[ruby-core:92986] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[ruby-core:92986] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been reported by Eregon (Benoit Daloze).

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: 
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:92987] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been updated by Eregon (Benoit Daloze).

Assignee set to Eregon (Benoit Daloze)

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78365

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: Eregon (Benoit Daloze)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:92989] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[jean.boussier]

Issue #15903 has been updated by byroot (Jean Boussier).


This is kinda tengential, so sorry if it's shifting the discussion. But if `resolve_feature_path` is to be made a first class public API, I wonder if it could be the occasion to make `Kernel#require` invoke `Kernel#resolve_feature_path` under the hood.

The reasoning is similar to [#11140] which made `Kernel#autoload` invoke `Kernel#require` and allowed tools like bootscale / bootsnap.

If the feature resolution logic was swappable it could make these tools much simpler, and open the door to avoiding $LOAD_PATH entirely. 


----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78367

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: Eregon (Benoit Daloze)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:92991] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[mame]

Issue #15903 has been updated by mame (Yusuke Endoh).


(I'm an author of `RubyVM.resolve_feature_path`.)

Sorry but I'm not so positive.  From perspective of module design, I agree that `Kernel` module looks the best place to add the method.  However, we can't be too careful to add anything to `Kernel` nowadays.  At least, I don't want to do that until we receive an actual request to make the method available in production.  Currently, I have no reason to move it to `Kernel`, except module design consistency.

This is just my opinion.  It is all right if matz accepted this.

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78369

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: Eregon (Benoit Daloze)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93000] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been updated by Eregon (Benoit Daloze).

Assignee changed from Eregon (Benoit Daloze) to matz (Yukihiro Matsumoto)

mame (Yusuke Endoh) wrote:
> However, we can't be too careful to add anything to `Kernel` nowadays.

I propose only as a class method, not an instance method, so I think there is literally no chance for conflicts. What's your concern?

> At least, I don't want to do that until we receive an actual request to make the method available in production.

We very rarely receive this, e.g., even for RubyVM::InstructionSequence which is now used in production (bootsnap).
I think it is not a good criteria, it's just too easy to use `RubyVM` in user code.

I understand we should have an actual use-case, but we already have since the feature was introduced.
It would be useful when wanting to have more control over loading files (e.g., I guess this could be useful in RubyGems), and potentially bootsnap as @byroot just said above.

> Currently, I have no reason to move it to `Kernel`, except module design consistency.

I think that's a good enough reason on its own.
`RubyVM` shouldn't become a random collections of classes & methods of which part of it are MRI-specific and part not, part stable and part not.
That's just so messy, so I'd like to fix that.
This issue is a trivial fix for I think an obvious case that does not belong under `RubyVM`.

> This is just my opinion.  It is all right if matz accepted this.

OK, I'll assign to him and add to the developer meeting's agenda.

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78377

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93001] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been updated by Eregon (Benoit Daloze).


byroot (Jean Boussier) wrote:
> This is kinda tengential, so sorry if it's shifting the discussion. But if `resolve_feature_path` is to be made a first class public API, I wonder if it could be the occasion to make `Kernel#require` invoke `Kernel#resolve_feature_path` under the hood.

Could you file a separate feature request and show or explain how bootsnap would use it?

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78378

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93002] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been updated by Eregon (Benoit Daloze).


deep-cover might be interested by this too, cc @marcandre

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78379

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93007] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[mame]

Issue #15903 has been updated by mame (Yusuke Endoh).


Eregon (Benoit Daloze) wrote:
> mame (Yusuke Endoh) wrote:
> > However, we can't be too careful to add anything to `Kernel` nowadays.
> 
> I propose only as a class method, not an instance method

Oh sorry I missed the point.  Fair enough.  I'll ask matz's opinion at the next meeting.


> > At least, I don't want to do that until we receive an actual request to make the method available in production.
> 
> We very rarely receive this, e.g., even for RubyVM::InstructionSequence which is now used in production (bootsnap).
> I think it is not a good criteria, it's just too easy to use `RubyVM` in user code.

Let's try to remove it and see how many people are killed.  (Joke)

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78383

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93097] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[nobu]

Issue #15903 has been updated by nobu (Nobuyoshi Nakada).


How about `$LOAD_PATH.resolve_feature_path`?

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78519

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93101] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been updated by Eregon (Benoit Daloze).


nobu (Nobuyoshi Nakada) wrote:
> How about `$LOAD_PATH.resolve_feature_path`?

As a singleton method, and because `$LOAD_PATH` cannot be re-assigned?
It's a fun idea, although I suspect this will be very hard to find and document properly, so I think it's better on `Kernel` as a class method.

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78524

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93123] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[mame]

Issue #15903 has been updated by mame (Yusuke Endoh).


This ticket was discussed at yesterday dev meeting.  Currently there is no singleton method to `Kernel`, so some people were reluctant.  Nobu counterproposed `$LOAD_PATH` as above, and matz said he waits for eregon's response to the counterproposal.

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78552

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93148] [Ruby trunk Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been updated by Eregon (Benoit Daloze).


Thanks for discussing the issue at the meeting.

I think having singleton-only methods on `Kernel` would be OK, and probably most of us agree having the instance method is not warranted for a rarely-used method.
`Kernel` makes sense to me, because it's where `require` and `load` are defined, and `resolve_feature_path` is a subset of those methods.

My main concern with `$LOAD_PATH` is documentation (e.g., where would it be listed on https://docs.ruby-lang.org/, there is no class, `globals.rdoc` doesn't seem right) and discoverability (hard to find it when searching for it), but otherwise it sounds fine.
Another concern with defining it on `$LOAD_PATH` is `resolve_feature_path` does not depend only on `$LOAD_PATH` but also on other values such as `Dir.pwd`, and maybe other things if the feature lookup changes behavior in the future.

Dear @matz, could you decide between `Kernel` and `$LOAD_PATH`?
I would be happy with either, and I believe that would be much better than `RubyVM` (as explained in the description).

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-78576

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93660] [Ruby master Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[matz]

Issue #15903 has been updated by matz (Yukihiro Matsumoto).


I vote for `$LOAD_PATH.resolve_feature_path`. We need to improve the documentation as well.

Matz.


----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-79271

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93740] [Ruby master Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been updated by Eregon (Benoit Daloze).


Thanks for the decision, and thanks to @nobu for already moving the method.
I noticed the documentation is still on RubyVM, I'll try to fix that.

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-79377

* Author: Eregon (Benoit Daloze)
* Status: Closed
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/

[ruby-core:93741] [Ruby master Feature#15903] Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path

[eregontp]

Issue #15903 has been updated by Eregon (Benoit Daloze).


I documented the new method in [globals.rdoc](https://github.com/ruby/ruby/blob/master/doc/globals.rdoc) and added a NEWS entry.

----------------------------------------
Feature #15903: Move RubyVM.resolve_feature_path to Kernel.resolve_feature_path
https://bugs.ruby-lang.org/issues/15903#change-79378

* Author: Eregon (Benoit Daloze)
* Status: Closed
* Priority: Normal
* Assignee: matz (Yukihiro Matsumoto)
* Target version: 2.7
----------------------------------------
RubyVM contains mostly MRI-specific features but `resolve_feature_path` is clearly not MRI-specific.

So I propose to move it as a class method of `Kernel`.
I think this makes sense given the related `load` and `require` are defined in `Kernel` too.

Moreover, moving this method outside `RubyVM` is *necessary* for other Ruby implementations to implement it, and keep the clean separation that `RubyVM` is only defined on MRI (see #15752).

So, can I move `RubyVM.resolve_feature_path` to `Kernel.resolve_feature_path`?

Do we need to keep the method on RubyVM (and deprecate it), or can we just remove it since anyway API under RubyVM is not stable?

cc @mame



-- 
https://bugs.ruby-lang.org/