On Mon, Oct 25, 2004 at 07:37:34AM +0100, John Levon wrote: > On Sun, Oct 24, 2004 at 12:03:59PM +0200, Andre Poenitz wrote: > > > > > That's why I propose changing some/most/all asserts to something less > > > > brutish, i.e. an exception carrying the same information as the assert > > > > that will be caught in the main loop (i.e. the outermost dispatch or > > > > even in the frontends). > > > > > > This seems an excellent route towards dataloss*. > > > > Not true. You can catch all of these exceptions in the main loop and pop > > up a flashing red sign "something went wrong but I refuse to chrash > > right now". Things like this can be made obvious. > > Sure, but how's that going to encourage users to use 1.4.0cvs?
If the result is still usable, it's usable. Right now we crash e.g. when we can place the cursor correctly after an 'undo'. Of course it will be annoying to have the cursor jump to pos 0 in the current par after an exceptionally tricky 'undo', but people have been living with that since day one for e.g. mathed (cursor always left mathed on undo()). So this is 'more usable' than a plain crash. Of course, fixing the real problem is nicer and should be prefered in the long run, but I'd rather have a usable LyX in finite time than an unusable in infinite time. Andre'
