DConf 2013 Day 1 Talk 2: Copy and Move Semantics in D by Ali Cehreli

[Dicebot]

On Friday, 10 May 2013 at 12:08:10 UTC, Andrei Alexandrescu wrote:
> ...

After some more thinking on topic I get the feeling that this is 
yet another case when having defined and working "scope" for 
function parameters would have been extremely useful.

Lets imagine some string processing function I want to use:

     int foo(string);

That is like Phobos string processing functions are defined, they 
take immutable(char)[], not const(char)[]. That is good in a 
sense that guarantees that function thread safety, but becomes an 
issue when you want to use that function with a mutable char 
buffer (some typical network packet manipulation, for example).

Presentation recommends to use assumeUnique in such cases as you 
can be sure that mutable entity only exists in calling scope and 
is practically immutable for the duration of function call. 
However, this is potentially dangerous, because foo may save 
reference of some kind to its immutable argument in global state 
and then your program is in undefined behavior. It is OK with 
Phobos string processing functions because common sense and open 
sources guarantee that no such stuff happens, but in general this 
sounds like a type system hole.

scope may have solved it.