Re: C Hashmap implementation

"bison" <Sean.Gillespie@bisonofborg.com> wrote in message
> Hello. I'm looking into hashmap implementations for a VM based
> dynamically typed language. Since the implementation must allow me to
> resize on demand, I am a bit confused about the various approaches.
> Here is what I have found out so far:
>
> Almost everyone I've talked to has said that Chained Hashmaps are much
> easier to implement than Open Addressed maps.


I would agree here.
there are some cases where they make a lot more sense as well.




> Wikipedia suggests that an approach to resizing hashmaps is to
> allocate space for a newer hashmap and copy elements from to the new
> table, and in some cases do it incrementally.


yes.
but then you have 2 hashes, one of which is around until the old one is
emptied.


sometimes one can resize the table though, and use a gradiated scheme:
first try the largest hash;
if failed, try the smaller hash;
....


when done copying the entry to the largest hash.
though not tried personally, this could have contention issues though (esp
if one uses a permutative lookup scheme, where potentially removing an entry
could eliminate existing entries, implying that one can't simply stop when
they encounter an empty slot, but must first check some n permutations).




> Quick question about the last point: I'm curious about a starting
> point. How much space should a hashmap allocate initially, and when
> it's full...by how much would I typically increase it? I realize
> there are lots of different answers here, so a good starting point
> would really help out.
>


for resizable things, usually I start out with a low-max guestimate, say 256
to 4096.
usually I expand in increments of 50%, so:
sz'=sz+(sz>>1);


in practice, this seems a fairly good factor (avoids extra expansions while
still usually keeping a fairly tight fit on the data, vs say, doubling the
size each time).


if being more conservative, a 4/3 factor might be better than a 3/2 factor,
but I don't know...

Return to the comp.compilers page.
Search the comp.compilers archives again.