BTW, this is why we cannot ship with the old bytestring fork of 0.9
bumping around in GHC.
----- Forwarded message from Bryan O'Sullivan <[EMAIL PROTECTED]> -----
Date: Tue, 9 Sep 2008 14:11:45 -0700
From: "Bryan O'Sullivan" <[EMAIL PROTECTED]>
To: "Duncan Coutts" <[EMAIL PROTECTED]>,
"Don Stewart" <[EMAIL PROTECTED]>
Cc: "Ketil Malde" <[EMAIL PROTECTED]>
Subject: Nasty (but fixed) complexity-changing bug thingy in bytestring
Hi, gents -
Ketil and I were mystified by a huge performance discrepancy between
builds of the same code, which he found to be due to the versions of
the bytestring library we were using.
He had a fast build using bytestring 0.9.1, and I had a slow build
using 0.9.0.1. It took us a while to isolate this as the source of the
difference, but it makes a huge difference. Our application can
consume 61MB of data in 17 seconds with the good (new) bytestring, and
shows linear scaling, but behaves quadratically with 0.9.0.1, and
didn't complete for me after almost two hours on the 61MB input.
I assume (but have not confimed) that the culprit was a fusion bug. Do
you have tests that ensure that fusion rules are firing correctly, or
some other way to be sure that for example this fix might not get
accidentally undone?
----- End forwarded message -----
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
