"Changing tests breaks backwards compatibility relative to the official language. That's a huge deal."
In theory, that's 100% correct. In practice, it's pretty common for ...
... "the influence of others [to] actually change what we see as the correct choice in the moment and remember as the right thing after the fact."
"Fortunately, this effect has good points ...", but still, read on...
notice that a test was written in such a way that it didn't test what it meant to, and to fix that discrepancy
Yes. But what if the person who spotted the discrepancy is merely human?
For everything else in Raku, forgiveness > permission, but for roast we must be ultra careful about being OK with individuals making mistakes when correcting "mistakes". And while we can't ever afford to put the blame on an individual human, no matter what, we must collectively take professional responsibility for never making a consequential mistake with roast, where "consequential" means that Raku ever, even once, fixes one supposed mistake by making another such that this becomes one of Raku's brand associations.
For (re)binding, I think the question is very different when we're talking about parameters
Hmm. Yeah. But there's another problem there that I think needs to be addressed. I'd rather we fixed that too. And then we can leave the current behaviour unchanged.
But SO comments probably aren't the place to figure that out.
Fair enough. But since we were here I decided to write down my immediate reaction.