MFELDMAN@GWUVM.BITNET (Mike Feldman) (02/22/88)
In response to Susan Flynn regarding my delay test: I think you are correct that I've got an erroneous program here. I suppose it would've been better to time-stamp all the messages from the 2 tasks and write them to 2 separate files. The intention was to find an easy and obvious way to show that in the runtime systems I tested, the high-priority task was not waking up on time (or at all). I agree that writing the shared variable (the screen or whatever) should've incorporated synchronization, but of course that very synchronization would've probably defeated the example. Your discussion of I/O issues (again) was very instructive. The end purpose of the example (and the rest of the discussion) has served the purposes for which I, at least, intended it: 1. to highlight the fact that many people experienced in Ada including, apparently, some compiler vendors, are disagreeing with each other in good faith (or just plain baffled) about what is meant by the LRM in this critically important area; 2. to highlight the apparent weaknesses in the ACVC in testing this area which is apparently so "make-or-break" to some real-time people. I've I used the word "illegal" instead of "non-conforming"; perhaps I'm being too loose with my terminology but the idea is the same: in this area particularly, we all have especially weak signals from the ACVC and the validation process about just what we should expect from the tasking part of an Ada runtime system. I have heard it said (glibly to be sure) that "all Ada tasking programs are erroneous" - but I believe the confusion here makes it imperative that attention be sharply focused on this area so that we have a clear and unambiguous statement regarding what the runtime system is expected to do, so that we can write programs that we are certain are not erroneous. One rule of thumb I agree with is that Ada tasks should _always_ do some explicit synchronization now and then. A brief delay might do, if nothing else - but if _all_ tasks don't do it, there's _still_ no guarantee, priorities notwithstanding, that a sleeping task will ever wake up. I will continue to make this assertion until the ACVC tests the intent of the LRM and all implementations are known to conform! Call it illegal, non-conforming, or whatever buzzword you like - there is still confusion on the issue, and I think my tests prove it.