- From: James Graham <jgraham@opera.com>
- Date: Thu, 21 Mar 2013 21:19:14 +0100 (CET)
- To: Tobie Langel <tobie@w3.org>
- cc: Robin Berjon <robin@w3.org>, Dirk Pranke <dpranke@chromium.org>, public-test-infra <public-test-infra@w3.org>
On Thu, 21 Mar 2013, Tobie Langel wrote: > I was planning to look at what the requirements for that were based on > the existing tests. Should I expect this to be much more challenging > than a twist on an echo server? I guess using something like that could work for a subset of cases, but it seems particularly bad for readbility if it encourages people to make a substantial amount of the test URL-encoded. However there are a large number of things that it doesn't work for. > Any pointers as to which tests have specific requirements in that area? >> and also to store state across requests. > > Hadn't thought about that one. What are the use cases? Progress events. Tests that require reloading documents. Tests that need to count the requests made, or check that requests reach the server in the right order. Tests that depend on authentication (e.g. authenticate in one step and check that credentials are required, then check that credentials are sent automatically in following steps). Tests for the HTTP cache or appcache. >> For >> comparison, see the documentation of Mozilla's javascript based solution >> [1] (Opera have had more or less all the reequirements described in that >> document, but have historically implemented them on top of Apache + PHP >> instead of a custom solution). > > Lifting Mozilla's server might be a possible alternative, here. I believe it is heavily dependent on XPCOM, so it isn't at all portable. Using the same idea but in python could work. However note that we also need a solution that we can run on w3c-test.org where there could be multiple simultaneous clients accessing the server. That means that e.g. state can't be global. It also means that it shouldn't fall over under load. Not an easy problem :( >> I don't think the most significant property of the system is that it is >> simple to use (although of course making easy things easy is a good >> thing); I think the most significant property is that hard things are >> possible. > > Well, you can always fallback to writing custom code for this, can't you? Custom code where? It isn't clear from what you said that you were thinking of a server that can run custom code. But yes, I think my point is that the server has to be able to run custom code on a per-test basis to deal with the edge cases.
Received on Thursday, 21 March 2013 20:19:49 UTC