uws: route every HTTP/1 header-section termination through terminateHeaders()

[robobun]

Part of #43853

Behaviour change: none

Problem

• Eleven places in packages/bun-uws/src/HttpResponse.h end the HTTP/1 header section with their own writeMark() + "\r\n" block, and uws_res_prepare_for_sendfile (src/uws_sys/libuwsockets.cpp) has one more. Five of them also copy the Transfer-Encoding: chunked line.
• Only internalEnd can write a Connection: close line. The fix for node:http default agent reuses a socket after a Connection: close request in 1.4.2 (every second request ECONNRESET/EPIPE) #43853 needs that decision on every path, so it needs one place where every path ends its headers.

Fix

• Add HttpResponse::writeOwnedHeaders(chunked): Date (unless the caller wrote one) and Transfer-Encoding: chunked for a chunked body (unless the caller wrote one). terminateHeaders(chunked) adds the blank line. Every block in HttpResponse.h and uws_res_prepare_for_sendfile now call it. The Content-Length path of internalEnd calls writeOwnedHeaders and keeps the blank line in the Content-Length write.
• uws_res_end_without_body stays as it is: it writes no Date today, so routing it through would change the wire. The next PR does that.
• Correct because the wire output is the same: a raw socket probe of 26 response paths, each with a keep-alive, a Connection: close and an HTTP/1.0 request, is byte-identical to a build of main (Date masked).
• Verified: test/js/bun/http/bun-serve-headers.test.ts, bun-serve-static, bun-serve-file, request-smuggling, serve-direct-readable-stream, serve.test.ts, bun-server.test.ts, test/js/node/http/, websocket-server.test.ts.

Background

• writeMark() writes the Date header once per response and records that in HTTP_WROTE_DATE_HEADER. The blocks this PR folds all called it right before the blank line.
• Release .text of the stack: 80,656,785 to 80,655,761 bytes (size, -1,024). The HttpResponse<true|false> and uws_res_* symbols involved: 12,809 to 12,223 bytes (nm -S).

Notes

• An uncorked internalEnd (a large in-memory body from an async handler, a Bun.file end from a read callback) issues the same send() calls as before: writeOwnedHeaders is writeMark(), and the Content-Length line plus the blank line stay one write.
• uws_res_prepare_for_sendfile used getSendBuffer(2) for the blank line. Super::write("\r\n", 2) does the same when corked. When not corked, both end in one send for the two bytes: getSendBuffer took a cork slot and uncork() flushed it.
• Static instruction counts on bun-profile (objdump): HttpResponse<false>::internalEnd 493 to 465, write 300 to 276, terminateHeaders 60 (new, shared).
• Probe paths: in-memory, HEAD, async handler, 204, 304, ReadableStream, Bun.file small and above 1 MiB (sendfile), explicit Transfer-Encoding, user Connection: close, user Date, thrown error page, 404, static route GET/HEAD/POST, static route with Connection: close, file route GET/HEAD, file route above 1 MiB, node:http server default, Connection: close, removeHeader("connection"), chunked, HEAD, 204. Each with a keep-alive, a Connection: close and an HTTP/1.0 request.
• Not routed through terminateHeaders(): writeContinue() (100 Continue) and the raw 1xx writer. They are informational responses with no header section of their own.
• Open PR Bun.serve: send Connection: keep-alive and Keep-Alive: timeout=N by default #43850 edits the same blocks. After this PR its keep-alive header belongs in terminateHeaders().
• websocket-server.test.ts has 4 tests that time out in this container on main too (send() > Buffer (utf-8) and siblings). serve.test.ts has 2 that need a non-root user or a non-loopback interface.

---

no test proof · iteration 1 · the description declares no behaviour change, so there is no failing test to prove; the existing suite in CI is the check

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: oven-sh/bun/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Essentials

Run ID: 1cb8fb5e-8815-4a69-aa3d-5ae49753938e

📥 Commits

Reviewing files that changed from the base of the PR and between 81f8803 and c5b9d81.

📒 Files selected for processing (1)

• packages/bun-uws/src/HttpResponse.h

Included review availability: Your plan provides up to 10 included reviews per hour; 0 remain after this review.

---

Walkthrough

HttpResponse adds shared methods to write owned headers and terminate response headers. Response-ending, streaming, upgrade, and header-flush paths use these methods. Sendfile preparation also calls terminateHeaders for SSL and non-SSL responses.

Changes

HTTP response header termination

Layer / File(s)

Summary

Shared header termination packages/bun-uws/src/HttpResponse.h, src/uws_sys/libuwsockets.cpp

writeOwnedHeaders writes the Date header and, for chunked responses, adds Transfer-Encoding: chunked if the caller has not supplied it. terminateHeaders writes the owned headers and ends the header section. Response paths use these helpers; the Content-Length path writes its header line and the blank line separately. Sendfile preparation calls terminateHeaders in both SSL and non-SSL branches, which continue to uncork the response.

Suggested reviewers: jarred-sumner

Priority: ⬇️ Low

Merge Risk: ⚪ Minimal · up to c5b9d

Header termination preserves the inspected HTTP response output; no issue identified here prevents merging after normal checks.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly identifies the main change: routing HTTP/1 header-section termination through terminateHeaders().
Description check	✅ Passed	The description explains the problem, implementation, unchanged behavior, verification steps, and relevant exceptions. It provides the information required by the repository template, although it uses…

---

_{Comment @coderabbitai help to get the list of available commands.}

[claude]

Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.

Beyond the inline finding, I also checked the equivalence risks of the refactor: the terminateHeaders(true) call at HttpResponse.h:611 sits under an HTTP_WROTE_TRANSFER_ENCODING_HEADER guard, so it cannot emit a second Transfer-Encoding line; the second writeMark() inside terminateHeaders() on the Content-Length path is a no-op via HTTP_WROTE_DATE_HEADER, so Date still precedes Content-Length; and uws_res_prepare_for_sendfile runs inside the corker callback, so Super::write("\r\n") lands in the same cork buffer getSendBuffer(2) used.

Extended reasoning...

The change is a refactor of the vendored uWebSockets HttpResponse.h and the uws_res_prepare_for_sendfile C shim, consolidating eleven hand-inlined Date/Transfer-Encoding/blank-line sequences into one terminateHeaders(bool chunked) helper, with no tests added. It touches HTTP/1 response framing (which headers are emitted and in what order) but no auth, crypto, or input-parsing surface. Each replaced site was compared against its old sequence and the chunked/non-chunked argument matches what the old code wrote at every site; the one inline finding is an extra send() on the uncorked Content-Length end path rather than a wire-format difference.

[claude]

Code review found no issues

No high-confidence issues detected in this change.