On Mon, 2019-09-30 at 13:13 -0400, Laine Stump wrote:
On 9/30/19 10:05 AM, Andrea Bolognani wrote:
I see your point about backports being more painful when you have a bunch of unrelated changes mixed in, but I would still prefer if we converted everything at once and at the same time introduced a suitable syntax-check rule preventing more instances of whatever function we just removed all callers of from creeping back in, or actually just dropping the function altogether.
Don't forget that make syntax-check doesn't work properly for many downstream maintenance branches that would be backported to (it has to be disabled due to copyright date checks failing, or something like that).
That's a problem for downstream to solve. By the same token, all the existing syntax-check rules are pointless because they can't be guaranteed to hold for downstream branches.
In order to allay Andrea's fears of new usage of VIR_AUTO* that just draws out the conversion, maybe we could (temporarily, until the conversion is complete) put a commit hook in place to disallow new use of VIR_AUTO ? Or just, you know, pay attention in reviews (but of course part of the point of all of this is to eliminate the potential for human error, by depending less on humans paying attention, so... :-P)
Writing a check that compares the situation before a commit and after it is not as easy as a point-in-time check. Instead of spending a non-trival amount of time implementing something like that, I'd rather spend my time dealing with the fallout of a one-time conversion.
(BTW, I'm not firmly in *either* camp, although I may lean a bit more towards a gradual change (but with a *very* steep slope to minimize the period of confusion)
That's just a big-bang conversion with extra steps! -- Andrea Bolognani / Red Hat / Virtualization