INN Issue Triage

INN Issue Triage

Open issues of InterNetNews/inn, checked against upstream main at 6843efceb (2026-09-12): still valid or already fixed, effort, whether the e2e test harness can test it, and a suggested action. Triaged 2026-10-23. Click a row for the evidence and notes.

Issues marked our PR have a fix we opened; the PR column links it. These fixes are draft pull requests in the fork go-while/inn. They are reviewed there first and filed upstream afterwards.

Updated 2026-11-11: reconciled the fork against Julien ÉLIE's recent upstream commits. His own OpenBSD fixes are now on main — the ckpasswd DES/bcrypt test data (ebeb5e3c7, closes #201) and the xref test (2a1b44547, closes #203) — so our duplicates of both drop when fork PR #36 rebases, leaving upstream #381 with only the FreeBSD, NetBSD and DragonFly test fixes.

Updated 2026-11-10: 18 new fork draft PRs, #40 to #57: seven code changes (#159, #162, #179, #210, #45, #328, #333), the innfeed 500/501 handling of #40 and a first step for #317 (CAPABILITIES before MODE STREAM), and nine documentation PRs. Independent review is still pending. New upstream issue #384 for the timecaf lock wait (fork PR #38).

Earlier updates (2026-10-23 to 2026-11-09)
  • 2026-11-09: Julien ÉLIE closed #366 (innd accepted sizes) as completed; our e2e test had confirmed the fix. It and the closed #201 and #203 are off the list. New upstream question #383: innd and nnrpd pass a bare CR in article lines through, which RFC 3977 section 3.1.1 does not allow. #321 (old TLS versions in nnrpd) now has fork PR #39.
  • 2026-11-08, later: all 34 open fork PRs rebased onto upstream main 4c5218607, retested and checked again by GPT-6 Astra. Fork branch integration (main with all of them merged) passes CI everywhere and our full e2e suite (62 tests). New fork PR #38 (timecaf lock wait in expireover). The ovsqlite WAL visibility delay is now upstream issue #382.
  • 2026-11-08: opened #380 (Clang -Wpre-c11-compat with glibc 2.43) and #381 (make check on FreeBSD, NetBSD and DragonFly) upstream.
  • 2026-11-07: Julien ÉLIE committed our #379 (glibc 2.43 const qualifiers) as 39aad9cf3 and fixed #201 and #203 on OpenBSD himself.
  • 2026-11-06, evening: every open fork PR reviewed by two more models besides its author; 16 of 32 got follow-up commits. New CI branch main-matrix builds main and 2.7 on Ubuntu (gcc, clang, ARM64, ASan, UBSan), 20 Linux distributions, four BSDs and OmniOS. The e2e branch tests every storage, overview and history method.
  • 2026-11-06: Julien ÉLIE committed our #376, #377 (innd signals while idle or busy) and #378 (nnrpd 502 for COMPRESS while active), plus his own signal fixes. The e2e branch tests compressed POST and IHAVE (#269: 2.6.x to 2.7.1 hang, fixed in 2.7.2).
  • 2026-10-26: PR column and our PR marks added.
  • 2026-10-23: first triage of all open issues against main 6843efceb.
Who reviewed which fork PR
Fork PRsWritten byReview 1Review 2Final check
#1 #2 #3 #4 #6 #7 #10 #11 #12 #14 #15 #16 #17 #20 #21 #22 #23 #24 #27 #28 #29 #30 #31Claude OpusGPT-6.1 TerraGPT-6.1 SolGPT-6 Astra
#8 #9 #13 #18 #19 #25 #26 #32 #33Claude OpusGPT-6.1 SolGPT-6.1 TerraGPT-6 Astra
#34GPT-6.1 SolClaude Opus–GPT-6 Astra
#35GPT-6.1 TerraClaude OpusGPT-6.1 SolGPT-6 Astra
#36 (upstream #381)GPT-6.1 Sol, Claude OpusGPT-6.1 TerraClaude Fable 5.1GPT-6 Astra
#37 (upstream #380)Claude OpusGPT-6.1 SolGPT-6.1 Terra, Claude Fable 5.1GPT-6 Astra
#38GPT-6.1 SolGPT-6.1 Terra–GPT-6 Astra
#39 #46 #47 #48 #49 #50 #51 #52 #53 #54 #57Claude Opus 5.5pending
#40 #41 #42 #44 #45 #55Claude Sonnet 5.5pending
#43 #56Claude Sonnet 5.5, Claude Opus 5.5Claude Opus 5.5 (#56; fixes applied)pending

Each fork PR was also rebased onto 4c5218607 by GPT-6.1 Sol on 2026-11-08 and verified again by GPT-6 Astra. Details: one review file per PR.

Issue Title Action PR Status Effort Component Milestone Prio e2e

Our pull requests without an upstream issue

Bugs we found while fixing or testing other issues, or on the test machines (24 Linux distributions, Debian GNU/Hurd among them). There is no issue row for them in the table above. Updated 2026-11-06.

PRComponentWhat it fixesStatus
InterNetNews/inn#379
fork: go-while/inn#5
build glibc 2.43 const warnings
glibc 2.43 makes strchr, strstr, bsearch and friends return const pointers for const input. gcc warns at 16–19 places, and -Werror builds fail: since early November that includes the CI build on GitHub's ubuntu-26.04 runner, for upstream main too. Seen on Fedora 44/45/46, Arch, Gentoo, Ubuntu 26.04, Debian GNU/Hurd. Cherry-picks cleanly onto 2.7.
committed upstream unchanged (39aad9cf3, 2026-11-07), milestone 2.7.5; 2.7 backport ready in the fork
go-while/inn#21 innd Filter __BODY__ without the article terminator
The Perl and Python article filters of innd get a __BODY__ that ends with part of the . line ending the article, and the two filters see different bodies.
draft in fork; cross-reviewed
go-while/inn#28 overview tdx-util: same :bytes and :lines as innd
The overview that tdx-util -R rebuilds has other :bytes and :lines values than innd wrote; innd and makehistory also differ in how they count stuffing dots.
draft in fork; cross-reviewed
go-while/inn#29 samples filter_nnrpd.pl: checkincludedtext reads $body
The checkincludedtext check of the sample nnrpd filter reads $hdr{__BODY__}, which nnrpd doesn't set, so it never rejects anything.
draft in fork; cross-reviewed
go-while/inn#32 history (2.8) hissqlite: warn when SQLite cannot use WAL mode
Where SQLite cannot use WAL mode (no shared memory: some network filesystems, GNU/Hurd), innd writes the history silently and every read-only open (nnrpd, grephistory) fails. Now a warning; the unit test skips its read-only checks there. Found on Debian GNU/Hurd.
draft in fork; cross-reviewed
go-while/inn#33 overview (2.8) ovsqlite: warn when SQLite cannot use WAL mode
ovsqlite-server's "cannot enable WAL mode" warning never fired, because the pragma succeeds and keeps the old journal mode. nnrpd already falls back to the server. The integration test now skips its direct-reader checks there. Found on Debian GNU/Hurd.
draft in fork; cross-reviewed
go-while/inn#34 overview ovsqlite: keep the -I cutoff in makehistory rebuilds
makehistory -O -I on ovsqlite dropped articles found out of number order: the first article added to an empty group raised its low mark, and lower ones were then refused as too old. Found by the method-matrix e2e test.
draft in fork; written by GPT-6.1 Sol; reviewed by Claude Opus and GPT-6 Astra
go-while/inn#35 expire (2.8) expire: exit 0 after an in-place hissqlite expiry
With hismethod: hissqlite, expire did its work but exited 1 with cannot unpause server: it unpaused an innd it never paused. Found by the method-matrix e2e test.
draft in fork; written by GPT-6.1 Terra; reviewed by Claude Opus, GPT-6.1 Sol and GPT-6 Astra
InterNetNews/inn#381
fork: go-while/inn#36
tests tests: make check passes on FreeBSD, NetBSD and DragonFly
make check failed on FreeBSD, NetBSD and DragonFly BSD: a hard-coded /usr/bin/perl and a crash after a configuration read error. Test fixes only. Julien fixed the other two parts himself on 2026-11-07: DES-only crypt test data (ebeb5e3c7, closes #201) and the OpenBSD xref test (2a1b44547, closes #203).
opened upstream (2026-11-08), awaiting review; 2 commits; reviewed by GPT-6.1 Terra, Claude Fable 5.1 and GPT-6 Astra; CI green on all BSDs, macOS and OmniOS
InterNetNews/inn#380
fork: go-while/inn#37
build configure: Clang -Wpre-c11-compat with glibc 2.43
glibc 2.43 implements strchr and friends with _Generic; Clang's -Weverything then fails make warnings at the first call. With #379 now in main, this alone makes the Clang build pass.
opened upstream (2026-11-08), awaiting review; reviewed by GPT-6.1 Sol, GPT-6.1 Terra, Claude Fable 5.1 and GPT-6 Astra; CI green on main
InterNetNews/inn#376 innd innd: exit on SIGTERM and SIGHUP while idle
innd only acted on a signal when the next connection or data arrived, or after the timer interval.
committed upstream (98edb4382, 2026-11-03)
InterNetNews/inn#377 innd innd: exit promptly on a signal while busy
A signal arriving while innd was busy, for example in a slow filter, was not acted on promptly either.
committed upstream (ba5d445eb, 2026-11-03)
InterNetNews/inn#378 nnrpd nnrpd: 502 for COMPRESS while compression is active
RFC 8054 asks for 502; nnrpd answered 503 to COMPRESS with another algorithm on a compressed connection.
committed upstream (f62b7c1e2, 2026-11-03)