v5.38 and module_true: File "/app/x.pl" did not return an application object

[melo]

• Mojolicious version: 9.33
• Perl version: 5.38.0
• Operating system: Debian 12.1 (inside a container)

Steps to reproduce the behavior

This is not a Mojolicious bug but a side effect of using v5.38 feature bundle. I'm creating the issue just to leave a note on this, for others that might see this error message.

Morbo fails to load app with error File "/app/x.pl" did not return an application object.

The minimal script:

use v5.38;
use Mojolicious::Lite -signatures; 
app->start;

Run with morbo:

$ morbo x.pl 
Web application available at http://127.0.0.1:3000
File "/app/x.pl" did not return an application object.

This works:

use v5.38;
no feature 'module_true';
use Mojolicious::Lite -signatures; 
app->start;

The feature bundle v5.38 activates the module_true feature. This adds an implicit return 1; at the end of the script file. To make it work we just need to unimport the feature.

Bye,

[leonerd]

Does it help if you wrote return app->start instead?

[kraih]

That sucks. We probably need an FAQ entry for this and recommend against the use of the v5.38 bundle with Mojolicious.

[melo]

Does it help if you wrote return app->start instead?

No, same error.

[melo]

That sucks. We probably need an FAQ entry for this and recommend against the use of the v5.38 bundle with Mojolicious.

I'll search for the FAQ and send a PR, glad to help.

[leonerd]

That sucks. We probably need an FAQ entry for this and recommend against the use of the v5.38 bundle with Mojolicious.

You can just

feature->unimport( 'module_true' ) if $^V ge v5.38;

in your main sub import

[jberger]

That sucks. We probably need an FAQ entry for this and recommend against the use of the v5.38 bundle with Mojolicious.

You can just

feature->unimport( 'module_true' ) if $^V ge v5.38;

in your main sub import

That's a little unfortunate too as then class files will still have to end in 1; though the users might not expect that (some of them at least). Is there any way we can detect that we're in a ::Lite context (and even as I ask it I'm almost certain the answer is no).

[melo]

Using @leonerd suggestion and adding the following

feature->unimport( 'module_true' ) if $^V ge v5.38;

before the goto Mojo::Base::import logic solves the issue.

Not sure if that is the way to go. I like it, and it is just scoped to the use of Mojolicious::Lite.

[jberger]

That will work until Mojo::Base's import would load 5.38, but that's a while off and even then we could probably inspect the hints hash to see if it has already been unimported from outside or something. I like it.

[melo]

@jberger I like it too, seems the most local and clean way to deal with it.

I experimented with (caller(1))[10] to see if I could find the feature_module_true in there, and if present, unimport it. Doesn't seem to work. When you use a feature bundle like use v5.38, the hints hash is undef.

edited: cute...

This sets the hint hash and I can see feature_module_true there:

use feature ':5.38';

but this doesn't set the hints hash:

use v5.38;

I still think just removing the feature is the way to go.

edited, again: I actually believe that the hint hash not been set when implicit bundle load is used to be a perl bug, so I also opened this: https://github.com/Perl/perl5/issues/21387

[melo]

Updated version of the fix, more precise:

feature->unimport( 'module_true' ) if feature::feature_enabled('module_true', 1);

edited: the feature::feature_enabled() function was released on 5.36, so cannot be used with Mojolicious, target is 5.16. Back to @jberger solution.