[Haskell-cafe] Significant compilation slowdown when using Data.CaseInsensitive.mk

[Brendan Hay]

Hi,

I've recently run into an issue whereby changing textual case comparison
to use the Data.CaseInsensitive.mk
<http://hackage.haskell.org/package/case-insensitive-1.2.0.3/docs/Data-CaseInsensitive.html>
smart
constructor causes substantial
differences to outputted core language and a related compilation speed
regression.

The code which seems to cause this is here:
https://github.com/brendanhay/amazonka/blob/release/0.1.4/core/src/Network/AWS/Data/Internal/Text.hs#L51

With the downstream project having about 68 call sites for the above
function starting here:
https://github.com/brendanhay/amazonka/blob/release/0.1.4/amazonka-ec2/gen/Network/AWS/EC2/Types.hs#L1465

For comparison, the previous version used Attoparsec.Text.takeText
<http://hackage.haskell.org/package/attoparsec-0.12.1.2/docs/Data-Attoparsec-Text.html>
and Text values.

Here are some hand wavy points of interest:

* Before (Case Sensitive)
Function: takeText = Attoparsec.Text.takeText
Time (real): 2m27.735s
amazonka-ec2/gen/Network/AWS/EC2/Types.hs (simpl): {terms: 242,688, types:
310,463, coercions: 31,215}

* After (Case Insensitive)
Function: takeCI = Data.CaseInsensitive.mk <$> Attoparsec.Text.takeText
Time (real): 11m31.937s
amazonka-ec2/gen/Network/AWS/EC2/Types.hs (simpl): {terms: 1,384,779,
types: 465,656, coercions: 33,510}

I'm finding the outputted core rather unwieldy to analyse, but the main
difference seems to be the inclusion of huge swathes of case statements
like:
https://gist.github.com/brendanhay/5a57f073d5a7d196a1ad

It would be appreciated if someone could help me to understand in simple
terms why
the use of case-insensitive causes this.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.haskell.org/pipermail/haskell-cafe/attachments/20150106/fac33427/attachment-0001.html>