fix(networking): recover shared relay clients and preserve write order

Open

Salazareowants to merge intofrom

Distinguish relay failures from individual stream errors, discard failed shared clients with transport cleanup, and flush queued socket writes before open callbacks.

Relay runtime error classification needs validation against the deployed runtime.

Validation on this isolated branch: 112 sdk tests across 4 suites. git diff --check passed.

Activity

fix(networking): recover shared relay clients and preserve write order

f04c1b5Oct 27

github-actions[bot]3d ago

Coverage Report for puter.js SDK

Status	Category	Percentage	Covered / Total
🔵	Lines	59.26% ⬇️ -0.22 pp	5266 / 8885
🔵	Statements	57.38% ⬇️ -0.21 pp	5607 / 9771
🔵	Functions	58.83% ⬇️ -0.50 pp	972 / 1652
🔵	Branches	52.46% ⬇️ -0.23 pp	3549 / 6764

File Coverage

File	Stmts	Branches	Functions	Lines	Uncovered Lines
Changed Files
src/puter-js/src/modules/networking/PSocket.js	35.65% ⬇️ -0.48 pp	30.12% ⬇️ -1.38 pp	66.66% 🟰 ±0 pp	36.22% ⬇️ -0.22 pp	8, 12, 16, 29, 115, 131-155, 165-166, 179, 184-286, 301-326, 331, 341
src/puter-js/src/modules/networking/epoxy.js	4.1% ⬇️ -3.40 pp	0% 🟰 ±0 pp	0% 🟰 ±0 pp	4.16% ⬇️ -3.34 pp	7-137, 143-154
src/puter-js/src/modules/networking/index.js	57.97% ⬇️ -14.94 pp	41.02% ⬇️ -8.98 pp	72.72% ⬇️ -7.28 pp	57.97% ⬇️ -14.94 pp	22, 46, 65-69, 73-75, 83-88, 97, 105-107, 126, 144-163, 179
src/puter-js/src/modules/networking/requests.js	72.5% ⬆️ +2.74 pp	62.5% ⬆️ +0.74 pp	100% 🟰 ±0 pp	72.5% ⬆️ +2.74 pp	8-14, 31-34, 41, 79, 102-113

Generated in workflow #1671 for commit 053b1ca by the Vitest Coverage Report Action

r58Playzreviewed3d ago

r58Playz3d ago

from my codex: Found one issue in f04c1b5:

[P2] Writes can remain queued forever after the initial flush — PSocket.js:204–208. After the flush empties the queue, await yields before #flushing resets. A write arriving in that gap gets queued, but nothing drains it afterward. The socket emits open while the payload remains unsent and its callback never runs. Reset the flag synchronously with the final queue check.

other than that lgtm

fix(networking): clear the flush flag with the last queue check

053b1caOct 28