...

(1) At least, the definition should say that the behaviour is undefined if a program ever writes to a file that it has read with `readFile' or `hGetContents' before.

The Library Report is already stronger than this.

Yes, but the behaviour specified by the report isn't exactly useful. The problem is that a semi-closed handle becomes fully closed at some random point in the future, when the list returned from readFile is fully evaluated. In order to make sure that a subsequent writeFile won't fail due to a locking violation, you have to fully force the string returned from readFile. Incedentally, nhc98 doesn't seem to implement the locking (perhaps this is a known bug, I didn't check). Cheers, Simon