Ben Collins-Sussman wrote:
>cmpilato and I are witnessing the same bug on win32.
>
>I'm seeing it in the 1.1.0-rc1 binaries, and cmpilato is seeing it with
>a build of today's /trunk.
>
>1. Destroy the %APPDATA%\Subversion area, so there are no auth caches.
>
>2. run 'svn ls https://svn.red-bean.com/repos/sussman'
>
>The standard unknown client-cert prompt appears: do you want to
>(R)eject the cert, accept (t)emporarily, or accept (p)ermanently?
>
>But there's no opportunity to respond: this prompt is instantly
>followed by a failed PROPFIND error, exactly as if I had typed (R)eject.
>
>This bug isn't reproducible with a 1.0.1 win32 client.
>
>Breser, might this bug be related to your issue #1916 prompting
>changes? They were committed to trunk *just* before we made the 1.1.x
>branch.
>
>
I just looked at the apr_poll code, and unless I'm totally blind, it
will consistently return APR_EBADF on Windows for an APR_POLL_FILE type
poll. However, the wait_for_input functin in poll.c will blithely ignore
that error, returning APR_SUCCESS instead, because it thinks it got a
valid response in the pollset (i.e., it reads from the pollset what it
wrote into it before the call...). I think this is wrong everywhere, not
just on Windows.
I think there's no other choice at this time but to have wait_for_input
return APR_ENOTIMPL on Windows.
-- Brane
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Tue Jul 27 23:10:18 2004