feat(@opentelemetry/semantic-conventions): change enum to object literals

[echoontheway]

Which problem is this PR solving?

• the transpile outputs of enums seem a little weird，which build lock-in to TS and not that straightforward

TS:

outputs:

• transpile outputs are hard to be treeshaked, which increases bundle size of a Web App.
  As shown above, TS enums will be transpiled to self-executing anonymous functions, which won't be regarded as dead code and got eliminated by pack tools. So all these enum transpile outputs are left in the bundle even if they're not imported by the app.

bundle size of enum:

bundle size of object literal

(bundle size statistics are got with the help of rollup-plugin-visualizer . Detail pack config can be found in the simplified example-app in my repo)

Short description of the changes

• modify template and regenerate semantic conventions
• the benefits of using enums for code completions and type checking are retained
• the bundle size is decreased

[codecov]

Codecov Report

Merging #2532 (2a50988) into main (f9f43e5) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main    #2532   +/-   ##
=======================================
  Coverage   93.05%   93.05%           
=======================================
  Files         139      139           
  Lines        5172     5172           
  Branches     1110     1110           
=======================================
  Hits         4813     4813           
  Misses        359      359           

[rauno56]

Would const enums behave much differently?
May I ask how did you get the bundle size statistics? I'd like to understand the situation a little better myself. :)

[weyert]

Sorry, how does this impair debuggability? Wouldn't the benefits of using enums for code completions and type checking getting lost?

Just found: https://www.typescriptlang.org/docs/handbook/enums.html#objects-vs-enums
Maybe it could work if as const gets added as a postfix 🤔

[echoontheway]

Would const enums behave much differently? May I ask how did you get the bundle size statistics? I'd like to understand the situation a little better myself. :)

Const enum members are inlined at use sites, and they will be completely removed during compilation. We don't use the enums in semantic-conventions, so they won't exist in semantic-convention transpile outputs. Const enums seem not suitable for our 'export' scenario.

I pack my app using vite , tree-shaking is enabled by default for vite. And I get the bundle size statistics with the help of rollup-plugin-visualizer . I have uploaded the simplified example-app just now to my repo，you may have a look :)
pack config:

pack command:

TS enums will be transpiled to self-executing anonymous functions, which won't be regarded as dead code and got eliminated by pack tools. So all these enum transpile outputs are left in the bundle even if they're not imported by the app.

[echoontheway]

Sorry, how does this impair debuggability? Wouldn't the benefits of using enums for code completions and type checking getting lost?

Just found: https://www.typescriptlang.org/docs/handbook/enums.html#objects-vs-enums Maybe it could work if as const gets added as a postfix 🤔

It seems object is more straightforward and won't impair code completion.

Do you mean 'const enums'？I tried const enums, however , they seem not suitable for our 'export' scenario, for const enums will be removed during compilation. Then the other app that imported semantic-convention (transpiled code) can't find the 'const enums'.

I have tried to add the 'as const' as postfix for object to keep the original 'type checking' ability and it works, thanks for your advice!

[weyert]

Great @echoontheway , thanks for taking the time to try it out :)

I now also understand this is problem for when using it for client-side browsers

[weyert]

LGTM

[obecny]

this is really nice finding, lgtm

[Flarna]

Line 17 tells // DO NOT EDIT, this is an Auto-generated file from scripts/semconv/templates//templates/SemanticAttributes.ts.j2.

if this is true this change will vanish on next generator run. Maybe the generator needs some fix?

[weyert]

Line 17 tells // DO NOT EDIT, this is an Auto-generated file from scripts/semconv/templates//templates/SemanticAttributes.ts.j2.

if this is true this change will vanish on next generator run. Maybe the generator needs some fix?

The template file has been updated too. Should be all good