Enum prefix not dropped for enum names with consecutive capital letters

[ryanrhee]

Given the following protobuf contents:

enum TSEqualsType {
  TS_EQUALS_TYPE_UNSPECIFIED = 0;
  TS_EQUALS_TYPE_EQ = 1;
  TS_EQUALS_TYPE_DOUBLE_EQ = 2;
  TS_EQUALS_TYPE_TRIPLE_EQ = 3;
}

I would expect protobuf.ts to drop the TS_EQUALS_TYPE prefix from the values. But this does not happen. It does correctly drop the prefix, if I change the enum prefix to be wrong:

enum TSEqualsType {
  T_S_EQUALS_TYPE_UNSPECIFIED = 0;
  T_S_EQUALS_TYPE_EQ = 1;
  T_S_EQUALS_TYPE_DOUBLE_EQ = 2;
  T_S_EQUALS_TYPE_TRIPLE_EQ = 3;
}

This seems like a bug in the capital snake case -> pascal snake case conversion.

ETA:
Fixing this would indeed mean that the relationship between capital snake case and pascal case is no longer one-to-one. I.e. converting TS_EQUALS_TYPE_EQ to pascal case is now ambiguous, as the options are:

• TsEqualsTypeEq
• TSEqualsTypeEq
• TsEQUALSTypeEq
• TsEqualsTYPEEq
• etc.

But I don't think that maintaining a backwards mapping from capital snake back to pascal is necessary. At best it may be useful for some tests, but I don't think it should break for what I suspect is a fairly common case, where acronyms are used at the beginning of a enum type name.

[timostamm]

Thanks for the report, Ryan.

To find a shared prefix, we convert the enum name to SCREAMING_SNAKE_CASE first, then test whether all value names have that prefix. But the conversion from PascalCase to SCREAMING_SNAKE_CASE is very simplistic and inserts an underscore before each uppercase letter.

I don't think we need a backwards mapping. We do need the original name of each enum value for the JSON format, but we already persist the shared prefix in the generated code.

I am a bit hesitant about changing this behavior right now, because protobuf-es actually shares that shortcoming. The plan is to migrate to the base types of protobuf-es, so it makes sense to punt on a fix for a while, so we only need to fix it once. I'm labelling this with post-protobuf-es.