Move Dictionary/PDict, List/PList, and Queue/PQueue to zeek namespace

This commit is contained in:
Tim Wojtulewicz 2020-06-23 13:25:43 -07:00
parent 4a1e17f4e0
commit 464efbe66a
30 changed files with 172 additions and 138 deletions

View file

@ -13,17 +13,18 @@
// If the mean bucket length exceeds the following then Insert() will
// increase the size of the hash table.
#define DEFAULT_DENSITY_THRESH 3.0
constexpr double DEFAULT_DENSITY_THRESH = 3.0;
// Threshold above which we do not try to ensure that the hash size
// is prime.
#define PRIME_THRESH 1000
constexpr int PRIME_THRESH = 1000;
// Default number of hash buckets in dictionary. The dictionary will
// increase the size of the hash table as needed.
#define DEFAULT_DICT_SIZE 16
constexpr int DEFAULT_DICT_SIZE = 16;
TEST_SUITE_BEGIN("Dict");
namespace zeek {
namespace detail {
class DictEntry {
public:
@ -40,6 +41,8 @@ public:
void* value;
};
} //namespace detail
// The value of an iteration cookie is the bucket and offset within the
// bucket at which to start looking for the next value to return.
class IterCookie {
@ -47,25 +50,29 @@ public:
IterCookie(int b, int o) : bucket(b), offset(o) {}
int bucket, offset;
PList<DictEntry>** ttbl = nullptr;
zeek::PList<detail::DictEntry>** ttbl = nullptr;
const int* num_buckets_p = nullptr;
PList<DictEntry> inserted; // inserted while iterating
zeek::PList<detail::DictEntry> inserted; // inserted while iterating
};
} // namespace zeek
TEST_SUITE_BEGIN("Dict");
TEST_CASE("dict construction")
{
PDict<int> dict;
zeek::PDict<int> dict;
CHECK(dict.IsOrdered() == false);
CHECK(dict.Length() == 0);
PDict<int> dict2(ORDERED);
zeek::PDict<int> dict2(zeek::ORDERED);
CHECK(dict2.IsOrdered() == true);
CHECK(dict2.Length() == 0);
}
TEST_CASE("dict operation")
{
PDict<uint32_t> dict;
zeek::PDict<uint32_t> dict;
uint32_t val = 10;
uint32_t key_val = 5;
@ -111,8 +118,8 @@ TEST_CASE("dict operation")
TEST_CASE("dict nthentry")
{
PDict<uint32_t> unordered(UNORDERED);
PDict<uint32_t> ordered(ORDERED);
zeek::PDict<uint32_t> unordered(zeek::UNORDERED);
zeek::PDict<uint32_t> ordered(zeek::ORDERED);
uint32_t val = 15;
uint32_t key_val = 5;
@ -147,7 +154,7 @@ TEST_CASE("dict nthentry")
TEST_CASE("dict iteration")
{
PDict<uint32_t> dict;
zeek::PDict<uint32_t> dict;
uint32_t val = 15;
uint32_t key_val = 5;
@ -161,7 +168,7 @@ TEST_CASE("dict iteration")
dict.Insert(key2, &val2);
HashKey* it_key;
IterCookie* it = dict.InitForIteration();
zeek::IterCookie* it = dict.InitForIteration();
CHECK(it != nullptr);
int count = 0;
@ -186,10 +193,14 @@ TEST_CASE("dict iteration")
delete key2;
}
Dictionary::Dictionary(dict_order ordering, int initial_size)
TEST_SUITE_END();
namespace zeek {
Dictionary::Dictionary(DictOrder ordering, int initial_size)
{
if ( ordering == ORDERED )
order = new PList<DictEntry>;
order = new zeek::PList<detail::DictEntry>;
if ( initial_size > 0 )
Init(initial_size);
@ -218,7 +229,7 @@ void Dictionary::DeInit()
for ( int i = 0; i < num_buckets; ++i )
if ( tbl[i] )
{
PList<DictEntry>* chain = tbl[i];
zeek::PList<detail::DictEntry>* chain = tbl[i];
for ( const auto& e : *chain )
{
if ( delete_func )
@ -238,7 +249,7 @@ void Dictionary::DeInit()
for ( int i = 0; i < num_buckets2; ++i )
if ( tbl2[i] )
{
PList<DictEntry>* chain = tbl2[i];
zeek::PList<detail::DictEntry>* chain = tbl2[i];
for ( const auto& e : *chain )
{
if ( delete_func )
@ -259,7 +270,7 @@ void* Dictionary::Lookup(const void* key, int key_size, hash_t hash) const
return nullptr;
hash_t h;
PList<DictEntry>* chain;
zeek::PList<detail::DictEntry>* chain;
// Figure out which hash table to look in.
h = hash % num_buckets;
@ -287,12 +298,12 @@ void* Dictionary::Insert(void* key, int key_size, hash_t hash, void* val,
if ( ! tbl )
Init(DEFAULT_DICT_SIZE);
DictEntry* new_entry = new DictEntry(key, key_size, hash, val);
detail::DictEntry* new_entry = new detail::DictEntry(key, key_size, hash, val);
void* old_val = Insert(new_entry, copy_key);
if ( old_val )
{
// We didn't need the new DictEntry, the key was already
// We didn't need the new detail::DictEntry, the key was already
// present.
delete new_entry;
}
@ -315,7 +326,7 @@ void* Dictionary::Remove(const void* key, int key_size, hash_t hash,
return nullptr;
hash_t h;
PList<DictEntry>* chain;
zeek::PList<detail::DictEntry>* chain;
int* num_entries_ptr;
// Figure out which hash table to look in
@ -338,7 +349,7 @@ void* Dictionary::Remove(const void* key, int key_size, hash_t hash,
for ( auto i = 0u; i < chain_length; ++i )
{
DictEntry* entry = (*chain)[i];
detail::DictEntry* entry = (*chain)[i];
if ( entry->hash == hash && entry->len == key_size &&
! memcmp(key, entry->key, key_size) )
@ -357,8 +368,8 @@ void* Dictionary::Remove(const void* key, int key_size, hash_t hash,
return nullptr;
}
void* Dictionary::DoRemove(DictEntry* entry, hash_t h,
PList<DictEntry>* chain, int chain_offset)
void* Dictionary::DoRemove(detail::DictEntry* entry, hash_t h,
zeek::PList<detail::DictEntry>* chain, int chain_offset)
{
void* entry_value = entry->value;
@ -397,7 +408,7 @@ void* Dictionary::NthEntry(int n, const void*& key, int& key_len) const
if ( ! order || n < 0 || n >= Length() )
return nullptr;
DictEntry* entry = (*order)[n];
detail::DictEntry* entry = (*order)[n];
key = entry->key;
key_len = entry->len;
return entry->value;
@ -417,7 +428,7 @@ void* Dictionary::NextEntry(HashKey*& h, IterCookie*& cookie, int return_hash) c
{
if ( ! tbl && ! tbl2 )
{
const_cast<PList<IterCookie>*>(&cookies)->remove(cookie);
const_cast<zeek::PList<IterCookie>*>(&cookies)->remove(cookie);
delete cookie;
cookie = nullptr;
return nullptr;
@ -427,7 +438,7 @@ void* Dictionary::NextEntry(HashKey*& h, IterCookie*& cookie, int return_hash) c
// That keeps the list small and helps avoiding searching
// a large list when deleting an entry.
DictEntry* entry;
detail::DictEntry* entry;
if ( cookie->inserted.length() )
{
@ -442,7 +453,7 @@ void* Dictionary::NextEntry(HashKey*& h, IterCookie*& cookie, int return_hash) c
int b = cookie->bucket;
int o = cookie->offset;
PList<DictEntry>** ttbl;
zeek::PList<detail::DictEntry>** ttbl;
const int* num_buckets_p;
if ( ! cookie->ttbl )
@ -484,7 +495,7 @@ void* Dictionary::NextEntry(HashKey*& h, IterCookie*& cookie, int return_hash) c
// FIXME: I don't like removing the const here. But is there
// a better way?
const_cast<PList<IterCookie>*>(&cookies)->remove(cookie);
const_cast<zeek::PList<IterCookie>*>(&cookies)->remove(cookie);
delete cookie;
cookie = nullptr;
return nullptr;
@ -503,7 +514,7 @@ void* Dictionary::NextEntry(HashKey*& h, IterCookie*& cookie, int return_hash) c
void Dictionary::Init(int size)
{
num_buckets = NextPrime(size);
tbl = new PList<DictEntry>*[num_buckets];
tbl = new zeek::PList<detail::DictEntry>*[num_buckets];
for ( int i = 0; i < num_buckets; ++i )
tbl[i] = nullptr;
@ -515,7 +526,7 @@ void Dictionary::Init(int size)
void Dictionary::Init2(int size)
{
num_buckets2 = NextPrime(size);
tbl2 = new PList<DictEntry>*[num_buckets2];
tbl2 = new zeek::PList<detail::DictEntry>*[num_buckets2];
for ( int i = 0; i < num_buckets2; ++i )
tbl2[i] = nullptr;
@ -524,12 +535,12 @@ void Dictionary::Init2(int size)
}
// private
void* Dictionary::Insert(DictEntry* new_entry, bool copy_key)
void* Dictionary::Insert(detail::DictEntry* new_entry, bool copy_key)
{
if ( ! tbl )
Init(DEFAULT_DICT_SIZE);
PList<DictEntry>** ttbl;
zeek::PList<detail::DictEntry>** ttbl;
int* num_entries_ptr;
int* max_num_entries_ptr;
hash_t h = new_entry->hash % num_buckets;
@ -554,7 +565,7 @@ void* Dictionary::Insert(DictEntry* new_entry, bool copy_key)
max_num_entries_ptr = &max_num_entries2;
}
PList<DictEntry>* chain = ttbl[h];
zeek::PList<detail::DictEntry>* chain = ttbl[h];
int n = new_entry->len;
@ -562,7 +573,7 @@ void* Dictionary::Insert(DictEntry* new_entry, bool copy_key)
{
for ( int i = 0; i < chain->length(); ++i )
{
DictEntry* entry = (*chain)[i];
detail::DictEntry* entry = (*chain)[i];
if ( entry->hash == new_entry->hash &&
entry->len == n &&
@ -576,7 +587,7 @@ void* Dictionary::Insert(DictEntry* new_entry, bool copy_key)
}
else
// Create new chain.
chain = ttbl[h] = new PList<DictEntry>;
chain = ttbl[h] = new zeek::PList<detail::DictEntry>;
// If we got this far, then we couldn't use an existing copy
// of the key, so make a new one if necessary.
@ -660,7 +671,7 @@ void Dictionary::MoveChains()
do
{
PList<DictEntry>* chain = tbl[tbl_next_ind++];
zeek::PList<detail::DictEntry>* chain = tbl[tbl_next_ind++];
if ( ! chain )
continue;
@ -720,13 +731,13 @@ unsigned int Dictionary::MemoryAllocation() const
for ( int i = 0; i < num_buckets; ++i )
if ( tbl[i] )
{
PList<DictEntry>* chain = tbl[i];
zeek::PList<detail::DictEntry>* chain = tbl[i];
for ( const auto& c : *chain )
size += padded_sizeof(DictEntry) + pad_size(c->len);
size += padded_sizeof(detail::DictEntry) + pad_size(c->len);
size += chain->MemoryAllocation();
}
size += pad_size(num_buckets * sizeof(PList<DictEntry>*));
size += pad_size(num_buckets * sizeof(zeek::PList<detail::DictEntry>*));
if ( order )
size += order->MemoryAllocation();
@ -736,13 +747,13 @@ unsigned int Dictionary::MemoryAllocation() const
for ( int i = 0; i < num_buckets2; ++i )
if ( tbl2[i] )
{
PList<DictEntry>* chain = tbl2[i];
zeek::PList<detail::DictEntry>* chain = tbl2[i];
for ( const auto& c : *chain )
size += padded_sizeof(DictEntry) + pad_size(c->len);
size += padded_sizeof(detail::DictEntry) + pad_size(c->len);
size += chain->MemoryAllocation();
}
size += pad_size(num_buckets2 * sizeof(PList<DictEntry>*));
size += pad_size(num_buckets2 * sizeof(zeek::PList<detail::DictEntry>*));
}
return size;
@ -753,4 +764,4 @@ void generic_delete_func(void* v)
free(v);
}
TEST_SUITE_END();
} // namespace zeek