Handle undefined modules when patching and raise UndefinedFunctionError

[xpgdk]

I have noticed that trying to patch non-existing modules, results in an error, that may be quite difficult to understand. The following code

patch(NoSuchModule, :function_that_does_not_exist, fn _x -> 42 end)

will result in the following error

     ** (MatchError) no match of right hand side value: {:error, :nofile}
     code: patch(NoSuchModule, :function_that_does_not_exist, :patched)
     stacktrace:
       (patch 0.13.1) lib/patch.ex:492: Patch.patch/3
       (patch 0.13.1) lib/patch.ex:514: Patch.patch/3
       test/user/patch/unknown_module_test.exs:8: (test)

The first couple of times I saw this, I had absolutely no clue about what was going on. Only by looking into the patch source code, it became apparent, that I was doing something wrong.

This PR adds a simple check that handles the {:error, :nofile} case, and emits an error message that should be easier to understand:

     ** (UndefinedFunctionError) function NoSuchModule.function_that_does_not_exist/0 is undefined (module NoSuchModule is not available)
     code: patch(NoSuchModule, :function_that_does_not_exist, :patched)
     stacktrace:
       (patch 0.13.1) lib/patch.ex:497: Patch.patch/3
       (patch 0.13.1) lib/patch.ex:514: Patch.patch/3
       test/user/patch/unknown_module_test.exs:8: (test)

[ihumanable]

I'm ok with raising an exception here, I'm not sure about UndefinedFunctionError though.

Here's the reasons I'm less inclined to use UndefinedFunctionError:

1. It's not very descriptive to the situation. The changeset here handles the case that Patch.Mock.module/1 returns an error, the thing that's undefined is actually the module not the function.
2. Well, you may retort, if the module is missing then so is the function. I agree with this but the implementation will happily no-op on a missing function.

This leaves us in a situation where the only time patch raises an UndefinedFunctionError is when the module is undefined and not when the function is undefined which also seems like weird behavior.

I think it would make sense to do either of the following:

1. Introduce a custom error here which more clearly indicates the failure. UndefinedModuleError. Use that for this case and retain the current behavior of no-op on a missing function.
2. Change the behavior on missing function to raise and then this can remain UndefinedFunctionError.

Are you interested in either of those alternatives?

[xpgdk]

I completely agree that UndefinedFunctionError is not a good exception to throw. I only chose it, because it aligns with what Elixir itself does:

Erlang/OTP 27 [erts-15.0] [source] [64-bit] [smp:20:20] [ds:20:20:10] [async-threads:1] [jit:ns]

Interactive Elixir (1.17.0) - press Ctrl+C to exit (type h() ENTER for help)
iex(1)> NonExistingModule.call()
** (UndefinedFunctionError) function NonExistingModule.call/0 is undefined (module NonExistingModule is not available)
    NonExistingModule.call()
    iex:1: (file)
iex(1)> 

I think adding a custom UndefinedModuleError is the most clear thing to do.