- From: Alex Danilo <alex@abbra.com>
- Date: Thu, 18 Mar 2010 01:34:52 +1100
- To: "Erik Dahlstrom" <ed@opera.com>
- Cc: "Patrick Dengler" <patd@microsoft.com>, "www-svg@w3.org" <www-svg@w3.org>
Hi Erik, --Original Message--: ><snip/> >* Many of these tests are so simple that they should be joined into a >single test IMHO. I disagree. One of my pet hates with the current test suite is that most of the tests test a combination of many features and require many things to be implemented to pass a single test. So as unit tests they are useless. The descriptions usually describe what the test is testing but in order to pass a large number of other features need to be implemented. The result is that some tests only test one feature - e.g. coords-trans-02-t.svg whilst others such as udom-svg-237-t.svg test many individual features in one test. Comparing those 2 tests feature-wise, it's about 30x the number of features implemented to pass the latter versus the former. The result is that passing one test has no significant correlation with passing a different test so the number of passes/failures is completely lacking correlation with the correctness of an implementation. i.e. two implementations may pass 90% of the test suite each, whilst one is a highly conformant implementtion, yet the other fails in many areas. This is a problem with the current test suite - very few tests actually test the feature they are testing in isolation - and that has been a criticism from other people as well... Breaking down the test suite into simpler tests, one per feature would allow a more meaningful indicator of implemented feature coverage rather than the conglomerated tests we now have. I know it's nice to have the combined succinct test suite we now have, but it is basically useless to a new implementor starting out - there should be some incremental single feature tests to allow an implementor to validate each feature as it is implemented. Because of that, I don't agree that any of the tests are too simplistic. Breaking tests down to individual features benefits all implementors and I think the existing test suite suffers from an imbalance in the weight of implementation required to pass each of them. Best regards, Alex
Received on Wednesday, 17 March 2010 14:36:33 UTC