Discourse Keybase Proof

Ok, so I should just redirect back to the new_proof_url endpoint, fair enough :slight_smile:

yeah. redirecting is perfect. it doesn’t have to redirect to new it could also go to the user’s profile page or settings page or something like that as long as the flash notice tells the user something actionable. whatever makes the most sense. i’m not really sure. no opinions.

the logged-in user is the discourse user that is making the proof

absolutely. yes. it looked like here we’re only checking if it’s the correct user on create and i’d argue we should also do it on new. there’s no sense in trying to save a proof that we know will fail.

The code I pushed is not the latest. I rewrote the create method, removing the check for the username param as it should not be necessary anymore (as the proof_valid? check will fail).

The new method is just a placeholder for Discourse routing logic, it will never be called and it needs to allow non-logged-in requests to allow the user to log in if they logged off previously :slight_smile:

EDIT: this is the new code I just pushed (still not very polished, I pushed it just to let you see it).

2 Likes

About setting up a test domain: it would be great :slight_smile:

3 Likes

I am having issues updating the proof custom field. The first time I put a JSON object in it, it works, but any attempt to change the value of the custom field fails. This is the code:

    def create
      kb_username = params.fetch('kb_username')
      sig_hash = params.fetch('sig_hash')

      proof = Proof.new(current_user.username, kb_username, sig_hash)
      unless proof.valid?
        raise Discourse::InvalidParameters, I18n.t('keybase_proofs.invalid_proof')
      end

      proofs = current_user.custom_fields['keybase_proofs'] || {}
      # Override the signature because the old one might have been
      # revoked and not yet updated on our side.
      proofs[kb_username] = proof.signature

      current_user.custom_fields['keybase_proofs'] = proofs
      current_user.save
      
      render json: success_json
    end

the first time this code is called, it adds the proofs to the custom field named "keybase_proofs". If I call it again, it doesn’t update the custom field.

Some more information: if I change the type of the custom field to a string, it works correctly.

1 Like

Is that code up to date in GitHub?

I assume that

Is not returning a hash, so on the first time it works because you coalesce it to an empty Hash. But it fails on subsequent tries because it’s returning something else.

Adding a debugging break point there may help.

2 Likes

Unfortunately not. Since I am learning how everything works together, my code is almost always in a very sad shape, not really worth pushing. But the code that is in GitHub shows exactly the same issue (but I didn’t push the tests).

In the latest version of the code, which I have locally, I got rid of the JSON type and I am just using serialized JSON strings instead, and it is now working fine.

This is not the case. When I say that “it fails”, I mean “it fails to update” the field. So the first time the code runs, it saves the custom field correctly (I can also retrieve it, I use the data in the profile page and in the check proof endpoint). The second time I run the code (ie, if I add another Keybase identity to the same Discourse user), the custom field is not updated, it keeps only the first signature I added.

In case you’re curious and you want to debug the issue, I’ll create a branch with the original code and tests (that were failing) so you can investigate :slight_smile:

Unfortunately I don’t know how to do it for this codebase using VS Code :frowning: do you have any advice on an alternative editor? RubyMine?

Thanks!

You can add a byebug to any line in the controller, and when that line is hit the console where you started the webserver (the one you run bin/unicorn -x) will stop and allow you to inspect the variables.

9 Likes

The debugger confirms what I saw with previous testing: the JSON is stored the first time, and any attempt to change it later gets silently ignored. Anyway, for now I have fixed it by manually parsing/encoding the JSON string in the custom field.

Hello everyone! In particular @sam and @kb_xgess.

The plugin is in good enough shape now to get reviewed: GitHub - etamponi/discourse-keybase-proofs-plugin: Discourse Plugin for Keybase Proofs.

It should mostly work. There are a few things to fix before it is “shippable”:

  • The “new proof” UI needs some CSS love. It should also show an error if the logged in user is not the one for which the proof is requested.
  • The config endpoint /keybase-proofs/config needs a couple of SVG logos, but I don’t know where to get them from. Ideas? @kb_xgess, do they need to be SVG? Do you need both the black&white and the colored one?
  • redirect-after-login doesn’t work currently. @techAPJ is investigating, might be a bug in Discourse core or (more likely?) a bug somewhere in the plugin.

Some notes on code quality:

  • I didn’t write frontend tests. I think I’ve got the gist of how to write tests for components, but I didn’t want to spend more time figuring out how to mock requests and so on.
  • proof_controller.rb has a very ugly pic_url method, needed because I couldn’t make the call to the keybase API from the client because of CORS. I guess it can be moved to somewhere else or someone can figure out how to fetch the information from the client side.
  • I didn’t run any linter/formatter on the code. Should I?
  • The README might need some love :slight_smile:
7 Likes

I’d love to test it out, is there a test instance available?

2 Likes

What’s the endpoint URL? Maybe it can be requested from the devs?

As far as I know this needs to be allowed per instance by the Keybase folks.

@kb_xgess, can you add meta.discourse.org for us?

2 Likes

takin’ a look at some of your code now. :slight_smile: great progress! quick question about meta.discourse.org, is it running a different codebase? can we deploy your code to it? i ask because part of the back-and-forth protocol is keybase hitting discourse to check for the hosted proof. if it’s not there, it’ll all just fail. so, it might be easier for you to iterate on a public server that’s less production-y. if that makes sense. maybe a staging or hosted dev environment. or even if you have an account with ngrok, we can throw up a static url that always proxies to your localhost that’s only on when you want it (entirely from your end). if you have the flexibility you need on meta.discourse.org, i can turn it on for that site with the configs in your code and some other defaults.

other question about the SVGs. SVGs are ideal because we actually need a handful of PNGs of different sizes for the mobile app, desktop app, and website, and it’s likely easier for everyone if we can render them all from the same initial inputs.

5 Likes

That’s disappointing. Is that necessarily so, @kb_xgess? I was hoping that one could just, say, provide KeyBase a link to a profile on any Discourse server running the plugin (or any Discourse server if it gets pulled in to core) and you could then have proof on any Discourse server.

That was very clear on their announcement post when they launched the service.

If you can add domains easily, please do add one for keybase-test.demo.discourse.org which I will be spinning up soon for our tests.

4 Likes

Oh. I guess I haven’t been following as closely as I thought. :crying_cat_face:

Wait. That’s not what I get from the OP:

Ah, but Keybase ♥'s Mastodon, and how to get your site on Keybase says that they won’t link to

That makes sense, I’m afraid. :wink:

6 Likes

Yes, the plan is creating all the necessary tooling, and I’m sure @kb_xgess will enable it for communities where it makes sense. That doesn’t mean that every ghost town, or login required instance will get it.

4 Likes

Is this old-school way supported by the plugin? I am unclear.

1 Like

No, the new way is, but that requires whitelisting by Keybase

That’s the way that you prove to KeyBase that you own a web site. I thought a similar method might work. I was wrong.