tests: the one-line skip_all is copied into five scripts; lib.sh can carry it #1186

(posted by an agent, not by @lneto)

skip_all() { ktap_header; ktap_plan 1; ktap_skip "$1"; ktap_totals; exit 0; } is written identically in tests/device/held.sh:53, tests/rcu/entry_release.sh:42 and tests/runtime/collected.sh:29, and #1168 and #1169 copy it into tests/device/returns.sh and tests/control/session.sh. tests/lib.sh, which every test sources, has no skip helper, so each new test copies it again. Moving the one-liner into tests/lib.sh and deleting the copies is a tree-wide change of its own; the scripts whose skip_all skips each planned case, in notifier, probe, sched, skb, tc and xdp, keep theirs.

Found in the review of #1169.

Update: more than forty scripts skip a whole suite with echo "# SKIP: ..."; ktap_totals; exit 0 after ktap_plan N (61 such lines in 47 files at b64c6f9). The comment is not a KTAP result: the plan is left unmet, ktap_totals prints skip:0 since only ktap_skip counts, and tests/run.sh, which sums the # Totals: lines, reports no skip for the script. The counted form, one ktap_skip per planned case, is in tests/crypto/context.sh and tests/fsnotify/run.sh. Three scripts also spell the same check before a stimulus only a fixed module survives, the loaded module is the installed one and carries the refusal: tests/runtime/self_stop.sh, tests/netlink/rtnl.sh and #1247's tests/cpu/run.sh. tests/lib.sh can carry that check beside skip_all, and the sweep that moves the copies can turn the comments into counted skips. Found in the review of #1247.

Update: tests: the check that the loaded module is the installed build and carries a refusal is open coded in each test that needs it (#1342)

Before a stimulus that would wedge the host on a build without the fix, tests/netlink/rtnl.sh, tests/runtime/self_stop.sh, tests/socket/rtnl.sh and tests/crypto/context.sh each spell the same test: /sys/module//srcversion equal to modinfo -F srcversion <m>, and a grep -aF of the refusal in modinfo -n <m>. tests/hid/context.sh and tests/darken/run.sh from #1342 add two more copies, and #1194 and #1276 carry one each. A single helper in tests/lib.sh, for example carries <module> <message>, would replace them all and keep one spelling of what AGENTS.md, Tests, asks for.

Update: #1276 merged its copy as tests/runtime/percpu_netfilter.sh, so tests/runtime now spells the check twice, there and in self_stop.sh, which skips on each half separately where tests/netlink/rtnl.sh has one skip. A carries <module> <message> helper in tests/lib.sh folds both into one line each. Found in the review of #1276.

Update: tests: the one-line skip_all is spelled in five scripts instead of lib.sh (#1509)

Five test scripts define the same one-line skip_all() { ktap_header; ktap_plan 1; ktap_skip "$1"; ktap_totals; exit 0; }: tests/device/held.sh, tests/device/returns.sh, tests/runtime/collected.sh, tests/rcu/entry_release.sh and tests/hid/device.sh (#1509). tests/lib.sh, which every test sources, has ktap_skip but no way to skip a whole test. The multi-line variants in tests/notifier and tests/probe skip one line per planned case and do not fit the same helper.

Fix: define skip_all once in tests/lib.sh and drop the five copies.

Contract: undocumented

Update: tests: skip_all is spelled in five scripts and N-skip loops in four more; tests/lib.sh carries neither (#1551)

Two shapes are copied across test scripts instead of living in tests/lib.sh.

1. One case that skips the whole script. This one-liner is spelled identically in five places:

   skip_all() { ktap_header; ktap_plan 1; ktap_skip "$1"; ktap_totals; exit 0; }

   • tests/device/returns.sh:52
   • tests/device/held.sh:53
   • tests/runtime/collected.sh:29
   • tests/rcu/entry_release.sh:42
   • tests/netlink/umbrella.sh:28 (added by netlink: one call shape across session, genl, rt and nl80211 #1551)

2. N skips under a plan already printed, then the totals and exit:

   • tests/netlink/list_family.sh's skip_all (netlink: one call shape across session, genl, rt and nl80211 #1551)
   • tests/fsnotify/perm.sh:27 perm_skip
   • the inline loops in tests/socket/orphan.sh:80 and tests/netlink/addr_list.sh:65,75

   tests/notifier/{inside,stop,replay}.sh and tests/probe/{raise,dropreason}.sh define multi-line skip_all variants worth reading alongside.

Fix: one helper per shape in tests/lib.sh, next to ktap_skip, and the scripts call it. For example ktap_skip_all <reason> for the first shape, and ktap_skip_n <count> <reason> followed by the totals and exit 0 for the second. This touches only tests.

Contract: undocumented for shell; AGENTS.md Lua style, read here by analogy: "Code two scripts of one directory repeat goes in a module beside them that both require" (tests/lib.sh is the shell scripts' shared module)

Update: tests: the one-line skip_all is copied into five scripts; lib.sh can carry it (#1643)

#1643 adds a twelfth identical copy, skip_all() { ktap_header; ktap_plan 1; ktap_skip "$1"; ktap_totals; exit 0; } in tests/rcu/foreach_overflow.sh. At 3fb2799, eleven scripts carry the same line, among them tests/device/held.sh, tests/rcu/entry_release.sh, tests/rcu/cycle_release.sh and tests/runtime/collected.sh, and tests/lib.sh still has ktap_skip but nothing that skips a whole test. A skip_all in tests/lib.sh, with the copies deleted in one sweep, ends the copying. The variants that skip one line per planned case keep theirs. Found in the review of #1643.

Contract: undocumented (test harness cleanup; AGENTS.md C style: "When a two line pattern repeats in every method, collapse it into one helper or macro")

Update: tests/netlink: route_adddel.sh and rule_adddel.sh add two more uncounted whole-script skips (#1559)

On a kernel without CONFIG_IP_MULTIPLE_TABLES, the policy-routing skip in tests/netlink/route_adddel.sh (after ktap_plan 4) and tests/netlink/rule_adddel.sh (after ktap_plan 5) prints # SKIP: ..., calls ktap_totals and exits 0. That leaves the KTAP plan unmet and prints # Totals: pass:0 fail:0 skip:0. tests/run.sh sums those Totals lines, so it reports neither script as passed or skipped. Seen by running both scripts with an ip shim that refuses -4 rule show. Each file's own module skip has the same shape. tests/netlink/list_family.sh, which uses the same probe, counts its skips with one ktap_skip per planned case. Fix: the sweep #1186 describes, turning the echo "# SKIP" form into counted skips through a tests/lib.sh helper, covers these two sites with the rest.

Contract: undocumented: .agents/rules/tests.md says only "skip, do not fail, when the kernel lacks a required config", which both scripts meet

Update: tests/netlink: route_adddel.sh and rule_adddel.sh add two more uncounted whole-script skips (update to #1186) (#1798)

On a kernel without CONFIG_IP_MULTIPLE_TABLES, the policy-routing skip in tests/netlink/route_adddel.sh (after ktap_plan 4) and tests/netlink/rule_adddel.sh (after ktap_plan 5) prints # SKIP: ..., calls ktap_totals and exits 0. That leaves the KTAP plan unmet and prints # Totals: pass:0 fail:0 skip:0. tests/run.sh sums those Totals lines, so it reports neither script as passed or skipped. Seen by running both scripts with an ip shim that refuses -4 rule show. Each file's own module skip has the same shape. tests/netlink/list_family.sh, which uses the same probe, counts its skips with one ktap_skip per planned case. Fix: the sweep #1186 describes, turning the echo "# SKIP" form into counted skips through a tests/lib.sh helper, covers these two sites with the rest.

Contract: undocumented: .agents/rules/tests.md says only "skip, do not fail, when the kernel lacks a required config", which both scripts meet

Update: tests: passed() is copied into three tests/monitor scripts and tests/thread/driver.sh has the same read; tests/lib.sh carries none (#1879)

Three scripts in tests/monitor define the same one-line passed() { dmesg_since | grep -qE "PASS[[:space:]]$1\$"; }: collector.sh:27 and name.sh:26 on master, and hook.sh:36 on the monitor-hook-off-lock branch. Each one follows it with the same loop over its case names: passed "$case" || fail "$case: ..."; ktap_pass "$case". tests/thread/driver.sh writes the same read as reported() plus a verdict() that passes or fails a case, but only for the thread suite. tests/lib.sh, which every script sources, has no helper that reads the line tests.lib prints for a passed case, so each new suite that runs its cases through tests.lib copies it again.

Fix: move reported/verdict from tests/thread/driver.sh into tests/lib.sh, next to ktap_pass (or a ktap_cases <case>... that runs the loop), and delete the copies in tests/monitor/{collector,name,hook}.sh, in the same sweep that moves skip_all and the carries check. This touches only tests.

Found in the review of the monitor-hook-off-lock branch (#1878).

Contract: undocumented; by analogy, .agents/rules/lua.md: "Code two scripts of one directory repeat goes in a module beside them that both require" (tests/lib.sh is the shell scripts' shared module)