AddressSanitizer doesn't report the location of memory leaks in unloaded cores

[JesseTG]

This ticket is relevant to developers who are using LeakSanitizer to debug memory leaks in cores (not in the frontend).

When a libretro core is unloaded, it takes all the symbols with it. In general, this is desired behavior. However, tools that produce stack traces for programs after they close might still have needed any symbols within those cores.

This means, for example, that LeakSanitizer can't identify the origin of core memory leaks unless RetroArch crashes while the core is still loaded. The leaks themselves are still reported, but the stack trace contains (<unknown module>) instead of a library/function name. See these two tickets in the relevant repo for details.

I'd like to see an option to prevent cores from being unloaded, so that memory leaks within them are easier to identify. Someone on Discord suggested that this could be a CLI-only option, so that some hapless newbie doesn't accidentally cause problems for themself.

[JesseTG]

Huh, I think HAVE_PRESERVE_DYLIB fits this use case already. See this commit for details.

[LibretroAdmin]

This seems to be a relevant request.

[JesseTG]

One that might have been resolved eight years ago. Building now, will close if --enable-preserve_dylib fulfills this need.

[JesseTG]

Ah ha! Using --enable-preserve_dylib in the configure script solves this exact problem. Closing.