t/Plack-Util/load.t test #11 fails in environments that load FindBin early.

[hartzell]

Test #11 in t/Plack/Util/load.t fails in one of my environments.

That environment uses a sitecustomize.pl file that happens to load FindBin. Beyond loading FindBin, in this particular situation the sitecustomize.pl file is a no-op.

The problem seems to be that _load_sandbox, which is called by load_psgi tries to fake out "FindBin etc." by locally setting $0.

If FindBin's init sub has already been called (e.g. by a call to use elsewhere) then this fakery is insufficient. (See the known issues section of the FindBin man page).

The test can be made to pass in my environment changing t/Plack-Util/bin/findbin.psgi to include a call to FindBin::again() after the use FindBin statement.

This behavior of load_psgi, resetting $0 and @ARGV, does not seem to be documented in the Plack::Util::load_psgi documentation. If it is supported behavior then perhaps it should be described in the documentation and a warning included about this failure mode.

[miyagawa]

it's, for now, documented at https://github.com/plack/psgi-specs/wiki/PSGI-file-evaluation-convention

[hartzell]

If you're going to recommend a practice (for instance here: https://github.com/plack/psgi-specs/wiki/PSGI-file-evaluation-convention), then I think that the safer idiom is to call FindBin::again():

# app.psgi
use FindBin;
FindBin::again();
use lib "$FindBin::Bin/lib";
use MyApp; 
MyApp->psgi_app;

This follows the advice given in the KNOWN-ISSUES section of the FindBin docs.

In particular, it would be great if the Plack test case followed this practice so that automated tools (e.g. Carton, Pinto) can be installed without intervention.

[hartzell]

Spring cleaning, closing.