cm.events grows TCometFreqMatch / TCometFreqMatches plus
hook signatures: TCometOnFreqRequest (now returns matches
instead of per-file accept), TCometOnFreqNak,
TCometOnLstRequest, TCometOnLstItem, TCometOnLstEnd.
TCometSessionConfig wires them up.
cm.xfer handles all five wire packets:
NPKT_FREQ (in: HandleFreq -- pattern fed to consumer
hook, returned matches queued into its
own FFreqQueue, drained ahead of provider
files via StartFreqOutbound)
NPKT_FREQNAK (in: HandleFreqNak fires consumer hook +
decrements awaiting-responses counter)
NPKT_LSTREQ (in: HandleLstReq -- emits one
NPKT_LSTITEM per match + NPKT_LSTEND)
NPKT_LSTITEM (in: parses name/size/time/desc, fires
Cfg.OnLstItem)
NPKT_LSTEND (in: fires Cfg.OnLstEnd, decrements
awaiting-responses)
Consumer-side request API (RequestFiles, RequestListing)
queues patterns; xfer drains them as the FIRST action in
Step (DrainPendingRequests) so they reach the peer before
either side decides "nothing to send -> end-of-batch".
DeferEndOfBatch / CompleteBatch added too for the late-
request scenario.
Awaiting-responses tracking via FAwaitingResponses (separate
from FRxEndSeen so a late request emission can't clobber a
previously-set peer-end-of-batch state).
Three wind-down robustness fixes that surfaced while writing
the LST test:
1) ReadOneFrame on Recv = -1 used to FailSession + Exit
immediately, dropping any frames already buffered in
FRecvBuf. Now parses buffered bytes first; only fails
if no parseable frame remains AND the phase isn't a
tolerated wind-down.
2) cmpShutdown used to Pump exactly once before
transitioning to cmpDone. When peer queued multiple
trailing frames before its ShutdownWrite (LSTITEMs +
LSTEND + NPKT_END all in flight), the trailing ones
were dropped. Now while-Pump until both send queue
drained AND no more parseable frames.
3) NPKT_END was sent via raw CMWritePacket which bypasses
EmitFrame's encryption layer. On a CRYPT-active
session that meant NPKT_END went out plaintext, the
receiver tried to decrypt it as ciphertext, and got a
CRC garbage frame. Now NPKT_END uses EmitFrame so it
gets encrypted alongside everything else.
Spec maintenance: removed the "Hydra-style" comparative
language from FSP-COMET-001 section 12; the spec now
describes Comet on its own terms.
12 tests pass (added test_freq_lst); all 7 cross-targets
clean.
That closes the v0.1.0 spec-parity gap:
CRAM-MD5, ED25519, KEYEX/CRYPT, RPOS, ZLIB, FREQ, LST
-- all done.
Ken Johnson
8d0a27b82e