Build fails on NetBSD/vax

[puddles123fluffy]

Hello all,

Building ksh93 from pkgsrc on NetBSD/vax fails during the final compilation/linking of the ksh93 binary with this message:

# src/cmd/ksh93/Mamfile: 1096-1103: make ksh
+ link_binary ksh pmain.o 
+ cc -D_BLD_DLL -fPIC -Os -O2 -Wl,-R/usr/pkg/lib -o ksh pmain.o libshell.a /usr/pkgsrc/shells/ksh93/work/ksh-1.0.10/arch/netbsd.vax/lib/libast.a -lm /usr/pkgsrc/shells/ksh93/work/ksh-1.0.10/arch/netbsd.vax/lib/libcmd.a /usr/pkgsrc/shells/ksh93/work/ksh-1.0.10/arch/netbsd.vax/lib/libast.a -lm -lm -lutil /usr/pkgsrc/shells/ksh93/work/ksh-1.0.10/arch/netbsd.vax/lib/libdll.a /usr/pkgsrc/shells/ksh93/work/ksh-1.0.10/arch/netbsd.vax/lib/libast.a -lm -lm -lm -lutil /usr/pkgsrc/shells/ksh93/work/ksh-1.0.10/arch/netbsd.vax/lib/libast.a -lm -lm
ld: /usr/pkgsrc/shells/ksh93/work/ksh-1.0.10/arch/netbsd.vax/lib/libast.a(fmtuid.o): in function `fmtuid':
fmtuid.c:(.text+0x5c): warning: warning: reference to compatibility getpwuid(); include <pwd.h> to generate correct reference
ld: libshell.a(init.o): in function `sh_reseed_rand':
init.c:(.text+0x1c4c): undefined reference to `remainder'
mamake [cmd/ksh93]: *** exit code 1 making ksh
mamake: *** exit code 1 making cmd/ksh93
mamake: *** exit code 1 making all
package: make failed at Sun Apr 18 17:15:48 PDT 2026 in /usr/pkgsrc/shells/ksh93/work/ksh-1.0.10/arch/netbsd.vax

It seems the standard math library is missing remainder(). Is this caused by the VAX's non-standard float? Any way around it?

Thanks in advance!

[McDutchie]

I don't know anything about NetBSD on the VAX as I've never had access to a VAX, unfortunately. So I'll need your help in debugging this.

First, let's see whether your version of NetBSD is even supposed to have the remainder function. What does the command man 3 remainder show?

If it does document the remainder function, then I'd like you to try to build the current code from the git repo. There have been loads of fixes since 1.0.10, including a build system fix that could affect this (commit 714aeb5). Please do the following:

git clone -b 1.0 --depth 1 https://github.com/ksh93/ksh
cd ksh
bin/package make

If that succeeds, then do bin/package use to try it out or sudo bin/package install /usr/local to install it, and please let me know so I can close this issue :-)

If it doesn't succeed, please show that build log here so we can debug further.

[puddles123fluffy]

Hello,

Thank you for the prompt response!

remainder() does exist in section 3 of the manual on my VAX. The text is exactly alike to this html copy: remainder(3).

I have started the build process, but it will be a while before I can get back to you. Building ksh93 took about 3-5 hours last time.

Thank you again!

[McDutchie]

I have started the build process, but it will be a while before I can get back to you. Building ksh93 took about 3-5 hours last time.

If your VAX has multiple CPUs or CPU cores (find out with bin/package host cpu), then you can build in parallel, like bin/package make -j3 for 3 parallel build jobs. If so, and you're in the middle of one of those 3-5 hour build jobs, it's worth interrupting it to continue the build in parallel.

[puddles123fluffy]

Hello again,

Sorry for the late response. I was out while it was compilling. It takes a very long time with only one CPU/core.

It seems to be failling in the same way:

fmtuid.c:(.text+0x42): warning: warning: reference to compatibility getpwuid(); include <pwd.h> to generate correct reference
ld: libshell.a(init.o): in function `sh_reseed_rand':
init.c:(.text+0x185c): undefined reference to `remainder'
mamake [cmd/ksh93]: *** exit code 1 making ksh
mamake: *** exit code 1 making cmd/ksh93
mamake: *** exit code 1 making all
package: make failed at Sun Apr 19 09:21:45 PDT 2026 in /root/workdir/ksh/arch/netbsd.vax

Thank you again.

[McDutchie]

Can you show me the whole log entry starting with # src/cmd/ksh93/Mamfile: LINE-LINE: make ksh? We need to make sure the -lm linker flag is included and in the right place.

[puddles123fluffy]

Sorry, I meant to share the whole thing but my "selection aim" was off.

This is it:

# src/cmd/ksh93/Mamfile: 1039-1043: make ksh
+ link_binary ksh pmain.o
ld: /root/workdir/ksh/arch/netbsd.vax/lib/libast.a(fmtuid.o): in function `fmtuid':
fmtuid.c:(.text+0x42): warning: warning: reference to compatibility getpwuid(); include <pwd.h> to generate correct reference
ld: libshell.a(init.o): in function `sh_reseed_rand':
init.c:(.text+0x185c): undefined reference to `remainder'
mamake [cmd/ksh93]: *** exit code 1 making ksh
mamake: *** exit code 1 making cmd/ksh93
mamake: *** exit code 1 making all
package: make failed at Sun Apr 19 09:21:45 PDT 2026 in /root/workdir/ksh/arch/netbsd.vax

Although, I don't think that is the problem. I wrote my own extremely simple remainder-test.c, and compiled it with LDLIBS=-lm make remainder-test. It worked on my Linux PC but failed on the VAX with the same error as ksh93. I then suspected that remainder() is really not defined, so I did nm --defined -D /lib/libm.so | grep remainder and got nothing. It seems to be in the header but not in the library!

Thank you once again.

[McDutchie]

Looks like that bug should be reported to NetBSD, then.

However... I reconsidered the function, and it looks like my use of remainder(3) from commit af6a32d was wrong anyway, because remainder may return negative values for two positive inputs, and casting a negative floating-point value to an unsigned int is undefined behaviour.

What I really wanted was fmod(3), the floating point remainder function. So, let's hope your NetBSD does have that in its libm.

Does this patch make it build?

diff --git a/src/cmd/ksh93/sh/init.c b/src/cmd/ksh93/sh/init.c
index f148675aa..a24448fea 100644
--- a/src/cmd/ksh93/sh/init.c
+++ b/src/cmd/ksh93/sh/init.c
@@ -699,7 +699,7 @@ void sh_reseed_rand(struct rand *rp)
 	unsigned int		time;
 	static unsigned int	seq;
 	timeofday(&tp);
-	time = (unsigned int)remainder(dtime(&tp) * 10000.0, (double)UINT_MAX);
+	time = (unsigned int)fmod(dtime(&tp) * 10000.0, (double)UINT_MAX);
 	rp->rand_seed = (unsigned int)sh.current_pid ^ time ^ ++seq;
 	rp->rand_last = -1;
 }

[robohack]

Many people doing NetBSD/vax development run it under an emulator on a much faster system.

For example: https://www.netbsd.org/ports/vax/emulator-howto.html

However NetBSD/vax does not support multi-CPU systems.

The nm command is indeed the definitive way to find symbols in object files and libraries.

Fundamentally of course this problem is rooted in the fact that VAX hardware does not use IEEE-754 floating point (it uses its own format)[1], and GCC for VAX does't support -msoft-float either. libm on NetBSD/vax, or pretty much any Unix on VAX, is it's own beast.

There is an fmod(3) implementation in NetBSD's src/lib/libm/noieee_src/n_fmod.c so using it instead of remainder() should work on NetBSD/vax.

[1] http://www.eecs.berkeley.edu/~wkahan/ieee754status/754story.html

[puddles123fluffy]

Hello,

The patch seems to have worked. Funnily enough, I was just reading the float interview linked in the last comment the other day! Thank you again for your help and for maintaining ksh!

I will leave the issue open so you can close it after committing the fmod() patch.

Thanks again!

[McDutchie]

Fixed in 08fb672. Thanks @puddles123fluffy for the report and @robohack for the useful info!