Jesper Louis Andersen wrote:
I am not sure this is the correct mailing list. But the old Data.FiniteMap implementation states for the function ``addToFM'' (in the haddock documentation):
Adds an element to a FiniteMap. Any previous mapping with the same key is overwritten.
Whereas the equivalent ``insert'' in the new Data.Map module States:
O(log n). Insert a new key and value in the map.
You are right, this is not very clear. Can someone on this list update the documentation? The libraries for Data.Map, Data.Set, Data.IntMap, etc. are consistently using the following two rules for all operations: 1) operations are left-biased 2) the data structure comes as a last parameter Due to (2), the signature of "insert" takes the "Map" as its last parameter, and therefore, it overwrites an existing value (due to (1)). Same holds for "union" for example. I hope this helps, -- Daan Leijen.
---
I think it is good that there now is a running time asymptotic bound, but I do not know what happens if the (key, value) pair is already there. There is no information about this in the haddock documentation at all. It could either overwrite silently, or it could raise some kind of error. Both options are plausible, since one could just use ``Data.Map.adjust'' like:
f :: Ord k -> k -> a -> Map k a -> Map k a f k a fm = Data.Map.adjust (\_ -> a) k fm
And gain the explicit behaviour. I do not currently have a GHC 6.4 installed here, so I cannot test this behaviour, but some code I am working on needs to be fairly compatible. Therefore we have a layer on top of Data.FiniteMap which implements the new map structure. This difference just caught my eye. It also makes it rather hard to update the documentation myself and provide a patch. Where is the official repository located?