bradcray
commented
7 months ago Hi Damian —
To make sure I'm understanding properly, I think you are saying that reading i .. #L by S, your intuition is that it will contain L elements? And since it doesn't, you're proposing that the compiler / linter rule would look for expressions following the x..#y by z pattern and suggest the user might want to write it as x.. by z # y?
The
by Sclause is not a qualifier but a left associative operator with an argument.
You may already know this, and it's nearly moot for this discussion, but maybe not quite: note that #L is also an operator+argument in Chapel, like by S, such that 1..#L is not a range expression but the # operator applied to the unbounded range 1.. [see footnote]. You can also apply # to complete range, domain, or array, like (1..n)#L, myRange#L, or myArray#L. (lo..hi)#-4 or myRange#-4 can be a particularly useful expression, saying "give me the last 4 elements of the range (so, equivalent to hi-3..hi for the first case). Due to these negative operand cases, I tend not to think of # L as being "give me a thing of length L", but rather as a "take" or "restrict" operator, like "take the first four elements" or "take the last four elements".
Anyway, assuming I'm understanding correctly, my hesitation for adding an on-by-default linter rule or compiler warning for x..#y by z is that there are cases where it's a very useful expression—for example, to measure out a bounding box from a given point and then apply the stride to partition the resulting space. Maybe just because the superbowl has happened, I'm imagining that the ball is on the 'x' yard line, we want to measure the y=10 yards to the next first down and we want our video postprocessing to draw lines on the screen every z=5 yards, in which case x..#10 by 5 would be a very useful expression. Of course, as you say, in other cases, x.. by y #z can be very useful as well. But I guess neither seems inherently better/worse than the other to me without knowing the programmer's intention. That's what would make me disinclined to warn about such expressions by default (footnote2).
Do you buy that, or are you seeing some sort of pitfall I'm not?
footnote = in contrast, 1..<L is a range expression, but that's a different kettle of fish
footnote2 = though if this were a common mistake that a programmer or community of programmers made, I wouldn't be opposed to providing them with an opt-in option to be warned about it.
damianmoz
A range can be defined by its bounds
or by a length and a starting index.
Very occasionally, mostly for reasons of working with data whose structure is dictated by some other processing need such as an API, one works with a stride
S. A Chapel range of the first form is then written asor, with the example above
Every once in a while, one wants to have a range with a lowerBound, a Length and a Stride. Simply appending the by operator to the previous range declaration (as in):
introduces a gross error. The length of the second range is no longer
Leven though naively reading the syntax may appear this way. The bySclause is not a qualifier but a left associative operator with an argument.As a strided range (with a length) is likely to be used far less often than an unstrided range, the error above is not going to be immediately obvious to the programmer especially as it is correct syntactically. Such an error will take some time to find. This is especially true if one's working week also involves reading or writing, code in a language other than Chapel which describes a strode range by a qualifier rather than a left associative operator.
The correctly strode range should in fact be written as
One can have little scripts to scan one's code and look for such user errors. But it would be more rigorous if the compiler or a static analyzer or linter could look for such mistakes, especially as Chapel already issues warnings for other esoteric things.
It is arguably a low priority task as a quick and dirty script can be cobbled together to identify such things. But to those who has inflicted such an error on themselves and taken hours to identity their mistake, it might have a much higher priority, especially in the period where they are castigating themselves for their blunder! Just a thought.