Module sharedtables

Shared table support for Nim. Use plain old non GC'ed keys and values or you'll be in trouble. Uses a single lock to protect the table, lockfree implementations welcome but if lock contention is so high that you need a lockfree hash table, you're doing it wrong.

Imports

hashes, math, locks

Types

SharedTable[A; B] = object
  data: KeyValuePairSeq[A, B]
  counter, dataLen: int
  lock: Lock

generic hash SharedTable

Procs

proc mget[A, B](t: var SharedTable[A, B]; key: A): var B

retrieves the value at t[key]. The value can be modified. If key is not in t, the KeyError exception is raised.

proc mgetOrPut[A, B](t: var SharedTable[A, B]; key: A; val: B): var B

retrieves value at t[key] or puts val if not present, either way returning a value which can be modified. Note: This is inherently unsafe in the context of multi-threading since it returns a pointer to B.

proc hasKeyOrPut[A, B](t: var SharedTable[A, B]; key: A; val: B): bool

returns true iff key is in the table, otherwise inserts value.

proc withKey[A, B](t: var SharedTable[A, B]; key: A;
                 mapper: proc (key: A; val: var B; pairExists: var bool))

Computes a new mapping for the key with the specified mapper procedure.

The mapper takes 3 arguments:

1. key - the current key, if it exists, or the key passed to withKey otherwise;
2. val - the current value, if the key exists, or default value of the type otherwise;
3. pairExists - true if the key exists, false otherwise.

The mapper can can modify val and pairExists values to change the mapping of the key or delete it from the table. When adding a value, make sure to set pairExists to true along with modifying the val.

The operation is performed atomically and other operations on the table will be blocked while the mapper is invoked, so it should be short and simple.

Example usage:

# If value exists, decrement it.
# If it becomes zero or less, delete the key
t.withKey(1'i64) do (k: int64, v: var int, pairExists: var bool):
  if pairExists:
    dec v
    if v <= 0:
      pairExists = false

proc `[]=`[A, B](t: var SharedTable[A, B]; key: A; val: B)

puts a (key, value)-pair into t.

proc add[A, B](t: var SharedTable[A, B]; key: A; val: B)

puts a new (key, value)-pair into t even if t[key] already exists. This can introduce duplicate keys into the table!

proc del[A, B](t: var SharedTable[A, B]; key: A)

deletes key from hash table t.

proc init[A, B](t: var SharedTable[A, B]; initialSize = 64)

creates a new hash table that is empty.

initialSize needs to be a power of two. If you need to accept runtime values for this you could use the nextPowerOfTwo proc from the math module or the rightSize proc from this module.

proc deinitSharedTable[A, B](t: var SharedTable[A, B])

proc initSharedTable[A, B](initialSize = 64): SharedTable[A, B] {...}{.deprecated.}

Deprecated. Use init instead. This is not posix compliant, may introduce undefined behavior.

Templates

template withValue[A; B](t: var SharedTable[A, B]; key: A; value, body: untyped)

retrieves the value at t[key]. value can be modified in the scope of the withValue call.

sharedTable.withValue(key, value) do:
  # block is executed only if ``key`` in ``t``
  # value is threadsafe in block
  value.name = "username"
  value.uid = 1000

template withValue[A; B](t: var SharedTable[A, B]; key: A; value, body1, body2: untyped)

retrieves the value at t[key]. value can be modified in the scope of the withValue call.

sharedTable.withValue(key, value) do:
  # block is executed only if ``key`` in ``t``
  # value is threadsafe in block
  value.name = "username"
  value.uid = 1000
do:
  # block is executed when ``key`` not in ``t``
  raise newException(KeyError, "Key not found")