AGENTS.md

Guidance for LLM agents working in this repository. Human-facing documentation lives in README.md; the remaining work is tracked in todo.md.

What this is

A ground-up TypeScript rewrite of larvitsmpp 0.4.0, published as @larvit/smpp 1.0.0. The branch started from an orphan commit — no history from 0.4.0 is carried over. The 0.4.0 source is still readable on the master branch of the same repository and is the reference for protocol behaviour, not for structure or style.

The library's value is its very small API. Do not grow the public surface without being asked.

Hard rules

These are not preferences. Breaking one is a defect.

Nothing throws. Every fallible function returns (or resolves to) a DTO carrying an optional err. No throw, no rejected promises, no exceptions as control flow. Node APIs that throw are wrapped at the boundary and converted into a result. Programmer errors (bad arguments) are results too.
Log messages are static strings. Every dynamic value goes into the log metadata. Never interpolate, never concatenate.
- GOOD: log.debug('sendSms() - splitting message', { parts: msgs.length, to });
- BANNED: log.debug('sendSms() - splitting into ' + msgs.length + ' parts');
No error event. Node makes an unhandled error event throw, which would break rule 1. Sessions emit sessionError, servers emit serverError.
No casts, no non-null assertions. as, as unknown as and ! are all banned. Parse untyped input once through a type guard at the boundary; everything past it is typed. noUncheckedIndexedAccess is on, so every lookup into a record or buffer is T | undefined until you handle it — that is the point, not an obstacle to route around.

Architecture

src/
	index.ts             Public surface. Named exports only, no default export.
	client.ts            client() -> { err, session }
	server.ts            server() -> { err, server }, server owns the listener + close()
	session.ts           Session: dispatch, events, and the collaborators below
	sms.ts               The live handle emitted as the 'sms' event (sendResp/sendDlr)
	dlr.ts               Delivery receipts: text and TLV parsing, receipt status codes
	dlr-merger.ts        DlrMerger: per-segment receipts counted into one MessageDlr
	error-from.ts        errorFrom(): whatever was thrown or rejected, as an Error
	expiring-groups.ts   ExpiringGroups: the capped, expiring store both of those share
	incoming-requests.ts Every request the peer sends: messages, receipts, links, unknown commands
	link-timers.ts       LinkTimers: the enquire_link heartbeat and the idle timeout
	log.ts               SmppLog, the logger contract, and silentLog — the default
	message.ts           Encoding detection, splitting, bit counting, SMPP date formatting
	pdu.ts               pduToObj / objToPdu / pduReturn — synchronous, result-returning
	pdu-framer.ts        PduFramer: a byte stream cut into complete PDUs
	pending-requests.ts  PendingRequests: sequence numbers, correlation, timeout, abort
	reassembly.ts        Reassembler: capped, expiring multipart groups
	reconnect-loop.ts    ReconnectLoop: backoff, retry timer, stopped-ness
	result.ts            Result<T> — the shape every fallible call returns
	send-sms.ts          submitSms composition and the submitSmParams builder
	send-window.ts       SendWindow: the maxOutstanding semaphore
	session-options.ts   SessionOptions, ReconnectOptions, bind direction and the session defaults
	udh.ts               User data header: the concatenation fields of a long SMS
	uuid.ts              uuidv7() — the ids the library generates for messages
	defs/
		commands.ts      The 33 commands, their ids and ordered parameter lists
		constants.ts     consts + constsById, and the SMPP version constants
		encodings.ts     GSM 03.38, LATIN1, UCS2, detection, data_coding resolution
		errors.ts        errors + errorsById (ESME_*)
		tlvs.ts          TLV definitions, tlvsById
		types.ts         Wire types: int8/int16/int32/string/cstring/buffer/arrays

Dependency direction is one way: defs knows nothing above it, pdu uses defs, session uses pdu, and client/server use session. Nothing reaches back up.

Parameter order is wire order. The key order inside cmds.*.params is the order the fields are written to and read from the buffer. Never sort those alphabetically — the alphabetical-ordering convention applies everywhere else, but here it corrupts every PDU.

Toolchain

Run everything through the container; never invoke node or npm on the host.

docker compose run --rm node npm install
docker compose run --rm node npm test
docker compose run --rm node npm run build

Tests are .ts and run directly under Node's type stripping — no build step in the dev loop.
Source imports use .ts extensions; rewriteRelativeImportExtensions emits .js into dist.
erasableSyntaxOnly is on, so no enums, no namespaces, no parameter properties. Use as const objects plus union types.
The published floor is Node 18, but the dev container runs Node 24 (type stripping needs it). CI compiles the tests and runs them on 18/20/22/24, so the floor is verified rather than asserted.
typescript is pinned to the 6.x line because typescript-eslint peer-requires <6.1.0. Move to TypeScript 7 once that constraint lifts.

Defects found in 0.4.0

Confirmed by reading the 0.4.0 source. The rewrite fixes all of them; each needs a regression test naming the behaviour, and the wire-affecting ones are cross-checked against a reference implementation (see todo.md).

Defect	0.4.0 behaviour
LATIN1 never decodes	`decodeMsg` loops `consts.ENCODING` without breaking, so `data_coding` 0x03 lands on the alias `ISO_8859_1`, which has no decoder, and silently falls back to ASCII
Short segments	`splitMsg` accumulates a full segment then pushes `msgPart.slice(0, -1)`, so every segment is one character short: 152 GSM characters instead of 153, 66 UCS2 instead of 67. Long messages are split into more segments than they need, and each extra segment is billed
DLR month off by one	`smppDate()` uses `getMonth()` (0-based) without `+1`, so January renders as `00`
Non-standard DLR status	Receipts emit `stat:UNDELIVERABLE`; the spec's field is 7 characters (`UNDELIV`)
Flash destroys UCS2	`flash: true` overwrites `data_coding` with 0x10, discarding the UCS2 alphabet, which needs 0x18
Shared concat reference	The concatenation reference counter is a module-level global shared by every session in the process
`send()` never times out	Each call adds a listener keyed on the sequence number; a peer that never answers leaks it and the promise never settles
`tls: true` is not TLS	Constructs a bare `new tls.Socket()` with no handshake instead of `tls.connect()`
Alphanumeric sender TON	`sendSms` hardcodes `source_addr_ton` to 1 (international) even for alphanumeric senders, which require TON 5
Text-only DLRs refused	`deliver_sm` without both `message_state` and `receipted_message_id` TLVs is rejected with `ESME_RINVTLVSTREAM`, so Kannel-style receipts are unusable
Unbounded reassembly	Incomplete long-SMS groups are capped by nothing and swept only when other traffic arrives, after 24 hours
Dead DLR aggregation	`longSmsDlrs` is allocated to merge per-segment receipts and then never used
Trailing NULL truncation	`types.buffer.size()` subtracts one whenever the value's last octet is `0x00`, so the PDU is allocated one octet short while `sm_length` still reports the full length. Any UCS2 message ending in a character like U+4E00 or U+3000 goes out corrupt
Dormant filters	`defs.filters` is declared on commands and TLVs but never invoked anywhere. Dropped in the rewrite; SMPP time formatting is exported as `smppTime` instead
Unchecked reads	Wire reads index straight into the buffer, so a short or malformed PDU throws out of the codec. Reads are bounds-checked and return results now
Unrangechecked writes	Integer params are handed to `writeUInt8`/`writeUInt16BE` unvalidated, so an out-of-range value throws from inside Node
`submit_multi` missing `sm_length`	The field is commented out of the command table, so `short_message` never round-trips for that command
Per-parameter defaults never applied	`calcCmdLength` reads `paramType.default` (the wire type's) rather than the parameter's, so `interface_version: 0x50` on the bind commands did nothing and every bind declared version 0x00
`source_telematics_id` width	Defined as a 2-octet integer; SMPP 3.4 5.3.2.8 makes it 1 octet, unlike `dest_telematics_id`, which really is 2
Binary payloads decoded as text	`data_coding` 0x02, 0x04, 0x14 and 0xF4-0xF7 are 8-bit binary and land on the GSM 03.38 table, which rewrites every octet outside it. They resolve to LATIN1 now, so the payload survives as bytes
`ESME_RINVBCASTCHANIND` typo	Defined as `0x011`, three hex digits; the spec value is `0x0112`

Multipart sends and the send window

sendSms puts every segment of a message on the wire together instead of waiting for each response in turn. This is not an optimisation: this library's own server holds segments until the whole message is reassembled before it answers any of them, so sending them one-after-a-response deadlocks. It follows that a message with more segments than maxOutstanding cannot be delivered to a server that defers responses that way — real SMSCs answer each submit_sm immediately, so this only bites when both ends are this library.

GSM 7-bit is sent unpacked

Over SMPP the ESME puts one GSM character per octet in short_message and the SMSC packs it into septets. The 140-octet limit applies to that packed result, not to what goes on the wire here, which is why a concatenated segment is 153 characters plus a 6-octet UDH — 159 octets in short_message, and entirely correct. Do not "fix" this to 134; that number is the packed payload size and would truncate every long message by a fifth.

UCS2 is not packed, so there the two coincide: 67 characters = 134 octets, plus the 6-octet UDH is exactly 140.

Conventions

Hard tabs. Alphabetical ordering for keys, imports and lists unless order is logic-significant. Two deliberate exceptions: command parameters are in wire order (above), and the errors and TLV tables are ordered by their numeric id so they can be diffed against the spec and gaps stay visible.
Comments are the exception, not the default — see the root CLAUDE.md rules. Do not write file preambles or restate what the code says.
Test data uses real randomised UUID v7 values, never aaaa-0000 placeholders.
message_id values the library generates are UUID v7.
A test that needs a dummy peer must resume() its sockets. An unread socket never processes the peer's FIN, so server.close() hangs forever — that is a test bug, not a library one.
Everything a test opens gets its teardown registered as it is opened, never closed on the test's last line: an assertion that throws skips that line, and the listener it leaves behind keeps node --test alive until CI's ten-minute cap. test/teardown.ts covers a session, a server and a listener; anything else takes a bare t.after. Its close aborts rather than drains, so a test that fails holding the send window still ends.
t.after hooks run in registration order, so registering at creation tears the outermost resource down first. A teardown that waits on a listener must destroy that listener's own connections before it waits, or be registered after the hook that does — net.Server.close() does not call back until every connection on it is gone.
assert.equal from node:assert/strict narrows its first argument, so a following ?. on the same value is flagged as unnecessary. Assert once with assert.ok(x) and use plain access after.

Decisions

A close arriving after our own unbind is a clean unbind, not an error. Maintainer's call, 2026-08-26: most SMSCs drop the socket instead of answering, so the documented shutdown would otherwise always report a failure. It does mask a socket that died mid-unbind for an unrelated reason, which is accepted — the peer sees the same TCP close either way.
The published surface is frozen at what src/index.ts exports today. Session is exported and publicly constructible, which is why SessionOptions and ReconnectOptions are public too — that is correct, not a leak, and it has been raised twice. The collaborators session.ts delegates to (Reassembler, PendingRequests, SendWindow, ReconnectLoop, LinkTimers, DlrMerger, submitSms) stay unpublished so they can be reshaped.
The sub-3.4 optional-parameter rule is a predicate, not a chokepoint. acceptsOptionalParams() is consulted by the library's own senders; session.send({ tlvs }) is passed through as written, because silently stripping a caller's explicit TLVs off a deliberately public low-level surface would be worse than sending them. The guarantee is "what this library sends honours the rule", never "the session cannot send optional parameters to an old peer".
Both ends feed peerInterfaceVersion, and a peer that declared nothing is pre-3.4. acceptBind() records what the ESME declared in its bind request; the client's bind() records the sc_interface_version the SMSC answered with. A peer that declared no version is recorded as undeclaredInterfaceVersion (0x00) and is sent no optional parameters — the spec reads an absent sc_interface_version as an SMSC that supports none. undefined is left to mean one thing only: no bind has been accepted on this session yet.
The library speaks SMPP 3.4 on the wire, and defs/ keeps the 5.0 tables as a superset. Maintainer's call, 2026-08-26: 3.4 is what SMSCs actually run, while the wider tables let the codec parse and build whatever a peer sends. The declared version is an option on both client() and server(), so an implementation that needs 5.0 throughout can have it. The threshold at or above which a peer may be sent optional parameters is fixed at 0x34 by the spec and is not the same constant as the declared version.
Bind direction is enforced on the library's own senders and on everything incoming, not on send(). A receiver-bound ESME carries no submit_sm and a transmitter-bound one no deliver_sm; sendSms() and sendDlr() refuse locally, and an arriving PDU is answered ESME_RINVBNDSTS. bindAllows() is a predicate on the same footing as acceptsOptionalParams(), so the deliberately public low-level send() stays a passthrough. Only those two commands are policed, because they are the only ones the library sends and dispatches by direction.
The logger is a five-method contract this library declares, not a dependency. SmppLog in log.ts is what the code actually calls (debug, error, info, verbose, warn), so an application can satisfy it with an object literal and @larvit/smpp ships with no runtime dependencies. @larvit/log implements it structurally and stays a devDependency, where test/tls.test.ts passing a real Log as the server's logger keeps that compatibility compiled.
esm_class decides what a deliver_sm is, and the body is only read when it names nothing. Message type MC_DELIVERY_RECEIPT (0x04) makes it a receipt whatever the body parses to, so a receipt in a format dlrFromPdu() cannot read reaches dlr with smsId undefined instead of arriving as an inbound SMS. Any other named type — delivery or user acknowledgement, conversation abort, intermediate notification — is not a receipt and its body is not scraped. A message type of 0, or one of the ten the spec reserves, keeps the scrape: SMSCs that send text-only receipts leave esm_class at 0, and reading that as the spec's "default message type" would lose every one of them. A non-empty receipted_message_id TLV marks a receipt on the same footing there, since nothing but a receipt carries one. What gets scraped is the decoded short_message with any UDH stripped; a receipt this library recognises never reaches the reassembler, so an SMSC that splits one across segments gets a dlr per segment rather than one merged report. The message_state TLV is authoritative only where it names a state in the table — SMPP reserves 0x80-0xFF for MC-vendor-specific values, so an unnameable one keeps its raw statusId and leaves statusMsg to the body.
A listener that rejects is routed by Node's captureRejections, not by hand-dispatching. Both emitters construct with captureRejections: true and implement [EventEmitter.captureRejectionSymbol], which lands a rejected async listener on sessionError or serverError beside the synchronous guard in emit(). Dispatching rawListeners() from emit() instead needs a cast to call them with the event's argument tuple, which hard rule 4 forbids. A rejection reason is unknown and String() throws on a null-prototype object, so both handlers normalise through errorFrom() rather than inline — a route out of the handler would land on a bare process.nextTick with nothing to catch it.
Both emitters re-declare their listener methods to accept a promise. Maintainer's call, 2026-08-27: EventEmitter types every listener as void-returning, so the session.on('sms', async sms => …) the README documents reads as a misused promise in any strict consumer. declare on: … and its six siblings re-type the inherited methods to return unknown, which emits nothing, needs no cast and leaves the runtime method on the prototype. Overriding them as real methods instead cannot work: the super.on() call needs a cast to satisfy the conditional Listener type. The cost is that a subclass can no longer reach those seven through super or override them as methods — re-declaring them the same way is its way out. unknown rather than void | Promise<void> because a listener may return anything — session.on('close', () => set.delete(session)) returns a boolean.
A reconnect keeps the delivery-receipt merges; everything else the link held is dropped. onDeliverSm() answers each receipt before the group it belongs to is complete, and teardown() runs on every path — an idle timeout and a failed rebind, not only close() — so clearing the merges there loses receipts no peer has a reason to send again. They are cleared where the session is over instead: close(), or a drop with no reconnect loop left to bring the link back. Inbound segments stay in teardown(), because they go unanswered until the message is whole: the peer still holds them, and answering it on a later link with the old segments' sequence numbers would correlate with nothing. Surviving a process restart is a separate, public-surface question, and is in todo.md.
A message id base is merged at most once. A receipt carries nothing but <base>-<n>, so a straggler for a message whose group is gone cannot be told from a receipt for a later message the peer handed the same ids — an SMSC whose id counter restarts with its process is the realistic case. DlrMerger remembers the bases it has finished with, capped and expiring exactly like the groups, and refuses to open one a second time: the later message gets no messageDlr, and an earlier one whose receipts are still arriving is dropped rather than left to collect the later one's. Every segment still reaches the application as a dlr. The rule covers the bases the merger opened — expect() ignores a lone id, so a single-part message never claims one.
A deliberate shutdown drains; an unusable link and an abort do not. close() and unbind() refuse further sends and wait on the send window, not the pending map — the map misses a segment still queued behind a full window, and finishing a half-sent multipart message is the point. The window counts slots, never outcomes, so it says when to stop waiting and nothing about what happened: it empties on a drop too, where teardown() settles everything the link was carrying, which is why drain() reads closed before it reads the count. The wait covers what this end sent — a request the peer sent us is answered through sendReturn(), which never enters the window, so a server session waits for none of its inbound work; that half is in todo.md. shutdownTimeout bounds the drain alone, and 0 waits forever like every other timeout here; unbind() then waits responseTimeout for its own response, and sends that PDU through request() past both the window and the drain gate because it must go out either way. A stream the framer or the codec cannot read takes end() instead, and so does close({ signal }) on an aborted signal and a peer's own unbind — nothing on a dead link can answer, an abort means stop now, and a peer that has declared itself finished will not answer what it still owes us, so draining any of the three would only hold a socket open for the timeout. SmppServer.close() reports each session's unfinished drain through serverError, because its own result says nothing but that the listener stopped. shutdownTimeout stays a session option rather than a close() argument: server() builds sessions on the caller's behalf, so the option is the only composition point, and close({ signal }) already covers a hard deadline.
The TLS tests build their own self-signed certificate in DER (test/tls.test.ts) instead of adding a devDependency or shelling out to openssl. Maintainer's call, 2026-08-26: the dev image node:24.18.0-bookworm-slim ships no openssl binary, so a shelled-out fixture would pass in CI and fail on every developer machine, and a committed key leaks in a public repository. Valid while the dev image has no openssl.

22 KiB Raw Blame History