[Issue 8077] Regression (2.058): .reserve does not work for shared arrays

[d-bugmail]

http://d.puremagic.com/issues/show_bug.cgi?id=8077



--- Comment #2 from Andrej Mitrovic  2012-05-16 
11:30:15 PDT ---
Well I thought shared was going to become some sort of gateway to
safe-by-default concurrent programming, I guessed wrong.

This section needs to be updated (by someone who knows exactly what shared
does):
http://dlang.org/attribute.html#shared

Some partial information is here:
http://dlang.org/migrate-to-shared.html#shared

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
--- You are receiving this mail because: ---

[Issue 8077] Regression (2.058): .reserve does not work for shared arrays

[d-bugmail]

http://d.puremagic.com/issues/show_bug.cgi?id=8077


Walter Bright  changed:

   What|Removed |Added

 Status|NEW |RESOLVED
 CC||bugzi...@digitalmars.com
 Resolution||INVALID


--- Comment #1 from Walter Bright  2012-05-16 
11:06:45 PDT ---
I'm going to argue that although the code compiled in 2.057, it was not correct
and had latent concurrency bugs. The problems are:

1. being shared doesn't mean that assignments to an array are atomic

2. there is no obvious place to put a lock around updating a shared resource
like a shared array

3. locks can be very, very slow. Shared does not mean "automatically insert
locks"

Therefore, I'm going to mark this as invalid.

In essence, you (the programmer) has to decide how to manage the memory is to
be controlled by a shared array. You can allocate an array as thread local,
then cast it to shared as a last step (probably the best approach). If the
array is already shared, you'll have to protect the resize with a lock that
fits in with your data structure.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
--- You are receiving this mail because: ---