Re: Ada.Strings.Fixed.Count raises Storage_Error

[rieachus@comcast.net]

On Sunday, June 26, 2016 at 8:21:29 PM UTC-4, Jeffrey R. Carter wrote:
 
> While I mostly agree, this one, given the code presented earlier, is a 1-line
> change to correct, which hardly seems like it would make developers' lives
> miserable:
> 
>     while Ind <= Source'Last loop
>        if Pattern = Source (Ind .. Ind) then
>           ...
>        else
>           Ind := Ind + 1;
>        end if;
>     end loop;
> 
> Add
> 
> exit when Ind = Integer'Last;
> 
> after "else". I see no reason this code shouldn't correctly handle this corner case.

It is not a one-line change.  There is documentation, adding cases to regression test libraries, and finally the effort on the developer's part to fix this without producing significantly slower code.  How many times will that extra line be executed if the string is large, but does not end at Integer'Last?  This is extremely important since it is a branch in an innermost loop.  If I were responsible for fixing this I would hoist the case of strings with S'Last = Integer'Last up as high as possible, even if it required (almost) duplicating a lot of code.

This tradeoff between performance and corner cases has been made before.  Try writing a loop that runs from X to Y for some non-static values, then test the code when X = Integer'First and Y = Integer'Last.  You may want to print a period for each 100 million times around the loop...

Do the same thing for a large unsigned array with static bounds of U'First and and U'Last.  This case should work as expected.  Why?  Because looping over the entire range of an unsigned type is something that occurs often in radar, hash tables, and other cases. On a 64-bit machine it probably will not exit if you have the largest possible unsigned type, though I don't know how you could tell. (Even if you could do each interation in a nanosecond, that would take over 500 years...)