[vector] #79: Please provide memory-footprint "complexity" information in Haddock documentation

[vector]

#79: Please provide memory-footprint "complexity" information in Haddock
documentation
------------------------+---------------------------------------------------
Reporter:  hvr          |        Owner:     
    Type:  enhancement  |       Status:  new
Priority:  minor        |    Milestone:     
 Version:               |   Resolution:     
Keywords:               |  
------------------------+---------------------------------------------------

Old description:

> I've been surprised lately by the base-overhead of `Data.Vector.Vector`,
> as I assumed it to be somewhere near 4+N words, but it was actually more
> like 9+N words or so for N < 128, which is a lot on 64bit archs.
>
> IMHO, it's helpful to be aware of the cost-model in advance, instead of
> having to find out the hard way via heap-profiling... :-)

New description:

 I've been surprised lately by the base-overhead of `Data.Vector.Vector`,
 as I assumed it to be somewhere near 4+N words, but it was actually more
 like 9+N words or so for N < 128, which is a lot on 64bit archs.

 IMHO, it's helpful to be aware of the cost-model in advance, instead of
 having to find out the hard way via heap-profiling... :-)

--

Comment(by rl):

 Note that GHC allocates additional space for boxed arrays which it uses to
 mark the bits that have changed since the last GC. Updating a mutable
 array would be an O(n) operation otherwise since the GC would have to
 rescan the entire array. So vector should indeed add around 4 words to the
 array, it's GHC that allocates more than N and for good reasons! I
 wouldn't really want to precisely document these GHC internals in vector,
 it seems like the wrong thing to do. Do you have an idea for something
 more general?

-- 
Ticket URL: <http://trac.haskell.org/vector/ticket/79#comment:1>
vector <http://trac.haskell.org/vector>
Package vector