Re: port-mips/57680: printf("%.1f") shows wrong resultsonR3000mipseb

[Rin Okuyama <rokuyama.rk%gmail.com@localhost>]

The following reply was made to PR port-mips/57680; it has been noted by GNATS.

From: Rin Okuyama <rokuyama.rk%gmail.com@localhost>
To: Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>
Cc: gnats-bugs%netbsd.org@localhost, netbsd-bugs%netbsd.org@localhost
Subject: Re: port-mips/57680: printf("%.1f") shows wrong resultsonR3000mipseb
Date: Sat, 4 Nov 2023 22:38:28 +0900

 On 2023/11/04 19:16, Izumi Tsutsui wrote:
 >>>    Unless some MIPS guru has proper idea how this issue should be fixed,
 >>>    I'd like to commit workaround fix to disable -DHonor_FLT_ROUNDS in
 >>>    src/lib/libc/gdtoa/Makefile.inc in ${MACHINE_CPU}=="mips" case.
 >>
 >> (1) How about -current or netbsd-10, which have GCC 10.5?
 >>
 >>       Less likely, but if this is a temporal problem for specific
 >>       versions of GCC, we can restrict the workaround to netbsd-9.
 > 
 > I have not tried it, but last time I tried netbsd-10 kernel
 > on NWS-3260 it was very unstable and looked lost software interrupts
 > many times, so I'm currently testing NWS-3260 on netbsd-9.
 
 Thanks, I understand the situation :(
 
 > I doubt this problem was really compiler issue and
 > I wonder if it's okay to assume "FE_TONEAREST" is default
 > because FE_TOZERO works.
 > 
 >> (2) Can we exclude ``mipsn{,64}e[bl]''? -current and netbsd-10 have
 >>       ${MACHINE_MIPS64} macro for this purpose, but netbsd-9 does not :(
 > 
 > What's your motivation?
 > 
 > Performance? If so, is it visible or measurable?
 > Consistency? If so, is there any rationale of Honor_FLT_ROUNDS
 > implementation?  Actually we already have an exception (vax).
 > 
 > If you have a working patch to achive your intention in both
 > HEAD and release branches, it's fine.  If not, sorry I don't
 > have motivation to prepare such complexities.
 > 
 > I'm trying to fix visible issue (and real problems like invalid
 > strings stored in /etc/ntp.conf etc.).  If you want possible
 > performace or design consistency, it should be measured or
 > defined by people who want it, because we are on Tier-II ports
 > "keeping it working is the responsibility of the user community."
 >   https://wiki.netbsd.org/ports/
 
 I'm worried about consistency.
 
 IMO, it is less surprising that vax is treated differently here;
 it does not comply IEEE 754, and no working fenv implementation
 at the moment.
 
 This patch is working both for -current/netbsd-10 and netbsd-9:
 
 https://www.netbsd.org/~rin/gdtoa_port-mips_57680.patch
 
 Thanks,
 rin