[JAVA] Performance issues with AbstractJavaCodegen.toModelName()

Description

We are using swagger-codegen to generate Java code based on some OpenAPI specification. We noticed that the generation was taking a considerable amount of time so we decided to profile the generation. What we saw is that most of the time was spent inside AbstractJavaCodegen.toModelName(String name). After further investigation, we noticed that this method was being called 3390629 times and 3389750 of the times was for duplicated names. This means that only 879 times this method was being called with a unique name argument.

Just as a POC, we introduced a cache into this method and we got execution times up to 3x faster.

Here you can see a call stacktrace:

Let us know if there is more information we can provide.

Swagger-codegen version

4.0.0-beta3

Swagger declaration file content or url

Command line used for generation

Steps to reproduce

Related issues/PRs

I haven't found any

Suggest a fix/enhancement

We were wondering if the introduction of a cache may be something you may want to consider. But we are not sure about the implications of such a cache.

Start at AbstractJavaCodegen.toModelName(String name) and profile a Java generation run to confirm the repeated-name calls described in the issue. Investigate the implications of caching this method, then validate that generation remains correct and measure whether the reported performance improvement is reproduced; no specific test or reproduction command is provided.