In D44548#1068628, @ngie wrote:

Specifying this in the Makefile is significantly less code than doing it in the atf tests:

TEST_METADATA.lorder_test+=  required_programs="ar cc lorder nm"

This metadata ends up in the Kyuafile.

Update test to latest sync, and add nm in requiere list too.

New version using requiered_progs.
But anyway which port should insall all the missing sg_* binaries ? I don’t find instruction about them.
Because right now they are ignored:
https://ci.freebsd.org/job/FreeBSD-main-amd64-test/lastCompletedBuild/testReport/sys.cam.ctl/prevent/allow/

In D46853#1068383, @imp wrote:

I have a slight preference for require.prog since the counts are better aligne. Also, that lets people install the dist tests on their maybe customized systems for QA. I'll likely need to skip these for bsd-user too, but that's also beyond the scope. But I'm not insisting because this solves the immediate need

Thanks I will commit it

In D44548#1016031, @des wrote:

I think D44558 is a better approach.

In D44548#1015876, @des wrote:

Why are you testing lorder, which is part of the toolchain, on a system without a toolchain?

On 2 identical servers (Intel Xeon CPU E5-2697A v4 @ 2.60GHz), one running the unpatched and the other this patched review, here are the results of this script (extracting the "real" in seconds from the time command):

#!/bin/sh
set -eu
lenght=100000000000
sizes="8000 32000 64000 128000 256000 512000 1000000"
cmd=""
for s in $sizes; do
        echo "Benching with size $s (lenght: $lenght)..."
        for i in $(seq 3); do
                time -p -o bulk.$s.$i.txt ./unix_bulk $lenght $s $s
        done
        grep real bulk.$s.*.txt | cut -d ' ' -f 2 > bulk.$s.data
        cmd="$cmd bulk.$s.data"
done
ministat -w 74 -s $cmd

On similar review (https://reviews.freebsd.org/D44005) the idea was to still build and install it, but "skip" the test if jail binary was missing.
@jamie Do you have a preference ?

In D44005#1004021, @lwhsu wrote:

I don't really object this approach, but I don't really support it either. Do you think it's okay to let the test case check if jail utility exists, and skip running if no?