The `byedpi` egress kind, the `openwrt/byedpi` package (`ciadpi`), the readiness
endpoint and the panel plate are gone. D13 is not deleted from DECISIONS.md; it
is REVERSED there, with the reason, because the reason is the whole point.
D13 adopted an external desync process on an observation: the engine's own
`tls_fragment`/`tls_record_fragment` were tried against a live ISP and did not
get through, so the method was judged too weak for anything past "just fragment
the ClientHello". The method was never tried. `common/tlsfragment` dropped a
number of labels equal to the number of DOTS in the name, and a name always has
one more label than it has dots — so the cut always landed inside the FIRST
label. `www.youtube.com` was split inside `www` and `youtube` went to the wire
in one piece, which is the word the DPI matches on. Of six blocked names exactly
one got through: `youtube.com`, the one whose first label IS the blocked word.
That defect is fixed (815011dfb, efb2177f4). With it fixed the built-in presets
do the job the external process was brought in to do, and the process is 100 KB
of binary, a second procd service, a second UCI file, a port that agreed with
our egress by hand-written comment only, a readiness prober, a five-state
service model and a panel plate — all to work around fifteen lines of ours.
So this is not "ByeDPI turned out to be bad". It is a good tool that turned out
not to be needed, and the reason we thought it was needed was ours.
A CONFIG THAT STILL SAYS `type 'byedpi'` IS THE PART THAT NEEDED WORK. Nothing
is migrated and nothing is rewritten: the kind stays unbuildable, therefore
fail-closed — no outbound, no mark, no `ip rule`, no routing table, so every
node, group and rule bound to it is blocked rather than released onto the plain
WAN. A migration to `direct` was considered and rejected: it is the only rewrite
that leaves the egress routing at all, and it would silently turn a blocked
egress into a live plain-WAN path with the router's real address — by an
upgrade, on a config nobody touched. `CurrentSchemaVersion` is therefore not
bumped either: no stored field changes meaning, and a bump would only make this
build's configs unreadable to an older daemon for no gain.
What changes is what the operator is TOLD. `model.RetiredEgressTypes` is a
closed, positive table read by BOTH `ValidateEgresses` and the generator (one
copy of the sentence, because two copies drift). It names the removal, denies
that it is a typo, says nothing is built and that the traffic is blocked rather
than leaked, names the replacement (`direct`/`interface` with `dpi 'record'`),
refuses to promise which preset defeats a given ISP, and says `apk del byedpi`.
The generic "unknown type" is still there and still says something different, on
purpose: "we took this kind away" and "you mistyped something" send an operator
to different places, and a value that was correct on the day it was written must
not be reported as a spelling mistake. The type list stays closed and positive —
`interface`, `direct`, the alias `tunnel` — and `EgressTypeKnown` does NOT admit
the retired kind: being told it was removed and having it work anyway is worse
than either alone.
`Egress.Port` goes with the kind: no surviving egress dials anything, so the
option is no longer parsed and drains out of /etc/config/shater on the next
render, the same way the deleted per-group probe_url/probe_interval did.
Tests, verified by mutation, each failing by name:
- drop the retired branch in `ValidateEgresses` -> the retired kind is
reported as "is not one of interface/direct" and
TestRetiredEgressTypeIsReportedByTheValidator fails on both spellings;
- drop it in the generator -> "unknown type \"byedpi\"" and
TestRetiredEgressTypeIsReportedByTheGenerator fails;
- the FAIL-OPEN mutation, which is the one that matters: let `byedpi` fall
into the `direct` arm and be a known type -> four tests fail, including the
two that check no outbound is emitted. A removal that quietly starts routing
the traffic it used to block, under a reassuring message, is the failure with
the worst consequence;
- the panel half: empty RETIRED_EGRESS_TYPES -> two egressEdit tests fail.
Controls beside the claims: `interface`, `direct`, the `tunnel` alias and the
empty synonym must still resolve, warn about nothing and emit an outbound
(TestSupportedEgressTypesAreUntouched), and never-supported values — `proxy`,
`block`, `wireguard`, `byedpi2`, `bye dpi`, `sorcery` — must NOT draw the
removal sentence, which names a replacement for something that never existed.
CI and docs: the feed loses its fourth package everywhere the four were named —
`apk upgrade shaterd shater-core luci-app-shater`, in CLAUDE.md, both READMEs,
INSTALL.md, the release body and `shaterd`'s own diag bundle. The version
exception (byedpi carried upstream's version, ours come from the git tag) is
gone with it, so ci/version.sh and ci/sdk-build-apk.sh no longer have an
exception to remember and the "expected >=4 of OUR .apk" collect check is now 3.
INSTALL.md §5.3 gains the half a feed cannot do: dropping the package from the
feed does not take it off a router it is already on, so `apk del byedpi` is
written down, with what it removes and why it is safe.
Panel: 368 tests -> 339. Deleted with the mechanism they covered:
byedpiReady.test.ts, byedpiAge.test.ts, byedpiRefusal.test.ts (34 tests);
egressEdit.test.ts gains 5 for the retired-type sentence.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BHw89tdWddzhjUc4bAH4tS
omarandClaude Opus 5
4869d62e02