while this might work, not sure this is the proper place (or actually the proper fix).

Gem.paths should have been properly adjusted, otherwise we're kind of in a mess anyway... 🤷

see lib/warbler/templates/jar.erb

You're absolutely right.

We’ve looked into this quite extensively, and honestly, we couldn’t figure out why this issue occurs in some projects but not in others — even when the dependencies and environment are nearly identical.

It would definitely be better to understand and fix the root cause instead of working around it. Unfortunately, we haven’t been able to pinpoint it. Given that monkey patches are already present in the relevant issue comments — and much of this file is already composed of other monkey patches (example here) — we’ve put together this PR to help unblock affected users.

That said, a proper fix to ensure Gem.paths is set up correctly would definitely be preferable. We'd be happy to adjust or close this PR if someone can point us in the right direction.

Seems there is an alternate approach at #572 FWIW

not sure about that one either given the if defined?(Gem) && Gem.respond_to?(:paths=) part, which would have the intent of not always applying... my first question would be why 🤷

jruby-rack has ways user can control the ENV and wout looking into Warbler in details I can not tell which one is aligned with that.

also Bunlder supports a bunch of ENV variables and these days RubyGems and Bundler are tightly integrated, so that should be taken into account as well, in terms of isolation from the actual environment.

both PRs achieve something but they feel like quick hacks, would rather not be involved with these with my already limited time for OSS.

Fair enough.

It probably is quite sensitive to how the given servlet container is propagating env from surrounding environment (standalone tomcat et al, vs embedded etc)

Which then makes it a design decision about how warbler should (or should not) respect them and what is most idiomatic. I had to address a related issue with jruby-rack when reinstating the tests for modern bundler/rubygems (with and without mise/rvm) and adjusting paths to load the vendored rack so I might have a clue or another....

After a lot of digging, I've made an alternate approach at #575 which

• includes cleaning up the old hacks which confuse things here
• makes the change closer to the appropriate place (it's not really specific to bundler usage)
• includes analysis of the root cause of change in behaviour in JRuby 9.4

The core problem and attempt to fix it here is valid IMHO - just not in quite the right place or in an ideal state.

As to why it occurs in some places, but not others It is probably sensitive to

• whether the GEM_HOME and GEM_PATH are already set or propagated to the ENV before webserver loads
• whether you have config.webxml.jruby.rack.env.gem_path set to false or some path value for jruby-rack to work with before warbler boots
• jruby version, 9.3 will usually be fine.
• jruby-rack version (I have only really evaluated with 1.2.x)