Plack and apache child init phase

[ysth]

A new server is being developed, mostly by someone other than me, that has some code in a mod_perl child init handler; I'm trying to figure out if it's worthwhile to argue for using a module that expects to be run under Plack instead of reinventing that module's wheel. But I don't see any provision for child init code in Plack or PSGI. Am I missing something?

AFAICT we could just go ahead and use Plack and the child init handler, but that seems contrary to the spirit of PSGI (and would mean special code to enable plackup to work).

I know Starman has a similar child_init_hook; maybe how/if it solves this problem could provide an answer?

[sundialsvc4]

The Plack architecture is simply different in this respect.   You can either do something in the object constructor, or do something the first time through the request-handler method.   In the latter case, the child’s first request will experience a delay.

[Anonymous Monk]

that has some code in a mod_perl child init handler

Sets up some singletons right?

[ysth]

I'm not really clear on the purpose; I think it's not how I would have approached the problem. But I can certainly think of situations where I would use a child init handler.

Why do you ask?

[Anonymous Monk]

Why do you ask?

So I can frame the question better, and because I can't really think of any other purpose for the hook :)

I think you'll find https://metacpan.org/pod/Plack::Handler::Apache2#STARTUP-FILE interesting, as well as https://metacpan.org/pod/Plack::Component#OBJECT-LIFECYCLE and especially https://metacpan.org/pod/Plack::Builder#INLINE-MIDDLEWARE

So you can create your own init_hook as an anonymous inline Plack::Middleware

builder {
  ...
  enable \&child_init_phase;
  $app;
};
...
sub child_init_phase {
    my( $app ) = @_;
    ...   ## your init stuff here
    $app; ## return app unmodified
}

Thats your PlackChildInitHandler