_key, const void *new_key, void *value);
 static inline int ordered_hashmap_remove_and_replace(OrderedHashmap *h, const void *old_key, const void *new_key, void *value) {
-        return hashmap_remove_and_replace((Hashmap*) h, old_key, new_key, value);
+        return hashmap_remove_and_replace(PLAIN_HASHMAP(h), old_key, new_key, value);
 }
 
-int hashmap_merge(Hashmap *h, Hashmap *other);
-static inline int ordered_hashmap_merge(OrderedHashmap *h, OrderedHashmap *other) {
-        return hashmap_merge((Hashmap*) h, (Hashmap*) other);
+/* Since merging data from a OrderedHashmap into a Hashmap or vice-versa
+ * should just work, allow this by having looser type-checking here. */
+int internal_hashmap_merge(Hashmap *h, Hashmap *other);
+#define hashmap_merge(h, other) internal_hashmap_merge(PLAIN_HASHMAP(h), PLAIN_HASHMAP(other))
+#define ordered_hashmap_merge(h, other) hashmap_merge(h, other)
+
+int internal_hashmap_reserve(HashmapBase *h, unsigned entries_add);
+static inline int hashmap_reserve(Hashmap *h, unsigned entries_add) {
+        return internal_hashmap_reserve(HASHMAP_BASE(h), entries_add);
 }
-int hashmap_reserve(Hashmap *h, unsigned entries_add);
 static inline int ordered_hashmap_reserve(OrderedHashmap *h, unsigned entries_add) {
-        return hashmap_reserve((Hashmap*) h, entries_add);
+        return internal_hashmap_reserve(HASHMAP_BASE(h), entries_add);
+}
+
+int internal_hashmap_move(HashmapBase *h, HashmapBase *other);
+/* Unlike hashmap_merge, hashmap_move does not allow mixing the types. */
+static inline int hashmap_move(Hashmap *h, Hashmap *other) {
+        return internal_hashmap_move(HASHMAP_BASE(h), HASHMAP_BASE(other));
 }
-int hashmap_move(Hashmap *h, Hashmap *other);
 static inline int ordered_hashmap_move(OrderedHashmap *h, OrderedHashmap *other) {
-        return hashmap_move((Hashmap*) h, (Hashmap*) other);
+        return internal_hashmap_move(HASHMAP_BASE(h), HASHMAP_BASE(other));
+}
+
+int internal_hashmap_move_one(HashmapBase *h, HashmapBase *other, const void *key);
+static inline int hashmap_move_one(Hashmap *h, Hashmap *other, const void *key) {
+        return internal_hashmap_move_one(HASHMAP_BASE(h), HASHMAP_BASE(other), key);
 }
-int hashmap_move_one(Hashmap *h, Hashmap *other, const void *key);
 static inline int ordered_hashmap_move_one(OrderedHashmap *h, OrderedHashmap *other, const void *key) {
-        return hashmap_move_one((Hashmap*) h, (Hashmap*) other, key);
+        return internal_hashmap_move_one(HASHMAP_BASE(h), HASHMAP_BASE(other), key);
 }
 
-unsigned hashmap_size(Hashmap *h) _pure_;
+unsigned internal_hashmap_size(HashmapBase *h) _pure_;
+static inline unsigned hashmap_size(Hashmap *h) {
+        return internal_hashmap_size(HASHMAP_BASE(h));
+}
 static inline unsigned ordered_hashmap_size(OrderedHashmap *h) {
-        return hashmap_size((Hashmap*) h);
+        return internal_hashmap_size(HASHMAP_BASE(h));
+}
+
+static inline bool hashmap_isempty(Hashmap *h) {
+        return hashmap_size(h) == 0;
 }
-bool hashmap_isempty(Hashmap *h) _pure_;
 static inline bool ordered_hashmap_isempty(OrderedHashmap *h) {
-        return hashmap_isempty((Hashmap*) h);
+        return ordered_hashmap_size(h) == 0;
+}
+
+unsigned internal_hashmap_buckets(HashmapBase *h) _pure_;
+static inline unsigned hashmap_buckets(Hashmap *h) {
+        return internal_hashmap_buckets(HASHMAP_BASE(h));
 }
-unsigned hashmap_buckets(Hashmap *h) _pure_;
 static inline unsigned ordered_hashmap_buckets(OrderedHashmap *h) {
-        return hashmap_buckets((Hashmap*) h);
+        return internal_hashmap_buckets(HASHMAP_BASE(h));
 }
 
-void *hashmap_iterate(Hashmap *h, Iterator *i, const void **key);
+void *internal_hashmap_iterate(HashmapBase *h, Iterator *i, const void **key);
+static inline void *hashmap_iterate(Hashmap *h, Iterator *i, const void **key) {
+        return internal_hashmap_iterate(HASHMAP_BASE(h), i, key);
+}
 static inline void *ordered_hashmap_iterate(OrderedHashmap *h, Iterator *i, const void **key) {
-        return hashmap_iterate((Hashmap*) h, i, key);
+        return internal_hashmap_iterate(HASHMAP_BASE(h), i, key);
 }
 
-void hashmap_clear(Hashmap *h);
+void internal_hashmap_clear(HashmapBase *h);
+static inline void hashmap_clear(Hashmap *h) {
+        internal_hashmap_clear(HASHMAP_BASE(h));
+}
 static inline void ordered_hashmap_clear(OrderedHashmap *h) {
-        hashmap_clear((Hashmap*) h);
+        internal_hashmap_clear(HASHMAP_BASE(h));
+}
+
+void internal_hashmap_clear_free(HashmapBase *h);
+static inline void hashmap_clear_free(Hashmap *h) {
+        internal_hashmap_clear_free(HASHMAP_BASE(h));
 }
-void hashmap_clear_free(Hashmap *h);
 static inline void ordered_hashmap_clear_free(OrderedHashmap *h) {
-        hashmap_clear_free((Hashmap*) h);
+        internal_hashmap_clear_free(HASHMAP_BASE(h));
 }
+
 void hashmap_clear_free_free(Hashmap *h);
 static inline void ordered_hashmap_clear_free_free(OrderedHashmap *h) {
-        hashmap_clear_free_free((Hashmap*) h);
+        hashmap_clear_free_free(PLAIN_HASHMAP(h));
 }
 
-void *hashmap_steal_first(Hashmap *h);
+/*
+ * Note about all *_first*() functions
+ *
+ * For plain Hashmaps and Sets the order of entries is undefined.
+ * The functions find whatever entry is first in the implementation
+ * internal order.
+ *
+ * Only for OrderedHashmaps the order is well defined and finding
+ * the first entry is O(1).
+ */
+
+void *internal_hashmap_steal_first(HashmapBase *h);
+static inline void *hashmap_steal_first(Hashmap *h) {
+        return internal_hashmap_steal_first(HASHMAP_BASE(h));
+}
 static inline void *ordered_hashmap_steal_first(OrderedHashmap *h) {
-        return hashmap_steal_first((Hashmap*) h);
+        return internal_hashmap_steal_first(HASHMAP_BASE(h));
+}
+
+void *internal_hashmap_steal_first_key(HashmapBase *h);
+static inline void *hashmap_steal_first_key(Hashmap *h) {
+        return internal_hashmap_steal_first_key(HASHMAP_BASE(h));
 }
-void *hashmap_steal_first_key(Hashmap *h);
 static inline void *ordered_hashmap_steal_first_key(OrderedHashmap *h) {
-        return hashmap_steal_first_key((Hashmap*) h);
+        return internal_hashmap_steal_first_key(HASHMAP_BASE(h));
 }
-void *hashmap_first(Hashmap *h) _pure_;
-static inline void *ordered_hashmap_first(OrderedHashmap *h) {
-        return hashmap_first((Hashmap*) h);
+
+void *internal_hashmap_first_key(HashmapBase *h) _pure_;
+static inline void *hashmap_first_key(Hashmap *h) {
+        return internal_hashmap_first_key(HASHMAP_BASE(h));
 }
-void *hashmap_first_key(Hashmap *h) _pure_;
 static inline void *ordered_hashmap_first_key(OrderedHashmap *h) {
-        return hashmap_first_key((Hashmap*) h);
+        return internal_hashmap_first_key(HASHMAP_BASE(h));
 }
 
-void *hashmap_next(Hashmap *h, const void *key);
-static inline void *ordered_hashmap_next(OrderedHashmap *h, const void *key) {
-        return hashmap_next((Hashmap*) h, key);
+void *internal_hashmap_first(HashmapBase *h) _pure_;
+static inline void *hashmap_first(Hashmap *h) {
+        return internal_hashmap_first(HASHMAP_BASE(h));
 }
+static inline void *ordered_hashmap_first(OrderedHashmap *h) {
+        return internal_hashmap_first(HASHMAP_BASE(h));
+}
+
+/* no hashmap_next */
+void *ordered_hashmap_next(OrderedHashmap *h, const void *key);
 
-char **hashmap_get_strv(Hashmap *h);
+char **internal_hashmap_get_strv(HashmapBase *h);
+static inline char **hashmap_get_strv(Hashmap *h) {
+        return internal_hashmap_get_strv(HASHMAP_BASE(h));
+}
 static inline char **ordered_hashmap_get_strv(OrderedHashmap *h) {
-        return hashmap_get_strv((Hashmap*) h);
+        return internal_hashmap_get_strv(HASHMAP_BASE(h));
 }
 
+/*
+ * Hashmaps are iterated in unpredictable order.
+ * OrderedHashmaps are an exception to this. They are iterated in the order
+ * the entries were inserted.
+ * It is safe to remove the current entry.
+ */
 #define HASHMAP_FOREACH(e, h, i) \
-        for ((i) = ITERATOR_FIRST, (e) = hashmap_iterate((h), &(i), NULL); (e); (e) = hashmap_iterate((h), &(i), NULL))
+        for ((i) = ITERATOR_FIRST, (e) = hashmap_iterate((h), &(i), NULL); \
+             (e); \
+             (e) = hashmap_iterate((h), &(i), NULL))
 
 #define ORDERED_HASHMAP_FOREACH(e, h, i) \
         for ((i) = ITERATOR_FIRST, (e) = ordered_hashmap_iterate((h), &(i), NULL); \
@@ -236,7 +396,9 @@ static inline char **ordered_hashmap_get_strv(OrderedHashmap *h) {
              (e) = ordered_hashmap_iterate((h), &(i), NULL))
 
 #define HASHMAP_FOREACH_KEY(e, k, h, i) \
-        for ((i) = ITERATOR_FIRST, (e) = hashmap_iterate((h), &(i), (const void**) &(k)); (e); (e) = hashmap_iterate((h), &(i), (const void**) &(k)))
+        for ((i) = ITERATOR_FIRST, (e) = hashmap_iterate((h), &(i), (const void**) &(k)); \
+             (e); \
+             (e) = hashmap_iterate((h), &(i), (const void**) &(k)))
 
 #define ORDERED_HASHMAP_FOREACH_KEY(e, k, h, i) \
         for ((i) = ITERATOR_FIRST, (e) = ordered_hashmap_iterate((h), &(i), (const void**) &(k)); \
@@ -249,6 +411,7 @@ DEFINE_TRIVIAL_CLEANUP_FUNC(Hashmap*, hashmap_free_free_free);
 DEFINE_TRIVIAL_CLEANUP_FUNC(OrderedHashmap*, ordered_hashmap_free);
 DEFINE_TRIVIAL_CLEANUP_FUNC(OrderedHashmap*, ordered_hashmap_free_free);
 DEFINE_TRIVIAL_CLEANUP_FUNC(OrderedHashmap*, ordered_hashmap_free_free_free);
+
 #define _cleanup_hashmap_free_ _cleanup_(hashmap_freep)
 #define _cleanup_hashmap_free_free_ _cleanup_(hashmap_free_freep)
 #define _cleanup_hashmap_free_free_free_ _cleanup_(hashmap_free_free_freep)
diff --git a/src/shared/set.c b/src/shared/set.c
deleted file mode 100644 (file)
index 84ab82a..0000000
+++ /dev/null
@@ -1,166 +0,0 @@
-/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
-
-/***
-  This file is part of systemd.
-
-  Copyright 2010 Lennart Poettering
-
-  systemd is free software; you can redistribute it and/or modify it
-  under the terms of the GNU Lesser General Public License as published by
-  the Free Software Foundation; either version 2.1 of the License, or
-  (at your option) any later version.
-
-  systemd is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
-  Lesser General Public License for more details.
-
-  You should have received a copy of the GNU Lesser General Public License
-  along with systemd; If not, see <http://www.gnu.org/licenses/>.
-***/
-
-#include <stdlib.h>
-
-#include "set.h"
-#include "hashmap.h"
-#include "strv.h"
-
-#define MAKE_SET(h) ((Set*) (h))
-#define MAKE_HASHMAP(s) ((Hashmap*) (s))
-
-/* For now this is not much more than a wrapper around a hashmap */
-
-Set *set_new(const struct hash_ops *hash_ops) {
-        return MAKE_SET(hashmap_new(hash_ops));
-}
-
-void set_free(Set* s) {
-        hashmap_free(MAKE_HASHMAP(s));
-}
-
-void set_free_free(Set *s) {
-        hashmap_free_free(MAKE_HASHMAP(s));
-}
-
-int set_ensure_allocated(Set **s, const struct hash_ops *hash_ops) {
-        return hashmap_ensure_allocated((Hashmap**) s, hash_ops);
-}
-
-int set_put(Set *s, void *value) {
-        return hashmap_put(MAKE_HASHMAP(s), value, value);
-}
-
-int set_consume(Set *s, void *value) {
-        int r;
-
-        r = set_put(s, value);
-        if (r < 0)
-                free(value);
-
-        return r;
-}
-
-int set_put_strdup(Set *s, const char *p) {
-        char *c;
-        int r;
-
-        assert(s);
-        assert(p);
-
-        c = strdup(p);
-        if (!c)
-                return -ENOMEM;
-
-        r = set_consume(s, c);
-        if (r == -EEXIST)
-                return 0;
-
-        return r;
-}
-
-int set_put_strdupv(Set *s, char **l) {
-        int n = 0, r;
-        char **i;
-
-        STRV_FOREACH(i, l) {
-                r = set_put_strdup(s, *i);
-                if (r < 0)
-                        return r;
-
-                n += r;
-        }
-
-        return n;
-}
-
-int set_replace(Set *s, void *value) {
-        return hashmap_replace(MAKE_HASHMAP(s), value, value);
-}
-
-void *set_get(Set *s, void *value) {
-        return hashmap_get(MAKE_HASHMAP(s), value);
-}
-
-bool set_contains(Set *s, void *value) {
-        return hashmap_contains(MAKE_HASHMAP(s), value);
-}
-
-void *set_remove(Set *s, void *value) {
-        return hashmap_remove(MAKE_HASHMAP(s), value);
-}
-
-int set_remove_and_put(Set *s, void *old_value, void *new_value) {
-        return hashmap_remove_and_put(MAKE_HASHMAP(s), old_value, new_value, new_value);
-}
-
-unsigned set_size(Set *s) {
-        return hashmap_size(MAKE_HASHMAP(s));
-}
-
-bool set_isempty(Set *s) {
-        return hashmap_isempty(MAKE_HASHMAP(s));
-}
-
-void *set_iterate(Set *s, Iterator *i) {
-        return hashmap_iterate(MAKE_HASHMAP(s), i, NULL);
-}
-
-void *set_steal_first(Set *s) {
-        return hashmap_steal_first(MAKE_HASHMAP(s));
-}
-
-void* set_first(Set *s) {
-        return hashmap_first(MAKE_HASHMAP(s));
-}
-
-int set_merge(Set *s, Set *other) {
-        return hashmap_merge(MAKE_HASHMAP(s), MAKE_HASHMAP(other));
-}
-
-int set_reserve(Set *s, unsigned entries_add) {
-        return hashmap_reserve(MAKE_HASHMAP(s), entries_add);
-}
-
-int set_move(Set *s, Set *other) {
-        return hashmap_move(MAKE_HASHMAP(s), MAKE_HASHMAP(other));
-}
-
-int set_move_one(Set *s, Set *other, void *value) {
-        return hashmap_move_one(MAKE_HASHMAP(s), MAKE_HASHMAP(other), value);
-}
-
-Set* set_copy(Set *s) {
-        return MAKE_SET(hashmap_copy(MAKE_HASHMAP(s)));
-}
-
-void set_clear(Set *s) {
-        hashmap_clear(MAKE_HASHMAP(s));
-}
-
-void set_clear_free(Set *s) {
-        hashmap_clear_free(MAKE_HASHMAP(s));
-}
-
-char **set_get_strv(Set *s) {
-        return hashmap_get_strv(MAKE_HASHMAP(s));
-}
index d2622d1..4605ecd 100644 (file)
   along with systemd; If not, see <http://www.gnu.org/licenses/>.
 ***/
 
-/* Pretty straightforward set implementation. Internally based on the
- * hashmap. That means that as a minor optimization a NULL set
- * object will be treated as empty set for all read
- * operations. That way it is not necessary to instantiate an object
- * for each set use. */
-
 #include "hashmap.h"
 #include "util.h"
 
-typedef struct Set Set;
+Set *internal_set_new(const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS);
+#define set_new(ops) internal_set_new(ops  HASHMAP_DEBUG_SRC_ARGS)
 
-Set *set_new(const struct hash_ops *hash_ops);
-void set_free(Set* s);
-void set_free_free(Set *s);
 
-Set* set_copy(Set *s);
-int set_ensure_allocated(Set **s, const struct hash_ops *hash_ops);
+static inline void set_free(Set *s) {
+        internal_hashmap_free(HASHMAP_BASE(s));
+}
 
-int set_put(Set *s, void *value);
-int set_consume(Set *s, void *value);
-int set_put_strdup(Set *s, const char *p);
-int set_put_strdupv(Set *s, char **l);
-int set_replace(Set *s, void *value);
-void *set_get(Set *s, void *value);
-bool set_contains(Set *s, void *value);
-void *set_remove(Set *s, void *value);
-int set_remove_and_put(Set *s, void *old_value, void *new_value);
+static inline void set_free_free(Set *s) {
+        internal_hashmap_free_free(HASHMAP_BASE(s));
+}
+
+/* no set_free_free_free */
+
+static inline Set *set_copy(Set *s) {
+        return (Set*) internal_hashmap_copy(HASHMAP_BASE(s));
+}
+
+int internal_set_ensure_allocated(Set **s, const struct hash_ops *hash_ops  HASHMAP_DEBUG_PARAMS);
+#define set_ensure_allocated(h, ops) internal_set_ensure_allocated(h, ops  HASHMAP_DEBUG_SRC_ARGS)
 
+int set_put(Set *s, const void *key);
+/* no set_update */
+/* no set_replace */
+static inline void *set_get(Set *s, void *key) {
+        return internal_hashmap_get(HASHMAP_BASE(s), key);
+}
+/* no set_get2 */
+
+static inline bool set_contains(Set *s, const void *key) {
+        return internal_hashmap_contains(HASHMAP_BASE(s), key);
+}
+
+static inline void *set_remove(Set *s, void *key) {
+        return internal_hashmap_remove(HASHMAP_BASE(s), key);
+}
+
+/* no set_remove2 */
+/* no set_remove_value */
+int set_remove_and_put(Set *s, const void *old_key, const void *new_key);
+/* no set_remove_and_replace */
 int set_merge(Set *s, Set *other);
-int set_reserve(Set *s, unsigned entries_add);
-int set_move(Set *s, Set *other);
-int set_move_one(Set *s, Set *other, void *value);
 
-unsigned set_size(Set *s);
-bool set_isempty(Set *s);
+static inline int set_reserve(Set *h, unsigned entries_add) {
+        return internal_hashmap_reserve(HASHMAP_BASE(h), entries_add);
+}
+
+static inline int set_move(Set *s, Set *other) {
+        return internal_hashmap_move(HASHMAP_BASE(s), HASHMAP_BASE(other));
+}
+
+static inline int set_move_one(Set *s, Set *other, const void *key) {
+        return internal_hashmap_move_one(HASHMAP_BASE(s), HASHMAP_BASE(other), key);
+}
+
+static inline unsigned set_size(Set *s) {
+        return internal_hashmap_size(HASHMAP_BASE(s));
+}
+
+static inline bool set_isempty(Set *s) {
+        return set_size(s) == 0;
+}
+
+static inline unsigned set_buckets(Set *s) {
+        return internal_hashmap_buckets(HASHMAP_BASE(s));
+}
 
 void *set_iterate(Set *s, Iterator *i);
 
-void set_clear(Set *s);
-void set_clear_free(Set *s);
+static inline void set_clear(Set *s) {
+        internal_hashmap_clear(HASHMAP_BASE(s));
+}
+
+static inline void set_clear_free(Set *s) {
+        internal_hashmap_clear_free(HASHMAP_BASE(s));
+}
+
+/* no set_clear_free_free */
+
+static inline void *set_steal_first(Set *s) {
+        return internal_hashmap_steal_first(HASHMAP_BASE(s));
+}
+
+/* no set_steal_first_key */
+/* no set_first_key */
 
-void *set_steal_first(Set *s);
-void* set_first(Set *s);
+static inline void *set_first(Set *s) {
+        return internal_hashmap_first(HASHMAP_BASE(s));
+}
 
-char **set_get_strv(Set *s);
+/* no set_next */
+
+static inline char **set_get_strv(Set *s) {
+        return internal_hashmap_get_strv(HASHMAP_BASE(s));
+}
+
+int set_consume(Set *s, void *value);
+int set_put_strdup(Set *s, const char *p);
+int set_put_strdupv(Set *s, char **l);
 
 #define SET_FOREACH(e, s, i) \
         for ((i) = ITERATOR_FIRST, (e) = set_iterate((s), &(i)); (e); (e) = set_iterate((s), &(i)))
 
 DEFINE_TRIVIAL_CLEANUP_FUNC(Set*, set_free);
 DEFINE_TRIVIAL_CLEANUP_FUNC(Set*, set_free_free);
+
 #define _cleanup_set_free_ _cleanup_(set_freep)
 #define _cleanup_set_free_free_ _cleanup_(set_free_freep)