[eval] [FIXED] Fix subject tree intersection to match literals when followed by additional wildcards

[Uzay-G]

Mirror of nats-io#6827 (MERGED) for Orpheus review evaluation.

Upstream: nats-io#6827

---

Original PR description:

The subject tree intersection could miss non-wildcard matches if they were followed up by further extended wildcards in other sublist filters. This is now fixed.

Fixes nats-io#6824.

Signed-off-by: Neil Twigg neil@nats.io
Co-authored-by: Maurice van Veen github@mauricevanveen.com

[orpheus-by-fulcrum-dev-uzay-g]

Review started.

[orpheus-by-fulcrum-dev-uzay-g]

All checks passed.

Expected behavior

• A JetStream consumer with filter subjects ["stream.A", "stream.A.>"] delivers all messages on stream.A (the literal) and stream.A.* / stream.A.> (the wildcard children)
• The IntersectStree function finds literal subject matches even when the sublist level also contains wildcard (fwc/pwc) children
• No duplicate messages are delivered when filter subjects overlap structurally (literal + extended wildcard at the same prefix)
• Existing sublist intersection behavior (literals, PWC, FWC, overlapping patterns) is preserved without regressions

What happens

• ✅ Happy path works: consumer with ["stream.A", "stream.A.>"] delivers all 10 messages on stream.A and all 8 messages on child subjects, with no duplicates (18 total)
• ✅ Bug confirmed on pre-fix build: same consumer shows 0 unprocessed messages on the old code, proving the fix resolves the issue
• ✅ No duplicates: consumer pending count matches exactly the number of published messages across all tested patterns (["stream.A", "stream.A.>"], ["stream.A", "stream.A.>", "stream.B"], ["stream.A", "stream.B"])
• ✅ Real-time delivery works: messages published after consumer creation are also correctly delivered through overlapping filters
• ✅ All 61 sublist tests and the new TestJetStreamConsumerDeliverAllOverlappingFilterSubjects test pass with no regressions

Detailed evidence

Setup

export PATH=$PATH:/usr/local/go/bin:$HOME/go/bin

# Build the fixed server
go build -o /home/agent/nats-server/nats-server .
# → nats-server: v2.11.2-RC.2

# Build the pre-fix server for comparison
git checkout 82cb3edb0^ -- server/sublist.go
go build -o /tmp/nats-server-old .
git checkout HEAD -- server/sublist.go

Demo 1: Happy path — overlapping literal + FWC filters

# Start fixed server with JetStream
/home/agent/nats-server/nats-server -js -sd /tmp/nats-data -p 4222 &

# Create stream
nats stream add TEST --subjects "stream.>" --storage memory ...
# → Stream TEST was created

# Publish 10 messages to stream.A
for i in $(seq 1 10); do nats pub stream.A "message-$i"; done

# Create consumer with overlapping filter subjects
nats consumer add TEST OVERLAP_CONSUMER \
  --deliver all \
  --filter "stream.A" --filter "stream.A.>" \
  --ack explicit --pull --defaults
# → Unprocessed Messages: 10   ← correct, all 10 found

# Fetch all messages
nats consumer next TEST OVERLAP_CONSUMER --count 10 --timeout 5s
# → 10 messages received on stream.A, cons seq 1-10, str seq 1-10

Demo 2: Child subject delivery via FWC

# Publish to child subjects
for i in $(seq 1 5); do nats pub stream.A.child "child-msg-$i"; done
for i in $(seq 1 3); do nats pub stream.A.deep.nested "deep-msg-$i"; done

# Fetch from consumer
nats consumer next TEST OVERLAP_CONSUMER --count 10 --timeout 5s
# → 8 messages received: 5 on stream.A.child (cons seq 11-15), 3 on stream.A.deep.nested (cons seq 16-18)

Demo 3: No duplicates

# Create new consumer with ack=none to verify total count
nats consumer add TEST DEDUP_CHECK \
  --deliver all \
  --filter "stream.A" --filter "stream.A.>" \
  --ack none --pull --defaults
# → Unprocessed Messages: 18   ← exactly 10+5+3, no duplicates

# Fetch and verify unique stream sequences
nats consumer next TEST DEDUP_CHECK --count 20 --timeout 5s | grep "subj:" | sort
# → 18 lines, each with unique str seq 1-18

Demo 4: Bug confirmed on pre-fix build

# Start OLD server (built from parent commit before fix)
pkill -9 nats-server
/tmp/nats-server-old -js -sd /tmp/nats-data-old -p 4222 &

# Same setup: create stream, publish 10 messages to stream.A
nats stream add TEST --subjects "stream.>" --storage memory ...
for i in $(seq 1 10); do nats pub stream.A "message-$i"; done

# Create same consumer
nats consumer add TEST BUG_CHECK \
  --deliver all \
  --filter "stream.A" --filter "stream.A.>" \
  --ack none --pull --defaults
# → Unprocessed Messages: 0   ← BUG: should be 10, but the old intersection misses the literal

Demo 5: Mixed complex filters

# Restart fixed server, create stream, publish various messages

# stream.A (literal) + stream.A.> (fwc) + stream.B (literal)
nats consumer add TEST MIXED_COMPLEX \
  --deliver all \
  --filter "stream.A" --filter "stream.A.>" --filter "stream.B" \
  --ack none --pull --defaults
# → Unprocessed Messages: 19   ← 10+5+3+1, correct

# Overlapping PWC+literal correctly rejected by server
nats consumer add TEST PWC_LITERAL \
  --filter "stream.*" --filter "stream.A" ...
# → error: consumer subject filters cannot overlap (10138)

Demo 6: Real-time delivery

# Fresh server, create consumer FIRST, then publish
nats consumer add TEST PUSH_OVERLAP \
  --deliver all \
  --filter "stream.A" --filter "stream.A.>" \
  --ack explicit --pull --defaults
# → Unprocessed Messages: 0

# Publish after consumer creation
for i in $(seq 1 5); do nats pub stream.A "realtime-$i"; done
for i in $(seq 1 3); do nats pub stream.A.child "realtime-child-$i"; done

nats consumer info TEST PUSH_OVERLAP | grep Unprocessed
# → Unprocessed Messages: 8

nats consumer next TEST PUSH_OVERLAP --count 10 --timeout 3s | grep "subj:"
# → 5 stream.A messages + 3 stream.A.child messages, all delivered

Test suite

# Sublist intersection tests (including new FWCExtended)
go test -v -run "TestSublistInterestBasedIntersection" ./server/
# → PASS (all 10 subtests)

# JetStream consumer overlapping filter test
go test -v -run "TestJetStreamConsumerDeliverAllOverlappingFilterSubjects" ./server/
# → PASS

# All sublist tests (regression check)
go test -v -run "TestSublist" ./server/
# → PASS (61 tests, 0.034s)