FilterManager really is dreadful

I reported a serious FilterManager bug some time ago, which was severe enough for me to rewrite the RO4 module for my own sanity. In revisiting it today I remembered another serious bug I fixed and hadn’t mentioned.

This page is a lie – it would be nice if R1 was indeed the caller’s event buffer as it was when it left the Wimp, but unfortunately FilterManager corrupts it for its own purposes because it only has eight other registers it could have used but no.

It seems FilterManager has always had this bug too.

Why have you modified the documentation to match the bug? Should the bug not be fixed?

Why have you modified the documentation

Because every version of FilterManager in existence (from RO3.1) has this behaviour except for the copy on my computer.

There is simply no point in documenting something that is not true, however much we’d like it to be. The documentation exists to allow programmers to write stuff. They clearly can’t do that if the documentation is a lie.

Even once the bug is fixed, the API will have to clearly state that R1 is only the caller’s event buffer from version X.YZ of the FilterManager module, whereas it is available at [R13,#4] in all versions ever shipped.

My recommendation is to fix R1 and define the [R13,#4] too.

I just rewrote the *Filters command from scratch and the module is now 20% smaller. Crikey.