On Mon, Feb 21, 2011 at 11:57:50AM +0000, Philip Martin wrote: > Stefan Sperling <s...@elego.de> writes: > > > On Mon, Feb 21, 2011 at 11:21:10AM +0000, Philip Martin wrote: > >> Stefan Sperling <s...@elego.de> writes: > >> > >> > But I think there is value in seeing the error message in the test code. > >> > It makes it much easier for those reading the test code to follow > >> > what kind of error the test is expecting. If all people can go by > >> > is the error code, then they'll have to keep looking up error codes > >> > to understand what the test is trying to do. > >> > I think that's worse than adjusting tests when the message changes > >> > every once in a while. > >> > >> For those reading the test code a comment could be used, although I > >> admit that a comment is less reliable. On the other hand, if we just > >> check error numbers that might make it possible to run the regression > >> tests in some locale other than C, and some language other than English. > > > > Sure but is that something we really need? After all this would mean > > testing gettext, which we don't maintain. > > We have got this far without it, but it would be interesting to try it. > The regression tests depend on lots of things we don't maintain BDB, > libxml, etc.
Fair enough, but why not write a specific test for this instead of testing it as a side effect of making the code of other tests harder to read?