Does not respect defaults set in config.exs

[codehugger]

It seems that the Nanoid generate functions are not picking up the values set in config.exs.

I currently have this is my config/config.exs inside a phoenix project

config :nanoid,
  size: 21,
  alphabet: "0123456789abcdef"

When I open the shell iex -S mix I get the following

iex(1)> Application.get_env(:nanoid, :size)
21
iex(2)> Application.get_env(:nanoid, :alphabet)
"0123456789abcdef"
iex(3)> Nanoid.Secure.generate()
"yXR2LbmKTzohOefJ5ocCh"
iex(4)> Nanoid.NonSecure.generate()
"obxlccY4icVuAQO2l5vy_"

This shows that the correct configuration is loaded into the iex but it is not being picked up correctly by the module.

[codehugger]

This seems to be an issue with when the config is read inside a phoenix project. I did some digging and tried recompiling the project by deleting the _build directory. When I ran IEx again Nanoid picked up the updated config.exs.

I updated the config/config.exs file again now with the following settings.

config :nanoid,
  size: 4,
  alphabet: "0123456789abcdef"

But it still uses the old config when I run IEx

iex(1)> Nanoid.generate()
"efbf07642ca92851159ed"

So again I deleted the compiled files, now only the _build/dev/lib/nanoid folder.

Now when I run IEx it has picked up the new config.

iex(1)> Nanoid.generate()
"ad88"

This is a pretty acceptable workaround since this only seems to be an issue in Phoenix projects. I won't be changing the defaults for Nanoid any time soon in the project but still a little bit annoying having to figure this out.

[railsmechanic]

Many thanks for your testing and your workaround. I think the problem is the use of module attributes, which were assigned during compile time. It seems that the configuration in the (dev|test|prod).exs file(s) of phoenix is not available when nanoid gets compiled and therefore it uses the fallback value which is defined using Application.get_env/3.

To avoid this issue I'll introduce a Nanoid.Configuration module which also supports runtime configuration.

[railsmechanic]

@codehugger
I've added the Configuration module to master. Could you please test it with your application? When everything works, I'll release v2.0.3.

Many thanks so far.

[codehugger]

@railsmechanic I have tested this and I can confirm that a simple IEx restart is enough to pick up the new config :)

[florianb]

Does this have any impact on the performance?

I made the former config changes (for the test-env) working by calling

MIX_ENV=test mix deps.compile nanoid --force

[codehugger]

Not that I have noticed but I have not made any extensive testing.

On Tue, 16 Jun 2020 at 11:22, Florian Breisch ***@***.***> wrote: Does this have any impact on the performance? I made the former config changes (for the test-env) working by calling MIX_ENV=test mix deps.compile nanoid --force — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAAEZZQTCSRNJFZZDWBPTTRW5IWBANCNFSM4NLNH7IQ> .

-- ./bjarki

[railsmechanic]

Does this have any impact on the performance?

Name                    ips        average  deviation         median         99th %
nanoid_master        5.40 K      185.23 μs    ±58.53%      157.40 μs      468.39 μs
nanoid               6.29 K      159.06 μs    ±61.38%      135.91 μs      445.44 μs

Unfortunately yes. I've done a benchmark of using module attributes vs. reading the application environment: Using module attributes are always faster. It's questionable whether the effects are acceptable. But I think faster is always better.

But using a Configuration module as a single point of truth IMHO has its charme. Therefore I think using a combination of module attributes within the Configuration module is the (my) way to go.

If someone need runtime configuration it's possible to pass a custom alphabet and size to the generate/2 function.

[florianb]

Ah yeah - that sounds pretty reasonable! Thank you very much for your work!

[railsmechanic]

I released v2.0.3 which now uses the Configuration module.