Faster 'cabal update'

[tanakh]

Currently, 'cabal update' uses Distribution.Simple.Utils.writeFileAtomic.
But it requires data as String, so very very slow.
3.7MB of 00-index.tar.gz is extracted to 61MB of 00-index.tar.
It takes about 8 seconds to write 00-index.tar on my machine (i5-2400 CPU @ 3.10GHz).
cabal-install can depends on bytestring', so I rewritewriteFileAtomic` to use bytstring.
The bottleneck is resolved (< 1 second).

[kosmikus]

Thanks for the pull request. I'll try to look at it soon.

As a general policy, I think it would be best to submit pull requests to master, not release branches such as cabal-1.14 (except if they concern specific bugfixes in a release branch). We can then always decide to merge changes from master back into cabal-1.14.

[tanakh]

Sorry for wrong way. On next time, I do so.

[dcoutts]

Looks ok to me. Only suggestion is that we should really move the new bytestring version of writeFileAtomic into the Cabal lib's Utils module.

[tanakh]

I think so, too. Is there no reason that Cabal avoid to depend to bytestring?

[dcoutts]

Historically Cabal did not use ByteString because it did not work with hugs and nhc98. We don't worry about that anymore. In future we will want to make use of bytestring and text packages.

[tanakh]

I understand. I will look forward to that time.

[tibbe]

Merged to master as 85ad9f8.