Max Kellermann

Thanks @geneticdrift, your analysis was correct - commit 57212d5a35a588ec29ed3c2c642598c6541542df was bad.

6db4b818e679d24f46c1222ddf732ae60117cb18

That part only knows which decoder plugin is being used. We could provide that piece of information easily. Would that be good for you?

Then what *do* you want, exactly? "mp3" is neither a container/file format nor a codec. Neither is "mpeg". "mp3" is a commonly used filename suffix for MPEG Layer 3 audio,...

> If a codec goes by different names That's why I asked the codec name to be "canonical", so there is one reliable name for each codec. It does not...

CI failures due to a dumb typo; I force-pushed an amended version.

> The bug that was featured in your talk would not have been prevented by this refactoring. Of course not! This is just a source-code level change with no runtime...

> If we went for it, I would like to assess if it has any impact on the generated code. The generated code is *mostly* identical. In a perfect world,...

With this PR, the executable grows by 572 bytes. But if I apply https://github.com/MariaDB/server/pull/3531 first, then this PR adds only 256 bytes, because the compiler can now omit some of...

> Would the executable size grow less if you did not add any data member to your RAII guard, but instead made it refer directly to the singleton `dict_sys`, similar...