Re: Performance of records with variant parts

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On 2021-03-22 19:23, John Perry wrote:

> I think I understand that, but the check for "obj.kind" is there in both versions of the code. Isn't that a dynamic check even when I use a non-variant record? What's the difference in the check when I'm checking the discriminant of a variant record and the value of a field?

1. Depending on the optimization level the compiler might keep the 
second check in place when accessing the variant field regardless the 
selected case statement alternative.

2. A varying size object might be allocated on the secondary stack 
rather than on the program stack. That could indict a sufficient penalty

3. Variant alternatives could be allocated outside the object. That 
would require an additional indirection and thus penalties when copying 
objects and accessing fields.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de