in reply to a question on sharing data structures across threads

does this mean, that in my example the $data value form %el has to be explicitly shared as well, or will it become shared automatically, since it's a scalar value and part of %el?

Whenever you use threads::shared::share() on a hash or array, any previous contents of that hash or array are silently discarded. It's a PITA but "working as designed". That is, it is operating the way the originators intended share() to work.

To achieve your aim, it is necessary to explicitely share every compound data structure (array or hash) prior to populating it.

One way of achieving this is to use an existing nested structure traversal utility (eg. Data::Rmap), and copy the data at each level into an appropriate shared structure before assigning a reference to the shared copy to the shared parent.

It is slow and gets messy to do this yourself. I've had a couple of goes at doing this in the past, but it is really something that should be done once, internally, rather than having to be recreated by each programmer.


Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
"Science is about questioning the status quo. Questioning authority".
In the absence of evidence, opinion is indistinguishable from prejudice.
"Too many [] have been sedated by an oppressive environment of political correctness and risk aversion."

Replies are listed 'Best First'.
Re^2: a question on sharing data structures across threads
by TOD (Friar) on Oct 08, 2007 at 06:39 UTC
    ok, thank you for that information. another question: if i declare my $data : shared within an if { } block, does the shared value go out of scope after the block? will i have to explicitly call something like $planets{$some_id}->{'data'} = share($data); to keep the value?
    --------------------------------
    masses are the opiate for religion.
      if i declare my $data : shared within an if { } block, does the shared value go out of scope after the block?

      Yes, it will go out of scope--but it will not cease to exists if

      1. anything holds a reference to it.
      2. if it is a reference and you have assigned it to something who's scope is wider.

      That's not a good explanation. The problem is it depends upon what $data is (contains)?

      • If $data is a simple scalar value, then there is no need to share() it in order to assign it to a shared data structure.

        This is because assigning a simple scalar just copies its value into the (shared) destination.

      • However, if $data is a reference to a hash or array (or nested structure of either or both), then simply sharing that reference is not enough.

        This is because (as indicated above) when you share a reference to a compound datastructure (hash or array), it will (silently) empty the referenced structure!

        For example:

        use threads; use trheads::shared; ## Non-shared hash assigned some data: my %d = (1 .. 10 ); print Dumper \%d; $VAR1 = { '1' => 2, '3' => 4, '7' => 8, '9' => 10, '5' => 6 }; ## Share a reference to that hash and assign it to a shared scalar my $r:shared = share( %d ); ## And not only will the reference point to an empty hash print Dumper $r; $VAR1 = {}; ## But also the contents of the original hash will have been silently +discarded print Dumper \%d; $VAR1 = {};

      So, if as suggested by your OP code, $data contains a reference (as returned by Storable::retrieve(), then you will need to copy the contents of the referenced thing into a shared equivalent.

      To illustrate (Please read the comments carefully!):

      our %planets : shared; our $response = SomeClass::TCP_IP_Response->new(@some_arguments); [...] lock %planets; if (exists $planets{$some_id}) { lock $planets{$some_id}; $response->content($planets{$some_id}->{'data'}); $planets{$some_id}->{'atime'} = time; } else { my $data; [...] # read the serialized data from disk ## Assuming $data is a reference to a *simple*, *single-level* hash ## Copy the data into a shared equivalent; my %sharedHash :shared = %{ $data }; ## And then assign a reference to the *shared* copy. # my %el = ( 'data' => \%sharedHash, # 'atime' => time, # 'deleted' => 0, # 'modified' => 0 # ); ## But if you do this, *ALL THE CONTENTS OF %e1 WILL BE DISCARD* # $planets{$some_id} = share(%el); ## So make the local datastructure shared also my %el :shared = ( 'data' => \%sharedHash, 'atime' => time, 'deleted' => 0, 'modified' => 0 ); ## Now you do not need to use shared() ## as a reference to a shared object is shared. $planets{$some_id} = \%el; }

      And once you've placed a reference to a local (lexical) variable (shared or not) into a variable who's scope is wider, the contents of that variable (but not the name) will persist (not be GC'd) as long the reference persists.

      But also note that if $data is a reference to a hash or array that contains nested references to other hashes or arrays, then you will also need to copy this nested structures manually. That is where a recursive structure traversal tool is needed.


      Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
      "Science is about questioning the status quo. Questioning authority".
      In the absence of evidence, opinion is indistinguishable from prejudice.
        that's what i wanted to know. because since the data get passed in "frozen" state via tcp/ip, $data is a scalar value.
        thank you very much indeed for the competent and elaborate explanation!
        --------------------------------
        masses are the opiate for religion.
        hmm... i did what you advised me to do, but with a remarkable result:
        my %el : shared = ( 'data' => $scalar_data, 'atime' => time, 'deleted' => 0, 'modified' => 0 ); $planets{$some_id} = \%el; lock $planets{$some_id}; # results in an error: "lock can only be used on shared values."
        --------------------------------
        masses are the opiate for religion.