Reviewer Comments

• Ward and i talked today about putting together a pattern language that includes both test automation design patterns and more general process patterns describing how testers can fit into ExtremeProgramming. BrianMarick and i have been working on something like this that we sometimes called Clipper: http://www.testing.com/cgi-bin/clipper.pl (now one of the SisterSites). Brian has also hosted several test patterns workshops: http://www.testing.com/t ... terns/index.html. In my browsing today i stumbled accross another testing pattern language by HowardFear that looks very compatable: http://kw4.knightweb.com ... Appendix-A.shtml . -- BretPettichord

• It seems to me worth emphasizing that a goal of the techniques on this site are to help define CoachingTests (aka Guiding Tests or ProductTestsFirst). These are tests written before coding that help define the system. They are SpecificationsByExample (TestsAsCompleteSpecifications). In other words, there are statements of "requirements" (i hate that word). -- BretPettichord

• I've tested your framework on three different projects now. One interesting aspect is that our applications have to provide a gentle facade, so that we can write the TestFixture in the first place. This technique helps inform the design of the application-level interfaces, like its small-scale companion, TestDrivenDevelopment, helps inform the design of the class-level interfaces. -- FrankWestphal

• Consider representing the individual fields of value objects, e.g. GeoCoordinate, as multi-columns. Column heading could be "coord.lat" and "coord.lon". -- FrankWestphal

• If the checked data type is a String (of length > someThreshold) and the expected value differs from the actual, it'd be useful to highlight the segment where both strings differ. Could easily be included by using the new ComparisonFailure facility of JUnit 3.8 -- FrankWestphal

• I want to echo FrankWestphal's comment about acceptance tests being to external interface design as unit tests are to class-level design. In particular, I've found that they help decouple external interfaces in the way that MockObjects decouple internal ones. For example, I wanted to test startup and shutdown behavior of a desktop app. It was inconvenient to have the tests run in the same process as the app being shut down. I found myself naturally using what MartinFowler used to call a SessionFacade - that is, I put the business logic on a "server" on the network that the tests started up, shut down, and talked with. Now it's easy to plug in a command-line interface, probably easier to add a Cocoa interface than if I'd been thinking monolithic process, and web services fall out naturally. I think exploring how acceptance tests influence architecture is very important. I'm groping for how to talk about it. AbstractConcreteTests is sort of along these lines. --BrianMarick

• What about an Ant task that does most of the job of filtering the path(s) for fixture documents, running the framework and providing a summary report, linking to all other summary documents? That would be quite helpful a tool for the folks who don't use a WikiRunner. -- FrankWestphal . . . I'd love to do this but don't (yet) know how. -- WardCunningham

• The VideoStore sample shows why we should both escape metacharscters (newline, lessthan) in strings destine to html and process c-style escapes (\n, \t) in string data from html. -- WardCunningham

Browse OldComments and their resolutions.