Using strict language extension

[kazu-yamamoto]

In my opinion, network programming in Haskell should use strict evaluation except atomicModifiyIORef', to avoid space leak. Strict and StrictData language extension are introduced in GHC 8.0 and it's days of GHC 8.6. So, it's good time to introduce them to network.

For this purpose, all undefined should be removed. It's very easy. But during this work, I experienced that undefined are evaluated even if they are removed from the source code. The reason is here: https://gitlab.haskell.org/ghc/ghc/issues/16810

When such inlinable functions are fixed in GHC 8.8, I would like to add the following to network.cabal:

  if impl(ghc >= 8.8)
      Default-Extensions:  Strict StrictData

[eborden]

We should probably formulate some realistic benchmarks before introducing a change of this nature.

[kazu-yamamoto]

SPJ helped me to understand what's going on. This is not a bug of base. We can use these extensions for all versions of GHC 8 now.

https://github.com/kazu-yamamoto/network/tree/strict

[kazu-yamamoto]

warp, warp-tls, auto-update, wai-app-file-cgi and mighttpd2 have been released with Strict and StrictData enabled by default.

[kazu-yamamoto]

I should explain this. The purpose of this PR is to let us convinced that space leaks are not due to lazy evaluation when they occurs. This PR might not improve heap profiling but would save our souls very much. I'm very tired from finding sources of space leaks due to lazy evaluation for last 10 years.