Skip to content

Commit 2c06663

Browse files
docs: record v0.7.0 limitations and next pass
Co-authored-by: Cursor <cursoragent@cursor.com>
1 parent 6ca5608 commit 2c06663

1 file changed

Lines changed: 38 additions & 15 deletions

File tree

‎docs/ROADMAP.md‎

Lines changed: 38 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -13,7 +13,7 @@ Effort: S (under ~1 hour), M (a few hours), L (a day or more).
1313
## Shipped in 0.6.x / 0.7.0
1414

1515
### C1 (DEFERRED -> tracked in #23). Guard the events sends against a non-draining consumer
16-
- **Status:** deferred; `runner.go` is intentionally left untouched. The full reasoning lives in issue #23.
16+
- **Status:** deferred. PR1 made `EventError` drain so `EventDone` still arrives on a reliability abort, but `EventResult` and `EventDone` sends remain unguarded. The deferral still holds.
1717
- **Resolve when:** a consumer that stops draining the events channel actually exists. Fix both sends together with a runner test that hangs without the guard and passes with it.
1818

1919
### C2 (DONE). Remove dead `dns.Resolve` and its doc reference
@@ -50,17 +50,39 @@ one `*net.Resolver` per scan, wildcard fingerprints and per-branch detection,
5050
`-max-queries` plus recursion ceiling plus resolver preflight, hermetic DNS tests
5151
and release hygiene.
5252

53+
## Known limitations
54+
55+
These are constraints from the v0.7.0 resolver-hardening pass, not footnotes.
56+
57+
- **testdns serves A records only.** AAAA queries return NXDOMAIN, so the hermetic
58+
suite does not cover the AAAA path in `ResolveTypes` or the dual-type timeout
59+
budget. Anyone touching AAAA handling must extend the responder first.
60+
- **The reliability guard is exercised with injected timeouts, not simulate misses,**
61+
because simulate misses classify as NXDOMAIN and NXDOMAIN is excluded from the
62+
failure rate by design. Preserve that when refactoring the guard.
63+
- **Cap and ceiling notices reuse `EventWildcard`.** Working, but the kind name now
64+
understates what it carries. Renaming it to `EventNotice` is a clean follow-up
65+
that touches both consumers.
66+
- **The DNS wire format in `testdns_test.go` is hand-rolled** because
67+
`golang.org/x/net/dns/dnsmessage` is not in the module graph and PR6 forbade new
68+
dependencies. Adding `golang.org/x/net` later would let the responder shrink
69+
considerably.
70+
- **The `go` directive is `1.24.2`, not patchless `1.24`.** charmbracelet/bubbles
71+
requires 1.24.2, so `go build` refuses a `go 1.24` line until tidy bumps it.
72+
On PowerShell, unquoted `-go=1.24` is parsed as `-go=1`.
73+
5374
## Next pass
5475

5576
Do not expand this list into drive-by fixes in the same PR that notices them.
5677

57-
### N1. `-rate` ticker implementation
58-
The shared stdlib ticker gate works but is coarse (burst at tick boundaries, no
59-
token bucket). Revisit pacing without changing the `scan.Event` boundary.
78+
### N1. `-rate` ticker under-delivers
79+
A ticker channel drops ticks when workers are busy, so observed QPS can fall
80+
below `-rate` under load. Revisit pacing without changing the `scan.Event`
81+
boundary.
6082

6183
### N2. Text-format record display
62-
`text` output still prints `Found: <domain>` and drops record type/value. JSON
63-
and CSV already carry them.
84+
`text` output still prints `Found: <domain>` and discards record type/value.
85+
JSON and CSV already carry them.
6486

6587
### N3. Per-type shared timeout budget
6688
`ResolveTypes` uses one `context.WithTimeout` for the whole type loop, so AAAA
@@ -69,16 +91,17 @@ budget that cannot go negative.
6991

7092
### N4. `-dL` multi-domain input
7193
A domain-list file (one apex per line) so a single process can scan more than
72-
one target. Needs a clear interaction with `-recursive`, `-max-queries`, and
73-
the reliability guard.
94+
one target. This would retire the Windows-hostile `examples/multi_domain_scan.sh`.
95+
Needs a clear interaction with `-recursive`, `-max-queries`, and the reliability
96+
guard.
7497

75-
### N5. Verbose/progress stderr interleaving
76-
`-v` and `-progress` write to stderr without coordination, so verbose lines can
77-
overwrite the carriage-return progress line.
98+
### N5. Verbose logging bypasses the output mutex
99+
`-v` writes to stderr without going through the output writer mutex, so verbose
100+
lines mangle the carriage-return progress line.
78101

79-
### N6. TUI viewport clamp
80-
The scan-view viewport can scroll past content or fail to follow the tail on
81-
fast result streams.
102+
### N6. TUI viewport height goes negative below 9 rows
103+
The scan-view viewport height is unclamped and can go negative on short
104+
terminals.
82105

83106
### N7. Event-channel send guards (C1 / #23)
84-
Still deferred until a non-draining consumer exists.
107+
Still deferred. The EventError drain does not guard EventResult or EventDone.

0 commit comments

Comments
 (0)