patch applied (packages/bytestring): Make sure that we hClose
file handles when we are done with them
duncan.coutts at worc.ox.ac.uk
Sun Oct 5 16:09:37 EDT 2008
On Sun, 2008-10-05 at 20:36 +0100, Ian Lynagh wrote:
> Hi Don,
> On Sun, Oct 05, 2008 at 11:01:03AM -0700, Donald Bruce Stewart wrote:
> > igloo:
> > > Sun Oct 5 05:11:50 PDT 2008 Ian Lynagh <igloo at earth.li>
> > > * Make sure that we hClose file handles when we are done with them
> > This stuff worries me, ad hoc fixing of bugs already closed upstream.
> You said you'd resolve the situation 3 weeks ago, but now we're out of
> time, so I'm trying to do the best I can.
We've been saying all along though that the right thing to do is to
switch the repo to the correct one. The other solutions were just ways
of trying to avoid that.
The problem was we didn't all agree on the same approach so nothing got
done. The solution we favoured involved you doing something and the
solution you favoured involved us doing something :-)
> The bytestring code is more or less unchanged from the code that came
> with 6.8.1, and I don't remember it causing major problems in 6.8,
It did. It had the handle closing bug, but more generally we had the
problem that people could not use the ghc api package and a fixed
bytestring at the same time. We fixed the last performance bugs in the
> so I don't expect it will do so in 6.10 either. Making large changes
> to bytestring at this stage sounds much scarier to me.
The current bytestring also has a more comprehensive testsuite.
> The patch above should resolve the only serious bug I know of. If you
> have a simple patch to fix the performance problem then we can probably
> get that in too.
I'm really not happy with shipping anything other than the latest
bytestring. We have tested that with 6.10 with the full testsuite. That
branch of the code has been used by lots of people and we've had no
serious correctness bug reports in about a year.
> > Duncan's already sent the win32 dep removal patch already. Now we just need
> > GHCi to not use bytestring to do its little bit of IO.
> Like I said on IRC, I'm not comfortable changing package APIs without
> using the process for doing so, especially now that people have started
> testing the 6.10 branch.
That's fair enough.
Then the solution would seem to be that we apply a mega-patch to ghc's
fork of the bytestring repo to bring it up to the same content as the
current bytestring release.
I'll send a patch for that.
More information about the Cvs-libraries