From 7d3e2bd2c4cfd1296d1d6f996d7548de26540d41 Mon Sep 17 00:00:00 2001 From: Josh Haberman Date: Fri, 15 Feb 2013 16:27:18 -0800 Subject: Sync with 8 months of Google-internal development. Many things have changed and been simplified. The memory-management story for upb_def and upb_handlers is much more robust; upb_def and upb_handlers should be fairly stable interfaces now. There is still much work to do for the runtime component (upb_sink). --- upb/table.h | 103 +++++++++++++++++++++++++++++++++++++++++++++--------------- 1 file changed, 78 insertions(+), 25 deletions(-) (limited to 'upb/table.h') diff --git a/upb/table.h b/upb/table.h index f6bff66..80f6813 100644 --- a/upb/table.h +++ b/upb/table.h @@ -17,13 +17,16 @@ * * This header is internal to upb; its interface should not be considered * public or stable. + * + * The table must be homogenous (all values of the same type). We currently + * enforce this on insert but store the full upb_value (with type) anyway. + * This is required with the current interface because lookups vend a pointer + * to the table's internal storage. */ #ifndef UPB_TABLE_H_ #define UPB_TABLE_H_ -#include -#include #include "upb.h" #ifdef __cplusplus @@ -32,45 +35,80 @@ extern "C" { typedef union { uintptr_t num; - char *str; // We own, nullz. + const char *str; // We own, nullz. } upb_tabkey; +#define UPB_TABKEY_NUM(n) {n} +#ifdef UPB_C99 +#define UPB_TABKEY_STR(s) {.str = s} +#endif +// TODO(haberman): C++ +#define UPB_TABKEY_NONE {0} + typedef struct _upb_tabent { upb_tabkey key; + // Storing a upb_value here wastes a bit of memory in debug mode because + // we are storing the type for each value even though we enforce that all + // values are the same. But since this only affects debug mode, we don't + // worry too much about it. The same applies to upb_inttable.array below. upb_value val; - struct _upb_tabent *next; // Internal chaining. + // Internal chaining. This is const so we can create static initializers for + // tables. We cast away const sometimes, but *only* when the containing + // upb_table is known to be non-const. This requires a bit of care, but + // the subtlety is confined to table.c. + const struct _upb_tabent *next; } upb_tabent; typedef struct { - upb_tabent *entries; // Hash table. size_t count; // Number of entries in the hash part. size_t mask; // Mask to turn hash value -> bucket. + upb_ctype_t type; // Type of all values. uint8_t size_lg2; // Size of the hash table part is 2^size_lg2 entries. + const upb_tabent *entries; // Hash table. } upb_table; typedef struct { upb_table t; } upb_strtable; +#define UPB_STRTABLE_INIT(count, mask, type, size_lg2, entries) \ + {{count, mask, type, size_lg2, entries}} + typedef struct { - upb_table t; // For entries that don't fit in the array part. - upb_value *array; // Array part of the table. - size_t array_size; // Array part size. - size_t array_count; // Array part number of elements. + upb_table t; // For entries that don't fit in the array part. + const upb_value *array; // Array part of the table. + size_t array_size; // Array part size. + size_t array_count; // Array part number of elements. } upb_inttable; -INLINE upb_tabkey upb_intkey(uintptr_t key) { upb_tabkey k = {key}; return k; } +#define UPB_INTTABLE_INIT(count, mask, type, size_lg2, ent, a, asize, acount) \ + {{count, mask, type, size_lg2, ent}, a, asize, acount} -INLINE upb_tabent *upb_inthash(const upb_table *t, upb_tabkey key) { - return t->entries + ((uint32_t)key.num & t->mask); +#define UPB_EMPTY_INTTABLE_INIT(type) \ + UPB_INTTABLE_INIT(0, 0, type, 0, NULL, NULL, 0, 0) + +#define UPB_ARRAY_EMPTYENT UPB_VALUE_INIT_INT64(-1) + +INLINE size_t upb_table_size(const upb_table *t) { + if (t->size_lg2 == 0) + return 0; + else + return 1 << t->size_lg2; } +// Internal-only functions, in .h file only out of necessity. +INLINE bool upb_tabent_isempty(const upb_tabent *e) { return e->key.num == 0; } +INLINE upb_tabkey upb_intkey(uintptr_t key) { upb_tabkey k = {key}; return k; } +INLINE const upb_tabent *upb_inthash(const upb_table *t, upb_tabkey key) { + return t->entries + ((uint32_t)key.num & t->mask); +} INLINE bool upb_arrhas(upb_value v) { return v.val.uint64 != (uint64_t)-1; } +uint32_t MurmurHash2(const void *key, size_t len, uint32_t seed); // Initialize and uninitialize a table, respectively. If memory allocation // failed, false is returned that the table is uninitialized. -bool upb_inttable_init(upb_inttable *table); -bool upb_strtable_init(upb_strtable *table); +bool upb_inttable_init(upb_inttable *table, upb_ctype_t type); +bool upb_strtable_init(upb_strtable *table, upb_ctype_t type); void upb_inttable_uninit(upb_inttable *table); void upb_strtable_uninit(upb_strtable *table); @@ -90,14 +128,24 @@ bool upb_strtable_insert(upb_strtable *t, const char *key, upb_value val); // Looks up key in this table, returning a pointer to the table's internal copy // of the user's inserted data, or NULL if this key is not in the table. The -// user is free to modify the given upb_value, which will be reflected in any -// future lookups of this key. The returned pointer is invalidated by inserts. -upb_value *upb_inttable_lookup(const upb_inttable *t, uintptr_t key); -upb_value *upb_strtable_lookup(const upb_strtable *t, const char *key); +// returned pointer is invalidated by inserts. +const upb_value *upb_inttable_lookup(const upb_inttable *t, uintptr_t key); +const upb_value *upb_strtable_lookup(const upb_strtable *t, const char *key); // Removes an item from the table. Returns true if the remove was successful, // and stores the removed item in *val if non-NULL. bool upb_inttable_remove(upb_inttable *t, uintptr_t key, upb_value *val); +bool upb_strtable_remove(upb_strtable *t, const char *key, upb_value *val); + +// Handy routines for treating an inttable like a stack. May not be mixed with +// other insert/remove calls. +bool upb_inttable_push(upb_inttable *t, upb_value val); +upb_value upb_inttable_pop(upb_inttable *t); + +// Convenience routines for inttables with pointer keys. +bool upb_inttable_insertptr(upb_inttable *t, const void *key, upb_value val); +bool upb_inttable_removeptr(upb_inttable *t, const void *key, upb_value *val); +const upb_value *upb_inttable_lookupptr(const upb_inttable *t, const void *key); // Optimizes the table for the current set of entries, for both memory use and // lookup time. Client should call this after all entries have been inserted; @@ -105,12 +153,15 @@ bool upb_inttable_remove(upb_inttable *t, uintptr_t key, upb_value *val); void upb_inttable_compact(upb_inttable *t); // A special-case inlinable version of the lookup routine for 32-bit integers. -INLINE upb_value *upb_inttable_lookup32(const upb_inttable *t, uint32_t key) { +INLINE const upb_value *upb_inttable_lookup32(const upb_inttable *t, + uint32_t key) { if (key < t->array_size) { - upb_value *v = &t->array[key]; + const upb_value *v = &t->array[key]; return upb_arrhas(*v) ? v : NULL; } - for (upb_tabent *e = upb_inthash(&t->t, upb_intkey(key)); true; e = e->next) { + const upb_tabent *e; + if (t->t.entries == NULL) return NULL; + for (e = upb_inthash(&t->t, upb_intkey(key)); true; e = e->next) { if ((uint32_t)e->key.num == key) return &e->val; if (e->next == NULL) return NULL; } @@ -124,12 +175,12 @@ INLINE upb_value *upb_inttable_lookup32(const upb_inttable *t, uint32_t key) { // upb_strtable_begin(&i, t); // for(; !upb_strtable_done(&i); upb_strtable_next(&i)) { // const char *key = upb_strtable_iter_key(&i); -// const myval *val = upb_strtable_iter_value(&i); +// const upb_value val = upb_strtable_iter_value(&i); // // ... // } typedef struct { const upb_strtable *t; - upb_tabent *e; + const upb_tabent *e; } upb_strtable_iter; void upb_strtable_begin(upb_strtable_iter *i, const upb_strtable *t); @@ -149,13 +200,15 @@ INLINE upb_value upb_strtable_iter_value(upb_strtable_iter *i) { // upb_inttable_iter i; // upb_inttable_begin(&i, t); // for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { +// uintptr_t key = upb_inttable_iter_key(&i); +// upb_value val = upb_inttable_iter_value(&i); // // ... // } typedef struct { const upb_inttable *t; union { - upb_tabent *ent; // For hash iteration. - upb_value *val; // For array iteration. + const upb_tabent *ent; // For hash iteration. + const upb_value *val; // For array iteration. } ptr; uintptr_t arrkey; bool array_part; -- cgit v1.2.3