Chip Salzenberg wrote:
According to Alan Burlison:IMHO it would be better to *remove* it and deliberately break the small number of things compiled against 5.8.1.Who are you, and what have you done with Alan the software engineer?
I became even older, greyer and more cynical ;-)
Allowing a mistake to perpetuate doesn't seem to like the best choice.But Alan, *we* made the mistake. We suffer by maintaining a vestigial variable. It's a small cross to bear, but it's *our* cross. Making users suffer for a developer's mistake -- on purpose -- is just wrong, and users won't forgive us for it easily or quickly.
Normally I'd be in 100% agreement with you, but I'm concerned about people unknowingly continuing to use 5.8.1 if we think it is broken. If you are sure that the proposed fix won't cause any problems then fine, let's do that. I'm not that bothered by the extra variable per se, but if there is any chance of silent breakage I think it is better to break at link-time rather than have mysterious errors, and removing the variable is one sure-fire way of doing that.
Note we have broken binary compatibility on nearly every perl release so far - ISTR that 5.8.0 had something in the README that more or less said 'We broke binary compatibility with 5.6.1, tough'. In this case I personally care more about correct behaviour than I do about binary compatibility.
-- Alan Burlison --