Re: A question about certain language features
- Posted by Robert Craig <rds at RapidEuphoria.com> Feb 12, 2002
- 617 views
Ed Davis writes: > 1) Variables can not be initialized when declared, but > rather, must be initialized via an assignment statement. > Based on other languages, this seems like one of those > convenience type issues. Is there some programming > philosophy that says that an initialization in a declaration > is a 'bad thing'? It's not a "bad thing". In fact it seems like an obvious bit of convenience that couldn't possibly cause any harm. Since a lot of people have asked lately, here is my (very subjective) opinion on the matter. I've also included your question about private constant declarations. Suppose we allow: #1. variable inits: integer x = 99 and if you allow #1, it would be even stranger (than it is now) not to allow: #2. constants declared inside a routine for use within that routine only So now you've got routines that start off with an arbitrary mixture of variable and constant declarations, some of the variables are assigned an initial value, some aren't. The only visual difference between a constant declaration and a variable declaration, is that people have to remind themselves that "constant" is not a type, it's something totally different. e.g. function aaa(integer z) object d, e, f constant a = 99 * foo() integer b = 99 + bar() * 7, h, i, g = 777 atom c g += 1 h = g etc. Now whenever you examine a routine, instead of quickly skipping over the declarations, you will have to carefully wade through all the declarations, some containing function calls and complicated expressions, to figure out which code is going to be executed. Simply stated, you've got executable code hidden amongst the declarations, and constant declarations now look no different from variable declarations. Things used to be neatly separated and you could visually skip the declarations when mentally tracing execution. Now things are jumbled together. I once spent hours debugging something in another language, simply because the author of the code had hidden a variable declaration with initialization in amongst 20 or so other declarations. I just couldn't "see" that the subroutine did not really "start" where I thought it did. It actually started in the middle of the declarations somewhere. I like it better the way it is. You could argue that I don't have to use variable inits if I don't want to. You could argue that I don't have to use goto if I don't want to. A language does not exist just to serve the isolated programmer. It exists to serve a community of programmers. In situations where it really doesn't matter how something is written, I think there are advantages to reducing the number of choices. > 2) No support of call by reference. I understand that call > by reference can lead to unexpected side-effects, but since > changing global variables in a subroutine seems to > essentially cause the same problem, I don't understand this > omission. 1. Call by value is appropriate 90% of the time, and no one has any complaints. 2. Call by value is fundamentally easier to understand and less likely to cause surprises. 3. In Euphoria, a variable can only be modified by explicitly assigning a new value to it. No aliases. No pointers. No hocus pocus. Certainly I would never want to have to look up the source code for each routine to find out if it was going to secretly modify the argument I was passing. 4. If a sequence parameter is not modified in the subroutine, then only a pointer is passed internally, so it's very cheap. 5. If the sequence *is* modified, then copy-on-write comes into effect, and a copy will be made of the argument sequence. In some cases this copying could be considered expensive, but only when a small portion of the sequence is modified. If you are going to rewrite the whole sequence, as in a sort routine, then the overhead from copying ( a few machine cycles per element) will be far less than the cost of sorting (or adding/subtracting/multiplying etc.) each element. There's an optimization which is near the top of my list, that would eliminate the need for copy-on-write in many cases. Consider: function sort(sequence s) s[i] = ... end function sequence x ... x = {5,6,7,8} ... x = sort(x) It would be possible to detect that the value of x, as passed to sort(), will never be needed again. The implementation could then arrange that the value passed to sort() will have only *one* reference count (from s), rather than 2 (from x and s). This would eliminate the need for a copy. Regards, Rob Craig Rapid Deployment Software http://www.RapidEuphoria.com