On Fri, Aug 09, 2002 at 11:45:20AM +0200, Arthur Bergman wrote:
> 
> Mind blowing work, I am very impressed and feel ready to work on a new 
> version of PoCo::HTTPD,
> 
> A quick comment is that it seems to lack any transactional 
> environment/rules, is this intentional?

Sorry, no.  Thanks for bringing it up.  I've summarized the discussion
re: transactions so far and added it to the document.

  === Transactions ===

  Arthur Bergman noticed that transactions were not mentioned.  Torvald
  Riegel pointed out that transactions would solve many problems:

  It would be really cool to support transactions.  If you have full
  control of messages, you can do rollback, make snapshots, and other
  useful things.  Ideally, it would be transparent to the component.

  These features are generally not needed except in the most fanatically
  robust systems, and they are difficult to implement.  They also tend
  to be performance bottlenecks.

  Protocols with atomic messages can usually work around the problems
  that transactions exist to solve.  They should tide us over until
  transactions are worked out.

  Design goals: Leave hooks for a future transactional system, but don't
  implement one now.  One aspect of this goal is to control even
  internal access to a component.  We cannot take snapshots of a
  component or roll back its state unless we can control access to its
  internal data.  This basically requires ALL access to $self to go
  through accessors.

-- Rocco Caputo / [EMAIL PROTECTED] / poe.perl.org / poe.sf.net

Reply via email to