Hans van Kranenburg wrote on Sat, Nov 05, 2016 at 17:51:13 +0100: > On 11/05/2016 05:15 PM, Daniel Shahaf wrote: > > Hans van Kranenburg wrote on Sat, Nov 05, 2016 at 16:39:44 +0100: > >> On 11/05/2016 11:44 AM, Stefan Fuhrmann wrote: > >> I also guess that ideally the tests need to be updated with cases which > >> show that authz file changes while doing requests are not seen when > >> caching is enabled and doing multiple requests uwing HTTP/1.1 keepalive > >> on a single connection, while they *are* seen when AuthzSVNDisableCache > >> is enabled. From what I see so far, this is not a trivial undertaking. > > > > The catch is, how to write code that commits twice to a single > > repository over a single connection. I don't think the command-line > > client lets you do that. Given that we want to test both svnserve and > > mod_dav_svn, using httplib directly would be suboptimal. I think that > > leaves (a) write a C test, (b) write a C helper tool for doing multiple > > commits over a single connection and write a Python test using that. > > (Example: atomic-ra-revprop-change.) > > Let httpd proxy to itself using keepalive on the backend and do multiple > actions? Gheh, but, very tricky, since you don't control the behaviour.
It'd probably be easier to perform a commit 'by hand' using httplib from the Python tests, but that would test just the mod_dav_svn caching, not the svnserve caching. I agree with the remainder (snipped). Cheers, Daniel
