From: Linus Torvalds > Sent: 28 July 2024 22:01 > > On Sun, 28 Jul 2024 at 13:23, David Laight <David.Laight@xxxxxxxxxx> wrote: > > > > At least you can 'just do it' :-) > > I decided to use my powers for good. Or at least goodish. > > I went through a lot of 'min_t()' and 'max_t()' users, and I think I > found them all. There's a possibility that some driver that I can't > easily build-test has issues, but I tried to manually check all the > architecture ones, and did an allmodconfig build on arm64 and x86-64. > > And by visual inspection I found a 32-bit x86 PAE case. Which makes me > think my visual inspection was not entirely broken. I've been through a lot of them in the past. About 95% could now just be min/max. The 'fun' ones are where the cast reduces the size of one type. One of those got fixed because it was a real bug. But there are some strange (u8) ones in some terminal window size code which manage to convert 0 to -1 then ~0u and finally to the max size - which might even be desirable! > Anyway, I don't love the timing, since I'm going to cut 6.11-rc1 asap, > but I also don't want to unnecessarily leave this pending for later, > so I just committed the simplifications for min_t/max_t. Better than just before -rc6! At least these changes tend to generate build errors. David > Doing the same for min/max (no more C constant expression worries!) > would be very good, but I think that's going to be for later. > > Linus - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
