Hello Richard, On 2014-04-24 at 15:04:55 +0200, Richard Eisenberg wrote:

That map seems to store the set of variables during printing TH, for the purposes of disambiguating identifiers with the same name but different uniques. If blatting out a whole lot of program text, I could imagine the Map getting somewhat sizeable.

When does printing TH actually occur? If it doesn't occur during regular compilation, do we ever need to pretty print large amounts of TH?

But, it seems to only need the lookup and insert operations...

actually, it uses a specific combination of lookup+insert, so that it would suffice to have the single operation in the style of Python's dict.setdefault(), i.e. findWithDefaultInsert :: Ord k => k -> v -> Map k v -> (v, Map k v) findWithDefaultInsert k default m | Just v <- Map.lookup k m = (v, m) | otherwise = default `seq` (default, Map.insert k default m)

is there a simpler data structure that has only these operations efficiently?

Really, the patch is fine just as it is probably. Earlier me and Herbert were wondering if anyone would actually be affected by the O(n) lookup, but the only reason *I* really cared to know that is just so we don't have to inline part of `containers` if we don't have to. But at the end of the day it's barely 100 lines of extra code. So maybe not that big a deal (but kinda lame). To be on the conservative side and not regress (and this could easily be considered a regression from a performance POV) we can certainly just commit this and see if we can make the code smaller later, I'd say. On Fri, Apr 25, 2014 at 8:33 AM, Richard Eisenberg wrote:

On Apr 25, 2014, at 4:16 AM, Herbert Valerio Riedel wrote:

...

Hello Richard,

On 2014-04-24 at 15:04:55 +0200, Richard Eisenberg wrote:

...

That map seems to store the set of variables during printing TH, for the purposes of disambiguating identifiers with the same name but different uniques. If blatting out a whole lot of program text, I could imagine the Map getting somewhat sizeable.

When does printing TH actually occur? If it doesn't occur during regular compilation, do we ever need to pretty print large amounts of TH?

I guess it depends on who “we” are. In the process of routine TH hackery, I would say that printing is uncommon -- normally the TH is processed and then just spliced, without user interaction. But, I can conceive of a library that does some processing and then prints.

...

...

But, it seems to only need the lookup and insert operations...

Good observation.

...

actually, it uses a specific combination of lookup+insert, so that it would suffice to have the single operation in the style of Python's dict.setdefault(), i.e.

findWithDefaultInsert :: Ord k => k -> v -> Map k v -> (v, Map k v) findWithDefaultInsert k default m | Just v <- Map.lookup k m = (v, m) | otherwise = default `seq` (default, Map.insert k default m)

...

is there a simpler data structure that has only these operations efficiently?

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

-- Regards, Austin Seipp, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/