Interpreter: clear finished hooks after intepreting

[asterite]

Fixes #12172

Some explanation:

• once the program is parsed and analyzed it computes a list of "finished hooks": code that must be analyzed and executed when the program ends
• analyzing the code was actually happening already: that's why macro finished was able to declare vars, etc.
• executing finished hooks is not done yet. That is, if you write puts "Hi" inside a finished hook, you won't see it in interpreted mode (I'm leaving that for a later PR)
• however, the finished hooks were kept in memory and re-analyzed whenever we entered a new line in pry. And it seems analyzing finished hooks in pry doesn't work very well right now
• so, far now, I'm just clearing the list of finished hooks right after interpreting some code. Ideally we'd also execute any finished hooks at the end of the program, but I played a bit with it and it's not so easy to do, so for now I'm not doing that
• but at least this is a small step forward and you can debug apps that have a finished hook

[asterite]

Well, in the end I added the ability to execute finished hooks in interpreted mode.

So you can try this:

# foo.cr
macro finished
  puts "Bye!"
end

puts "Hello!"

If you execute that with crystal i foo.cr, previously the output was:

Hello!

Now it's:

Hello!
Bye!

You can also try this:

# foo.cr
macro finished
  puts "Bye!"
end

puts "Hello!"
debugger
puts "Hello again!"

Then with crystal i foo.cr this happens:

$ crystal i foo.cr
Hello!
From: foo.cr:8:6 <Program>#foo.cr:

    3:   puts "Bye!"
    4: end
    5:
    6: puts "Hello!"
    7: debugger
 => 8: puts "Hello again!"

pry> continue
Hello again!
Bye!

Which makes sense, because finished is executed at the very end of the program.

That said, I don't think executing code is common in finished hooks. It's usually used to define code. But I think it still makes sense to execute it, just like in non-interpreted mode.

[asterite]

Oh, and you can also try this:

$ crystal i
icr:1:0> macro finished; puts 1; end; puts 2
2
1
=> nil
icr:2:0> macro finished; puts 3; end; puts 4
4
3
=> nil

That is, you can actually define finished hooks in interactive mode, and they get executed.

[asterite]

@cyangle If you want/can, you can check if the interpreter works fine in your particular case with this PR

[cyangle]

@asterite Thanks a lot!

Now I can use the interpreter to debug my Lucky toy app:

$ cr i src/toy.cr 
Using compiled compiler at /home/chao/git/personal/crystal/.build/crystal
Listening on http://0.0.0.0:8080
2022-06-30 22:50:29 +00:00
From: expanded macro: setup_call_method:15:3 Api::V1::Users::Show#call:

    10:         begin
    11:   logger.info do
    12:     "user_id: #{id}"
    13:   end
    14:   debugger
 => 15:   json(UserTwin.new(current_user!))
    16: end
    17:       end
    18: 
    19:       __temp_139 = run_after_pipes
    20: 

pry> id
"774107312037625857"
pry> current_user

GET /api/v1/users/774107312037625857 (7fbb6ac9-6096-4da8-b1a1-33ff01e43c7c)
 ▸ user_id: 774107312037625857
#<User:0x7f4b24983840 id: 774107312037625857, name: "Erick Schmidt", email: "[email protected]", lock_version: 0, age: 18, created_at: 2022-06-27 05:53:23.634084000 UTC, updated_at: 2022-06-27 05:53:23.634085000 UTC, deleted_at: nil>
pry> current_user!
#<User:0x7f4b24983840 id: 774107312037625857, name: "Erick Schmidt", email: "[email protected]", lock_version: 0, age: 18, created_at: 2022-06-27 05:53:23.634084000 UTC, updated_at: 2022-06-27 05:53:23.634085000 UTC, deleted_at: nil>
pry> current_user!.id
774107312037625857
pry> current_user!.name
"Erick Schmidt"
pry> UserTwin.new(current_user!)
#<UserTwin:0x7f4b309a1460 @age=18, @id=774107312037625857, @name="Erick Schmidt", @email="[email protected]", @lock_version=0, @created_at=2022-06-27 05:53:23.634084000 UTC, @updated_at=2022-06-27 05:53:23.634085000 UTC, @deleted_at=nil>

I also tried it with the Lucky website, it worked. @jwoertink

This is awesome!

[jwoertink]

whoa!! That's awesome @cyangle! Great work on this @asterite ❤️

[beta-ziliani]

Awesome! @asterite would it be hard to add specs? Do you want me to do that?

Also, for when this is merged, is this a reasonable change in the description? Thx!

---

Fixes #12172

Some explanation:

• once the program is parsed and analyzed it computes a list of "finished hooks": code that must be analyzed and executed when the program ends
• analyzing the code was actually happening already: that's why macro finished was able to declare vars, etc.
• executing finished hooks is not done yet. That is, if you write puts "Hi" inside a finished hook, you won't see it in interpreted mode (I'm leaving that for a later PR)
• however, the finished hooks were kept in memory and re-analyzed whenever we entered a new line in pry. And it seems analyzing finished hooks in pry doesn't work very well right now
• so, far now, I'm just clearing the list of finished hooks right after interpreting some code. Ideally we'd also execute any finished hooks at the end of the program, but I played a bit with it and it's not so easy to do, so for now I'm not doing that
• but at least this is a small step forward and you can debug apps that have a finished hook

[asterite]

Hi @beta-ziliani! There are no specs for pry yet and doing that is a huge task, and for now I won't have time to do that.

[asterite]

Another thing is the interpreter is mostly experimental. If it doesn't work well, it's not a big deal because compiled programs are the ones shipped.

But I promise I'll introduce tests for pry later on, otherwise it will be hard to make it evolve in a safe way. But I'd like to leave it for later.