[atharrva01]

Summary

Closes out three gaps flagged in issue #66's deliverables: tokengen documentation, a real bug in the e2e harness that was masking failures on every platform, and PLATFORM=xdev's e2e tests — which now pass end to end.

• Tokengen docs: a new "Public parameters (tokengen)" section under Development — what go tool tokengen generates and why every TMS (and the fabric3 chaincode image) needs to agree on it byte-for-byte, that make setup already runs it automatically, how to inspect an existing parameters file, and when to regenerate by hand. Consolidates the one existing passing mention into a link to this section.
• curl_with_retry never actually retried a failure: if CMD; then return 0; fi followed by local exit_code=$? always captured 0, not CMD's real exit code — bash resets $? to 0 for an if whose condition is false when no branch executes. Confirmed with an isolated repro before trusting it, then again against the real function with a stubbed curl. This means curl_with_retry has silently returned success after exhausting every retry since it was written, so every CI run using it (every platform) hasn't actually been asserting what it looks like it asserts. Fixed by capturing the exit code inside an explicit else, where $? still reflects the failed command.
• Now that a fabricx CI failure would actually be visible, HTLC_TEST=1 is turned on for the fabricx matrix leg — the real, honest verification for whether HTLC works on Fabric-X, since that network can't be fully reproduced locally (ansible-provisioned infrastructure) the way fabric3 can.

PLATFORM=xdev: now passes end to end

Three config/version-drift bugs, fixed early on:

1. configtx.yaml set Application.MetaNamespaceVerificationKeyPath, a field fabric-x-common removed in Feb 2026 (commit d03614c, PR #74, resolving fabric-x-common issue #58) in anticipation of a LifecycleEndorsementPolicy replacement that was never actually built. Confirmed via GitHub code search: the field exists nowhere in fabric-x/fabric-x-common, released or on main. configtxgen no longer errors with it removed; there's no replacement key to add.
2. fxconfig's CLI changed from flags (--channel, --orderer, --mspID, --mspConfigPath, --pk) to a config file / FXCONFIG_* env vars entirely — the old flags don't exist anymore.
3. compose-xdev.yml pinned the test-committer image at 0.1.7 while the Go module graph is already on fabric-x-committer v1.0.4 — the same version-drift class as the chaincode Dockerfile's Panurus commit pin fixed in the v0.22.0 bump. Bumped to match.

That got configtxgen/fxconfig namespace create working, but the app itself still couldn't reach a running network — reported here as an unroot-caused "orderer connection times out." It turned out to be two unrelated bugs, not one, and not an upstream gap:

• The hand-rolled compose-xdev.yml test-committer had no TLS configured, while the committer image requires mTLS on committer-sidecar/committer-query-service and the mock orderer alike. Every connection was a plaintext client hitting a TLS listener.
• Once TLS was fixed, FSC's generic broadcaster (platform/fabric/core/generic/ordering/bft.go) refused to submit anything at all: it hard-requires a 4-orderer BFT quorum, which a single mock orderer can never provide.

I'd filed fabric-x-common#202 believing LifecycleEndorsementPolicy was a missing upstream feature. That was wrong, and I'm correcting it separately: it's an ordinary Signature policy, same as classic Fabric. devnet/configtx.yaml already sets it exactly this way, and a clean make -C devnet init-namespace commits and lists the namespace with no upstream changes needed. Both real bugs were local.

Since both bugs are TLS/quorum-shaped and devnet/ already solves them correctly, this drops the hand-rolled compose-xdev.yml/configtx.yaml/crypto-config.yaml entirely and delegates PLATFORM=xdev's network lifecycle to devnet (../devnet), used unmodified via its own documented integration pattern, running its real 4-party Arma orderer instead of a single mock orderer.

Two further bugs surfaced once the network itself worked:

• A genuine Panurus bug. endorser/service/fsc.go's Init() called DeployTMSs(), which (Panurus v0.18.0, token/services/network/fabricx/tms/deployer.go) fetches the public parameters it's about to deploy from the ledger itself — the same reader used to load an already-existing TMS. On a namespace that's never been initialized, that read returns nothing, and the deployer commits that nothing right back: a bootstrap that can never bootstrap. Confirmed directly against the committer (the setup key's hash matched sha256(""), not the real ~22KB parameters file), and confirmed no caller anywhere in Panurus's own code or tests exercises DeployTMSs/DeployTMS/DeployTMSWithPP, so this had never been exercised end to end before. Fixed by reading the parameters from the local file core.yaml already points at (token.tms.mytms.publicParameters.path) and calling DeployTMSWithPP directly.
• FSC-native endorsement identity binding. The same mechanism fabric3 already needed (BindEndorsingIdentities in common/fsc.go) was never wired up for xdev: issuer/owner1/owner2 need a local copy of each endorser's Fabric identity under a matching MSP label for a response signed by that identity to be recognized as coming from the expected P2P party. Added.

PLATFORM=xdev HTLC_TEST=1 ./scripts/test.sh now passes end to end: issue, balance, transfer, transaction history (correct Pending→Confirmed transitions), redeem, and the full HTLC suite (lock, claim, timeout, reclaim, rejected-claim-after-reclaim) — the same assertions fabric3 already passes, all green.

Test plan

• go build/go vet clean on both the default and fabricx build tags
• curl_with_retry's fix verified both in an isolated repro and against the real function with a stubbed curl (returns the real failure code instead of silently succeeding)
• Full clean PLATFORM=fabric3 ./scripts/test.sh passes end to end, including the HTLC block (regression check — nothing above touches fabric3's own path)
• Full clean PLATFORM=xdev HTLC_TEST=1 ./scripts/test.sh passes end to end: issue/transfer/redeem/transaction-history/HTLC lock/claim/timeout/reclaim, every assertion green, reproduced from a clean make teardown clean + fresh make setup start

[atharrva01]

Splitting this into two focused PRs — it bundled three unrelated changes (docs, a test-harness bug fix, and the xdev devnet migration), which made it harder to review and meant the smaller/safer pieces were waiting on the bigger one.

• docs(tokens): tokengen docs, fix curl_with_retry, enable HTLC on fabricx CI #38 — tokengen docs + the curl_with_retry fix + enabling HTLC_TEST=1 on fabricx CI
• feat(tokens): run PLATFORM=xdev against devnet's full Fabric-X stack #39 — the PLATFORM=xdev devnet migration

Closing this one in favor of those two.