test: fix "invert" commands in sharness tests

[laurentsenta]

Many tests assume that grep --invert-match "X" FILE means

Fail when FILE contains "X"

But the command actually means:

Fail when FILE contains only "X"

Which is wrong in most cases.

• We want to use test_should_not_contain instead.
• We probably want to forbid the use of grep -v, it's confusing maintainers.

Example:

kubo/test/sharness/t0140-swarm.sh

Lines 95 to 109 in 3bcc89b

	noAnnounceCfg='["/ip4/1.2.3.4/tcp/1234"]'
	test_expect_success "test_config_set succeeds" "
	ipfs config --json Addresses.NoAnnounce '$noAnnounceCfg'
	"
	test_launch_ipfs_daemon
	test_expect_success "Addresses.NoAnnounce affects addresses from Announce and AppendAnnounce" '
	ipfs swarm addrs local >actual &&
	grep -v "/ip4/1.2.3.4/tcp/1234" actual &&
	grep -v "/ip4/10.20.30.40/tcp/4321" actual &&
	ipfs id -f"<addrs>" | xargs -n1 echo >actual &&
	grep -v "/ip4/1.2.3.4/tcp/1234" actual &&
	grep -v "//ip4/10.20.30.40/tcp/4321" actual
	'

Test's Intent:

1. add the multiaddr "/ip4/1.2.3.4/tcp/1234" to the list NoAnnounce,
2. verify that the addr is not announced anymore.

But the command grep -v "/ip4/1.2.3.4/tcp/1234" actual succeeds even if the file contains /ip4/1.2.3.4/tcp/1234:

echo "/ip4/1.2.3.4/tcp/1234\nanything-else" > actual
grep -v "/ip4/1.2.3.4/tcp/1234" actual
# outputs "anything-else", exit code = 0

Which means that the assertion is a no-op.

suspects matching ack -l 'grep.*-v':

• cmd/ipfs/Rules.mk
• test/sharness/t0046-id-hash.sh
• test/sharness/t0119-prometheus.sh
• test/sharness/t0250-files-api.sh
• test/sharness/t0060-daemon.sh
• test/sharness/t0080-repo.sh
• test/sharness/t0230-channel-streaming-http-content-type.sh
• test/sharness/t0140-swarm.sh
• test/sharness/t0141-addfilter.sh
• test/sharness_test_coverage_helper.sh
• bin/mkreleaselog
• coverage/Rules.mk

Relates to #9651

[laurentsenta]

@guseggert From your work on sharness tests, could you advise which tests we should investigate more? And which we'll drop soon?

I'm quite confident with the fix on t0118 because it reveals an error, so I know the new test is correct,
but for tests like t0046, the fix doesn't expose an error which hurts my red-green reflex.

[lidel]

@laurentsenta is this still something you want to do? If so, I think below miss &&, and that is why error produces false-positive.

[laurentsenta]

I'll update this PR.

@lidel I think we still want to finish these, and maybe forbid the use of the --invert-match/-v it's confusing maintainers.

The tests here used grep -v "X" to check that X is NOT in the result,
but actually, the command succeeds as long as there are other lines.

echo "A\nB\nERROR" | grep -v "ERROR" 
echo $? # success because A anb B matches the regexp.

[laurentsenta]

I kept it but that // looks like a typo

[lidel]

Nice catch! Yes, it looks like copy&paste bug of sorts 🙈
Anyway, the test was broken because after #8177 we should test both addresses added via Announce and AppendAnnounce (so we need two specimens):

-noAnnounceCfg='["/ip4/1.2.3.4/tcp/1234"]'
+noAnnounceCfg='["/ip4/1.2.3.4/tcp/1234", "/ip4/10.20.30.40/tcp/4321"]'

Local test passed because other code filtered out non-LAN addr, but second one would fail, if we had no typo 🤯

I've pushed fix in 038e56d and should be green now 🤞

[lidel]

@laurentsenta yes, banning use of --invert-match/-v sgtm. We have somehow related pre-check in ./t0018-indent.sh, could you add something similar after we get rid of them, so it is not re-introduced?