On Sat, Aug 15, 2026 at 08:29:27PM +0900, Simon Richter wrote: > Hi, Hi Simon,
> On 8/15/26 19:06, Adrian Bunk wrote: > > > The usual i386-specific problem is the excess precision of the x87, > > which goes away with a higher baseline. > > Exactly, but that, too, breaks compatibility, and the sole reason for the > i386 architecture these days is compatibility. > > Especially with games, I would expect that quite a few of them use the > excess precision to avoid z-fighting, that's precisely the sort of hack > you'd do one week before release. you can compile software relying on the old behaviour with -mfpmath=387. AFAIK the number of i386-only games in Debian is zero. > It would make sense to raise the baseline if someone actually wanted to run > scientific software in 32-bit mode, but I believe there is consensus that > this is nonsensical. The x87 excess precision has been a constant source of test failures on i386 (both FTBFS and autopkgtest). This has been the main pain point of the i386 port so far, and it doesn't make sense to continue to have this burden of work when we are no longer supporting CPUs without SSE2. > Simon cu Adrian
