Repository navigation
Conversation
The bundler gives each output its type. For an output that is in memory, set_blob_mime stored the type but did not mark it as set, so every reader that sends only a set type dropped it: a Response route, a fetch() upload, and the stream of the output. An output written with outdir is a file, and a file's type already counts as set.
|
Updated 9:02 AM PT - Oct 31st, 2026
❌ @robobun, your commit 79d3949 has 1 failures in
🧪 To try this PR locally: bunx bun-pr 47547That installs a local version of the PR into your bun-47547 --bun |
|
Caution Review failedAn error occurred during the review process. Please try again later. No actionable comments were generated in the recent review.
 |
| Layer / File(s) | Summary |
|---|---|
MIME flag and propagation coverage src/runtime/api/output_file_jsc.rs, test/bundler/bun-build-api.test.ts |
set_blob_mime marks the content type as explicitly set. Tests check JavaScript and JSON content types and verify that an explicit text/plain request header takes precedence. |
Suggested reviewers: jarred-sumner, alii
Merge Risk: 
79d39
-
No actionable merge-blocking risk is identified in the MIME propagation change; merge after normal checks pass.
Pre-merge checks: 
-
Check name Status Explanation Title check The title clearly identifies the main change: marking in-memory build output MIME types as set so routes and uploads send them. Description check The description explains the problem, fix, behavior changes, downsides, and verification results. It does not use the template headings exactly, but it provides the required information in equivalent … Linked Issues check Check skipped because no linked issues were found for this pull request. Out of Scope Changes check Check skipped because no linked issues were found for this pull request. Full details: Description check
-
Explanation
The description explains the problem, fix, behavior changes, downsides, and verification results. It does not use the template headings exactly, but it provides the required information in equivalent sections.
-
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
LGTM — a one-line flag set that brings in-memory build artifacts in line with file-backed ones and new Blob(bytes, { type }).
What was reviewed:
set_blob_mimenow mirrorsstandalone_graph_jsc.rs:52; the flag is consumed byhas_content_type_from_user(fetch uploads),Body.rs(.blob()fallback to text) andByteBlobLoader::setup(stream()), andBlob::slicepropagates it, so all nine assertions in the new test exercise the change (eight fail on base,uploadWithHeaderis the negative contract).- Checked the
file-loader / unknown-extension case:to_mime_typereturnsapplication/octet-stream, which now gets sent — same as theoutdirpath already does, so not a regression. - HTML-route caller (
HTMLBundle.rs:523) readscontent_type_or_mime_type(), which ignores the flag — unaffected.
Extended reasoning...
The diff adds blob.content_type_was_set.set(true) in the shared set_blob_mime helper in src/runtime/api/output_file_jsc.rs (used by both to_js and to_blob for the in-memory Buffer variant) and one hermetic test in test/bundler/bun-build-api.test.ts using port: 0 and await using. It touches no security-sensitive surface; it only affects which Content-Type header is emitted for in-memory Bun.build outputs on routes, streams and fetch uploads. Approve was decided by the change being a single line that copies an existing sibling pattern, the test covering every consumer of the flag plus the explicit-header precedence case, and no CODEOWNERS entry or outstanding reviewer objection on the changed files.
|
CI build 136099 is complete: 183 of 188 jobs passed. The 5 red jobs have one cause. |
Related to #3011
Problem
routes: { "/app.js": new Response(output) }sends noContent-Typefor an in-memoryBun.build()output (nooutdir).response.headersreports the type.set_blob_mime(src/runtime/api/output_file_jsc.rs:23) stores the type and does not setcontent_type_was_set. Readers that send only a set type drop it.Fix
set_blob_mimesetscontent_type_was_set, as embedded files do (standalone_graph_jsc.rs:52).new Blob(bytes, { type }). Of 22 readers compared on both builds, 6 change (Notes).test/bundler/bun-build-api.test.ts(the new test fails on main in 8 values). Also the route suites,blob,body,response.StaticRoute::from_js. The review asked for the producer and a declared upload change. Both are here.Background
BuildArtifactwraps aBlob: a file withoutdir, bytes without it.fetch()request andblob.stream()send only a set type.StaticRoute::from_js. It repairs one reader of ten, and http: consolidate the open fetch, Bun.serve, node:http, node:http2 and HTTP/3 pull requests #46528 rewrites that function.Downsides
fetch(url, { body: output })now sends theContent-Typeof the output, as anoutdiroutput does.output.stream()as a body sends it too. Both sent none. A caller that needs noContent-Typemust set its own.new Response(output.stream())sends that type, notapplication/octet-stream..text+256 bytes, same file size (GNUsize).Notes
What changes.
outputis the entry point of an in-memory build, a slice of it, or its source map. Debug builds of main 34cbc70 and of this branch, one script, 22 readers for each of 6 bodies. Every row that is not listed is the same on both builds.new Blob(bytes, { type })outdiroutputnew Response(output):Content-Typenew Response(output.stream()):Content-Typeapplication/octet-streamfetch(url, { body: output })sendsfetch(new Request(url, { body: output }))sendsfetch(url, { body: output.stream() })sends(await new Response(output.stream()).blob()).typetext/plain;charset=utf-8The same on both builds:
output.type,output.slice().type,new Response(output).headers, a handler'snew Response(output)for GET and HEAD,new Request(url, { body: output }).headers, a request with its ownContent-Type,(await new Response(output).blob()).type,Bun.readableStreamToBlob(output.stream()),new Blob([output]),new File([output], name),fetch(URL.createObjectURL(output)).Each kind of output. Release builds of main 34cbc70 and of this branch, one build with an entry point, a CSS asset, a source map, a
.pngasset and an asset whose extension has no known type (its type isapplication/octet-stream). On this branch a route, a handler and an upload send the same type for the in-memory output as for theoutdiroutput, for every kind. On main the route and the upload of the in-memory output send none.A deleted header. A route sends the type of an output also after
response.headers.delete("content-type"). That is main's rule for a set type: a route does so for a typed Blob on main, and afetchhandler does so for an output on main. AContent-Typethat the caller gives to theResponsereplaces the type, as before.The second caller.
set_blob_mimealso types the Blobs of HTML routes (OutputFile::to_blob,src/runtime/server/HTMLBundle.rs:519). Those routes write theirContent-Typefromcontent_type_or_mime_type(), which does not read the flag.bun-serve-html.test.ts,bun-serve-html-entry.test.tsandbun-serve-html-manifest.test.tspass.Other producers.
blob.slice(start, end, type), the body of adata:URL and a multipart part typed by its file name lose their type on a route in the same way (measured on 1.4.3). They are other producers, and this PR does not change them. #46516 and #47335 are open forslice().Overlap. #46114 removes the
store.mime_typewrite fromset_blob_mime, one line below this one. #46528 does not touch this file.The ETag of #3011 is not in this PR. It needs a maintainer decision. The issue has the details.
Suites on the debug ASAN build.
bun-build-api(92 pass, 1 skip, 1 todo),bun-serve-static,bun-serve-html,bun-serve-html-entry,bun-serve-html-manifest,bun-serve-html-405,bun-serve-routes,bun-serve-file,serve-if-none-match,blob,body,response,stream-fast-path,bun-cryptohasher, regression10004.Tools. The container has no
bloaty. The size is from GNUsize.