abitdodgy
The other day I wrote a small lib to interact with AWS IAM (on top of ExAws). I wanted the internal API to have one Parser.parse/2 function that would handle any response. ExAws executes the request and calls this function with the response and the action name as the two arguments. For example:
parse.call(response, action)
Where response is {:ok, %{body: xml, status_code: status}} and action is the AWS IAM operation (“CreateUser”, “DeleteUser”, etc…).
I created a single module with a parse/2 action that dispatched to a function, which in turn called the right parser in the right module (pattern matching on the action name).
defmodule ExAws.Iam.Parser do
alias ExAws.Iam.Parsers.{AccessKey, User}
def parse({:ok, %{body: xml, status_code: status} = resp}, action) when status in 200..299 do
parsed_body = dispatch(xml, action)
{:ok, %{resp | body: parsed_body}}
end
def parse(resp, _), do: resp
@user_actions ~w[ListUsers CreateUser]
defp dispatch(xml, action) when action in @user_actions do
User.parse(xml, action)
end
@access_key_actions ~w[ListAccessKeys GetAccessKeyLastUsed]
defp dispatch(xml, action) when action in @access_key_actions do
AccessKey.parse(xml, action)
end
end
I then created individual parsing modules for each entity:
defmodule ExAws.Iam.Parsers.User do
def parse(xml, "ListUsers") do
# parse the xml
end
def parse(xml, "GetUser") do
# parse the xml
end
# etc...
end
defmodule ExAws.Iam.Parsers.AccessKey do
# ...
end
I’m not a fan, though. The Parsers.AccessKey and Parsers.User should be private (which we don’t have in Elixir). I can’t import the functions because the names would clash. I also feel like I’m abusing pattern matching.
Protocols won’t work because the data types don’t change. What other patterns can I use to achieve the same API?
Trending in Questions
Other Trending Topics
Categories:
Sub Categories:
Forums
Popular Tags
- #ecto
- #liveview
- #troubleshooting
- #learning-elixir
- #deployment
- #library
- #erlang
- #testing
- #genserver
- #mix
- #absinthe
- #remote-other
- #otp
- #plug
- #how-to-question
- #macros
- #postgres
- #channels
- #elixirconf
- #exunit
- #discussion
- #code-sync
- #javascript
- #podcasts
- #onsite
- #dialyzer
- #docker
- #authentication
- #umbrella
- #full-time-contract
- #podcasts-by-brainlid
- #ecto-query
- #elixir-ls
- #blog-post
- #phoenix_html
- #iex
- #graphql
- #ai
- #genstage
- #elixirconf-us
- #websockets
- #supervisor
- #advent-of-code
- #distillery
- #processes
- #api
- #forms
- #metaprogramming
- #security
- #hex










First 10 of 11 Posts
abitdodgy
Thinking out loud, I suppose behaviours could be a solution? If I’m willing to accept that passing the right parser as an argument is better than having a polymorphic
parse/2function.I could declare
parse/2as a behaviour and implement it in each module (Parsers.User, andParsers.AccessKey). I would then pass the right parser as an argument.Still, it would be nice to just call
parse/2and have the API figure out which module to call.peerreynders
Have you considered:
abitdodgy
I don’t control the calling code. ExAws does that part. It just calls the parser I provide as an argument.
Here’s how I currently send the parser:
So it’s easy enough to replace
|> to_op(parser: &Parser.parse/2)with|> to_op(parser: &Parsers.User.parse/2). But I was looking for a better way.yurko
I do this kind of things with adapter pattern where I set the adapter in config, have a behavior that is an interface for adapters and few adapter modules (some call API, some use ETS some just serve hardcoced responses). A “public” wrapper module can then call configured adapter and depending on the env different implementations can be used, the context (that is the only really public module) can delegate to that wrapper module then.
dimitarvp
The best way in Elixir I could see is a mix of your original approach, and behaviours.
dispatchto use router functions like so:Code is not tested. But IMO this brings you close enough to some practical polymorphism while remaining terse enough.
(Another alternative would be to have a big module where you match on all possible combos but I don’t think it would be an improvement over your current approach – except maybe for having less source files.)
peerreynders
But as it is you’ll probably be best off by pursuing the behaviour angle further if just to capture the implementation commonalities between your various parsers.
abitdodgy
That’s essentially what I had before I went into a refactoring rabbit hole. It might just be simpler to pass the parser in the public function. For example:
abitdodgy
That’s pretty cool, but it sacrifices explicitness especially as the codebase gets bigger (my current approach also sacrifices explicitness).
dimitarvp
Valid reservation. I just feel we can’t go all the way explicit if we need polymorphism though.
As an extra step I would recommend you just make a separate configuration file and just use that in order not to have too much routing logic hardcoded in a module (where other team members might have trouble finding it or it will become hard to maintain as you pointed out).
If you opt for that you should definitely put that string_action->module map that @peerreynders suggested into that config.
I understand the nagging feeling of “it just doesn’t feel right” but don’t get stuck into paralysis analysis. Truth is, functional languages are not ideal when polymorphism is involved and that’s okay because it’s not their main focus. But behaviours and map routers get you close enough, and it’s not like the OOP languages don’t do the same anyway!
IMO just go for the minimal maintenance approach and move on. Even if the code is a bit more. Important part is to be able to add/remove consumers in a minute.
peerreynders
In some contexts “passing a function” can be seen as the functional equivalent of OO’s strategy pattern.