[abandoned] Proposal for multiple "~p" sigils

derekkraan

I have been using a multi-endpoint setup in my app, just because I think it makes the most sense for a multi-subdomain app. This worked well enough with the old style of route helpers, since I could call them with their fully qualified module names.

Now with the introduction of ~p, things have become more complicated. The crux of the issue is: there can only be one ~p sigil defined per module. This means that it is not possible to use verified routes for multiple routers in the same template.

The result is that I have resorted to using just string literals in the places where cross-domain links are present.

My proposal is to make the sigil that is defined customizable, with a default to p when no other has been defined.

Usage would look something like this:

use Phoenix.VerifiedRoutes,
  endpoint: MyApp.Endpoint,
  router: MyApp.Router,
  statics: MyApp.statics,
  sigil: "MYAPP"

And then you would be able to use the sigil like this: ~MYAPP"/my_path".

This would make it possible for several sigils to be defined in a single module.

I’d be happy to provide a PR if there is agreement that this would be a good addition to Phoenix.

[edit: apparently path/3 and url/3 exist and I just didn’t know about them. In light of that it seems unlikely that this proposal will be accepted]

12 comments

/phoenix

0 594 12

2023-07-18 01:27:15 UTC

Showing Posts 1 to 10

Show Best Posts
Show All (oldest first)
Show All (newest first)

hubertlepicki

This is the only reason I haven’t touchec ~p sigil at all and stick to the old routing model.

Post #1

LostKobrakai

You can use verified routers also with path(conn_or_socket_or_endpoint_or_uri, router, ~p"…").

Post #2

NobbZ

This is not valid syntax. As far as I remember, sigils are limited to a single lating letter, restricting their namespace to a size of 52.

If at all, you might be lucky using the “options” argument to the sigil. I am not aware though if the order is guaranteed to be kept, or if there are already options defined by sigil_p/2 that might conflict.

Though I have not yet used verified routes at all, as I didn’t have had any usecase for using route helpers at all in the last 2 years.

I usually do not generate paths on the server side.

So here I have to throw in some assumptions:

I’m pretty sure there is an alternative way to use verified paths rather than just using the sigil. And from what I can find, it seems as if the sigil is bound to a specific endpoint within a module when Phoenix.VerifiedRoutes was used with the :endpoint and :router options, while usable with any endpoint when the phoenix module was imported and then sigil_p/2 is used within a path/3 macro.

You should be able to build your own macros on top of this:

defmacro myapp_p(og_ast) do
  quote do
    path(MyApp.Endpoint, MyApp.Router, unquote(og_ast))
  end
end

Plus/Minus some quoting adjustements and perhaps some useage helper that deals with importing Phoenix.VerifiedRoutes for you.

Post #3

derekkraan

From the Elixir 1.15 changelog:

[Kernel] Support for multi-letter uppercase sigils

Post #4

derekkraan

Not sure how I overlooked this , still getting used to all the new goodies in Phoenix / LV lately (as you have probably noticed from my presence on Slack). Thanks!

Post #5

NobbZ

Ooof, I missed that one, still I am not a friend of that.

sigil_p/2 seems to allows for interpolation, while uppercase sigils do usually not do that by convention.

Personally I would probably prefer my explicit macro (maybe even a self-written uppercase multiletter sigil like yours, if you insist on that) over yet another auto-generated macro which every developer has to guess what it means.

With an explicitely self-implemented macro I get “docs on mouseover” for free in most editors. Sigils don’t receive this benefit.

And if they are generated by phoenix, you can not even use CTRL/CMD + click without falling into a deep rabbit hole.

Post #6

derekkraan

still I am not a friend of that

Discussion on that change was pretty heated so I don’t think you’re the only one. Still, it’s there.

by convention

I think with new capabilities there will be new conventions.

Personally I’m very much in favour of the new sigil capabilities. There are a thousand new sigils that one could imagine, and the all upper caps means that you will readily be able to see where magic is happening.

Post #7

NobbZ

I meant the autogenerated sigils for the various endpoints. I was not complaining about the multiletter sigils in general.

I only argued against them in this context, and more against autogenerating them.

Yes. I agree.

I hope though that we at least keep the convention for single-letter sigils.

Agreed, though then I am wondering, why only uppercase, not lowercase?

Why do I need to call it sigil_GRAPHQL/2 rather than sigil_graphql?

PS: Sorry for derailing.

Post #8

DianaOlympos

The whole point of the all uppercase sigil was to allow interpolation in the way the implementation wants it. We already have an example in wide use today with Heex. ~H"{magic_assign}" in attributes

Post #9