Re: Mild bug?

Ben Collins-Sussman <sussman@collab.net> writes:

> > Did you mean detect that diff has exited with SIGPIPE? APR doesn't
> > appear to allow us to determine which signal was received.
>
> !!
>
> > Even if we could detect it svn still has to exit.
>
> Why? So the child died with SIGPIPE, big deal. Why not allow svn to
> ignore the death, continue on, and just eventually exit with no error?

Well, in this case all subsequent diff processes are also going to get
SIGPIPE.

>
> > I suppose we could omit the error message but I'm not sure it's an
> > improvement.
>
> That would definitely be a user-interface improvement. The user
> simply terminated the 'more' process... why should an error come up in
> the user's face? I pipe things into 'more' all the time, and hit 'q'
> to exit. I've never seen the original data-producing program yell at
> me about that.

It's a question of reliability. I don't know enough to guarantee that
the *only* time diff gets SIGPIPE is when the user deliberately closes
the output stream. Are there other ways it can happen? Can it happen
if the machine runs out of some resource? Can it happen if there is a
bug in diff? Can we guarantee that

$ svn diff > my.patch

will never silently produce an incomplete patch?

-- 
Philip
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org