Native WSP POST Research Notes

Date: 2026-03-07

Scope:

connectionless WSP POST behavior for native desktop/Kannel interop
WML go/postfield submission semantics relevant to T0-30
implementation guidance for the current pre-alpha MVP slice

Why this note exists

The native desktop GET lane is already working against Kannel. The next feature, native form POST, exposed a wire-level interop gap. This note records the relevant source findings so the implementation is driven by spec text and observed gateway behavior instead of trial-and-error packet layout changes.

Primary source findings

1. Connectionless WSP `Post` PDU layout

Canonical source:

OMA-WAP-TS-WSP-V1_0-20020920-C.cleaned.md

The key connectionless Post layout is:

TID | PDU | UriLen | HeadersLen | Uri | ContentType | Headers | Data

Important details from the source text:

UriLen is a uintvar
HeadersLen is a uintvar
HeadersLen covers ContentType + Headers
Uri must not include a null terminator
Data starts immediately after the header block and runs to the end of the SDU

Why this matters:

earlier local encoder experiments assumed the wrong field ordering and boundary markers
Kannel failures that looked like malformed URLs or shifted bodies are consistent with a bad split between Uri, ContentType, Headers, and Data

2. Connectionless WSP `Get` PDU layout remains the control case

Canonical source:

OMA-WAP-TS-WSP-V1_0-20020920-C.cleaned.md

The Get layout is:

TID | PDU | URILen | URI | Headers

This is already proven live in the native desktop path and should remain the comparison baseline while fixing POST.

3. WML `go`/`postfield` submission behavior

Canonical sources:

Relevant rules:

substitute variables before submission
choose submission charset from accept-charset, or fall back to document encoding
preserve postfield order
URI-escape names and values for application/x-www-form-urlencoded
for method="post", send the payload in the HTTP POST body
Content-Type must include a charset parameter when required by the submitted content
same-document/same-deck navigation must ignore postfield unless cache-control="no-cache"

For this MVP slice, the most important rules are:

field ordering
variable substitution timing
application/x-www-form-urlencoded body generation
same-deck suppression semantics

4. WMLScript is not the transport pivot here

Relevant source set:

Conclusion:

WMLScript may influence form values indirectly through variables
the actual submission semantics still live in WML task execution plus transport encoding
T0-30 should stay focused on go/postfield plumbing and WSP request shaping, not WMLScript-specific transport behavior

Local codebase findings

Engine/runtime status

Relevant path:

navigation.rs

Current state:

the engine now preserves method="post" and postfield metadata
it serializes the body as application/x-www-form-urlencoded
it currently emits postContext with content type application/x-www-form-urlencoded

Gap still to keep in mind:

charset handling is not yet surfaced in emitted content type

Transport/runtime status

Relevant paths:

Current state:

transport selection already supports native handling for wap:// GET
current POST work is additive and uses the same stable entrypoint
the live blocker is wire framing, not the high-level request-policy contract

Kannel environment assumptions

Relevant path:

kannel.conf

Current lab mapping:

gateway HTTP base resolves through Kannel on localhost:13002
Kannel rewrites gateway-facing URLs to http://wml-server:3000/*
the desktop native lane should continue targeting this stack via WSP/WDP, not by adding more direct HTTP shortcuts

Observed gateway behavior

Observed from local Kannel logs during native POST debugging:

Kannel recognized the request as a connectionless Post attempt
malformed requests caused URI/header/body boundary corruption
symptoms included:
- malformed target URI
- truncated or shifted Content-Type
- request body bytes being interpreted as part of the path

Interpretation:

Kannel behavior strongly supports the spec reading that the problem is PDU layout, especially the ordering and sizing of UriLen, HeadersLen, and the ContentType block

External source cross-checks

Useful external references consulted:

What these contributed:

Kannel references confirmed the gateway role and expected HTTP-bridging behavior after WSP decode
Wireshark references are useful for protocol naming and packet sanity checks
the OMA corpus remains the authoritative source for exact PDU field order

Limitations of this pass:

exact Kannel and Wireshark source files were not reliably retrievable through the web tool in this pass
implementation guidance therefore relies on:
- canonical local OMA/WAP specs in the repo
- local repo code
- live Kannel behavior already observed in the dev environment

Implementation guidance for `T0-30`

Immediate next step

Fix the connectionless POST encoder to follow:

TID | PDU | UriLen | HeadersLen | Uri | ContentType | Headers | Data

before changing broader transport behavior.

Scope to keep

application/x-www-form-urlencoded only
ASCII-safe lab payloads first
login/register training-environment flows only
native connectionless mode only

Scope to defer

multipart/form-data
connection-oriented WSP session work
broader WTP session/client state
browser-side protocol logic
non-ASCII charset edge cases beyond the first explicit charset plumbing pass

Practical implementation order

Correct the wire layout for native connectionless POST.
Re-run transport-level ignored Kannel smoke until register/login succeed.
Add charset-aware Content-Type plumbing once the ASCII baseline is stable.
Expand host/browser smokes on top of the validated transport path.

Decisions reinforced by this research

do not add another HTTP workaround for desktop wap:// navigation
keep native POST additive under the existing transport entrypoint
treat Kannel interop logs as a validation aid, not as the source of truth over the OMA/WAP specs
keep T0-30 narrowly focused on form submission, not full session architecture