bake: remap server-side stack frames against the server chunk prefix

[robobun]

What does this PR do?

Stack traces for server-side code running in the bake dev server (framework mode, where route modules are evaluated on the server) point at the wrong line. With this route module:

// pages/error-page.tsx
export default function ErrorPage() {
  throwError();                        // line 2
  return <div />;
}

function throwError() {                // line 6
  throw new Error("HMR error test");   // line 7
}

a request prints

6 | function throwError() {
    ^
error: HMR error test
      at throwError (pages/error-page.tsx:6:1)
      at ErrorPage (pages/error-page.tsx:1:16)

and with this change it prints

7 |   throw new Error("HMR error test");
            ^
error: HMR error test
      at throwError (pages/error-page.tsx:7:9)
      at ErrorPage (pages/error-page.tsx:2:3)

test/bake/dev/server-sourcemap.test.ts had the wrong positions baked into its expectations (6:16, 6:1, 5:1, 1:28, 3:38, and the churn test's 6 + i), with a comment rationalizing them as "frames remap to the declaration position". This has been the behavior since server source maps were added; the client side is not affected.

Cause

The server HMR chunk and its source map disagree about what precedes the module code.

• IncrementalGraph::take_js_bundle_to_list_server emits a server HmrChunk as ({ + modules + }). The prefix contains no newline, so the first module starts on generated line 0.
• source_map_store::Entry::join_vlq ignored its side argument (let _ = side;) and, for every HmrChunk, started the mappings after the line count of the client prefix, self[Symbol.for("bun:hmr")]({\n, i.e. on generated line 1.

So the map for the server chunk is shifted down by one generated line relative to the code. A frame on generated line L is looked up at map line L, which holds the mappings recorded for generated line L - 1, so every frame reports the original position of the previous generated line. For ordinary code that is the line above the statement; the column is whichever mapping on that line happens to precede the frame's generated column (hence 6:16, the ( of function myFunc(, or 3:38, the { of the function declaration). In denser code it can be further off: a react-server-dom frame in the same chunk moved from 3971:15 (a try { line) to 3972:20, the actual Component(props, void 0) call.

The client chunks use the same function but with the client prefix, which is why test/bake/dev/sourcemap.test.ts (exact positions for the initial response and an HMR chunk) passes before and after.

Fix

ChunkKind::prefix(side) returns the code a chunk of that kind starts with on that side (the full HMR runtime for InitialResponse, the one-line client wrapper or the bare ({ for HmrChunk). Both take_js_bundle_to_list* and join_vlq take it from there, and HmrRuntime::from_static derives line_count from the literal, so the emitted code and the map cannot disagree again. join_vlq's side parameter is now used; render_json was already being called with Side::Server for server chunks.

Verification

test/bake/dev/server-sourcemap.test.ts now asserts the full remapped frames (at fn (.../file:line:col)) for the throw site and for each caller, across the initial bundle, an HMR update, a throw in an imported non-page module, and four successive reloads. Without the change in src/ all four tests fail with the positions shown above (6:16 / 1:1, 6:1 / 1:16, 5:1 / 1:28 / 3:38, 6 + i:1); with it they pass.

Also run: test/bake/dev/sourcemap.test.ts (client maps, unchanged behavior) and test/bake/dev/html.test.ts (the /_bun/report_error remapping path, which renders client maps through the same function); cargo clippy -p bun_runtime is clean.

Note on columns: in the first test the page component is async, so React reads error.stack before the error is printed, and that rendering currently places a construct frame at the callee (Error, 7:13) where the printer places it at the new keyword (7:9, what the other tests see). #37396 makes the two agree; once it lands, that one expectation becomes 7:9 (its current version of this file will conflict with this change either way).

While looking at this I also noticed that after any hot update, frames from modules loaded by an earlier server patch stop remapping (every patch is registered under bake://server.patch.js, so the newest patch's map replaces the older ones; visible as at react-stack-bottom-frame (bake://server.patch.js:3242:29) in the HMR test output). That is a separate problem and is not addressed here.

---

[review] gate passed · iteration 0 · 5 files touched

fails on main (without fix)

ASAN without fix: 4 FAILED
$ BUN_DEBUG_QUIET_LOGS=1 bun scripts/build.ts --profile=debug --quiet test "--reporter=junit" "--reporter-outfile=/tmp/mechgate.xml" test/bake/dev/server-sourcemap.test.ts
bun test v1.4.0 (178906144)

test/bake/dev/server-sourcemap.test.ts:
Dev server testing directory: /tmp/bun-dev-test-O6wHS9
bun add v1.4.0 (178906144)
Resolving dependencies
Resolved, downloaded and extracted [2]
Saved lockfile

installed react@0.0.0-experimental-603e6108-20241029
installed react-dom@0.0.0-experimental-603e6108-20241029
installed react-server-dom-bun@0.0.0-experimental-603e6108-20241029
installed react-refresh@0.0.0-experimental-603e6108-20241029

6 packages installed [244.00ms]
bun install v1.4.0 (178906144)

Checked 6 installs across 7 packages (no changes) [100.00ms]
�[0;30mdev|�[0m Started development server: http://localhost:39561
�[0;30mdev|�[0m �[32mBundled page in 1920ms�[0m�[2m:�[0m pages/[...slug].tsx �[2m+ 2 more�[0m
�[0;30mdev|�[0m �[0m�[1m1 |�[0m �[0m�[35mexport�[0m �[0m�[35mdefault�[0m �[0m�[35masync�[0m �[0m�[35mfunction�[0m MyPage(params) {
�[0;30mdev|�[0m �[0m�[1m2 |�[0m   myFunc()�[0m�[2m;�[0m
�[0;30mdev|�[0m �[0m�[1m3 |�[0m   �[0m�[35mreturn�[0m �[0m<�[0mh1>{JSO
... (truncated)

release without fix: 4 FAILED
bun test v1.4.0-canary.1 (9008ae7ab)

test/bake/dev/server-sourcemap.test.ts:
Dev server testing directory: /tmp/bun-dev-test-LMzvrB
bun add v1.4.0-canary.1 (9008ae7ab)
Resolving dependencies
Resolved, downloaded and extracted [0]
Saved lockfile

installed react@0.0.0-experimental-603e6108-20241029
installed react-dom@0.0.0-experimental-603e6108-20241029
installed react-server-dom-bun@0.0.0-experimental-603e6108-20241029
installed react-refresh@0.0.0-experimental-603e6108-20241029

6 packages installed [10.00ms]
bun install v1.4.0-canary.1 (9008ae7ab)

Checked 6 installs across 7 packages (no changes) [1.00ms]
�[0;30mdev|�[0m Started development server: http://localhost:38961
�[0;30mdev|�[0m �[32mBundled page in 51ms�[0m�[2m:�[0m pages/[...slug].tsx �[2m+ 2 more�[0m
�[0;30mdev|�[0m �[0m�[1m1 |�[0m �[0m�[35mexport�[0m �[0m�[35mdefault�[0m �[0m�[35masync�[0m �[0m�[35mfunction�[0m MyPage(params) {
�[0;30mdev|�[0m �[0m�[1m2 |�[0m   myFunc()�[0m�[2m;�[0m
�[0;30mdev|�[0m �[0m�[1m3 |�[0m   �[0m�[35mreturn�[0m �[0m<�[0mh1>{JSON�[0m�[3m�[1m.stringify�[0m(params)}�[0m<�[0m/h1>�[0m�[2m;�[0m
�[0;30mdev|�[0m �[0m�[1m4 |�[0m }
�[0;30mdev|�[0m �[0m�[1m5 |�[0m 
�[0;30mdev|�[0m �[0m
... (truncated)

passes on PR (with fix)

ASAN with fix: all passed
$ BUN_DEBUG_QUIET_LOGS=1 bun scripts/build.ts --profile=debug --quiet test "--reporter=junit" "--reporter-outfile=/tmp/mechgate.xml" test/bake/dev/server-sourcemap.test.ts
bun test v1.4.0 (178906144)

test/bake/dev/server-sourcemap.test.ts:
Dev server testing directory: /tmp/bun-dev-test-fA4VAw
bun add v1.4.0 (178906144)
Resolving dependencies
Resolved, downloaded and extracted [0]
Saved lockfile

installed react@0.0.0-experimental-603e6108-20241029
installed react-dom@0.0.0-experimental-603e6108-20241029
installed react-server-dom-bun@0.0.0-experimental-603e6108-20241029
installed react-refresh@0.0.0-experimental-603e6108-20241029

6 packages installed [174.00ms]
bun install v1.4.0 (178906144)

Checked 6 installs across 7 packages (no changes) [100.00ms]
�[0;30mdev|�[0m Started development server: http://localhost:43795
�[0;30mdev|�[0m �[32mBundled page in 1945ms�[0m�[2m:�[0m pages/[...slug].tsx �[2m+ 2 more�[0m
�[0;30mdev|�[0m �[0m�[1m2 |�[0m   myFunc()�[0m�[2m;�[0m
�[0;30mdev|�[0m �[0m�[1m3 |�[0m   �[0m�[35mreturn�[0m �[0m<�[0mh1>{JSON�[0m�[3m�[1m.stringify�[0m(params)}�[0m<�[0m/h1>�[0m�[2m;�[0m
�[0;30mdev|�[0m �[0m�[1m4 |�[0m }
�[0;30mdev|�[0m �[0m�[1m5 |�[0m 
�
... (truncated)

release with fix: all passed
$ bun scripts/build.ts --profile=release
[configured] bun-profile → bun (stripped)
  target       linux-x64-gnu
  build type   Release
  build dir    ./build/release
  revision     178906144a
  features     baseline

22 deps, 107 codegen, 1176 objects in 805ms

ninja: Entering directory `/workspace/bun/build/release'
[1/1238] install /workspace/bun
bun install v1.4.0-canary.1 (9008ae7ab)

Checked 107 installs across 153 packages (no changes) [14.00ms]
[2/1238] install /workspace/bun/packages/bun-error
bun install v1.4.0-canary.1 (9008ae7ab)

Checked 1 install across 2 packages (no changes) [4.00ms]
[3/1238] gen bindgenv2
[4/1238] gen ErrorCode+*.h
[5/1238] fetch picohttpparser
[picohttpparser] up to date
[6/1238] fetch tinycc
[tinycc] up to date
[7/1237] install /workspace/bun/src/node-fallbacks
bun install v1.4.0-canary.1 (9008ae7ab)

Checked 129 installs across 147 packages (no changes) [14.00ms]
[8/1237] gen .bind.ts → GeneratedBindings.cpp
[9/1237] fetch libjpeg-turbo
[libjpeg-turbo] up to date
[10/1237] gen ProcessBindingConstants.lut.h
Generating /workspace/bun/build/release/codegen/ProcessBindingConstants.lut.h from /workspace/bun/src/jsc/bindings/Process
... (truncated)

diff hotspot

src/runtime/bake/dev_server/incremental_graph.rs |  17 +---
 src/runtime/bake/dev_server/mod.rs               |  18 +++-
 src/runtime/bake/dev_server/source_map_store.rs  |  17 +---
 src/runtime/bake/mod.rs                          |  14 ++++
 test/bake/dev/server-sourcemap.test.ts           | 102 +++++++++++------------
 5 files changed, 85 insertions(+), 83 deletions(-)

gate history · 1 passed · 0 rejected · iteration 0

evidence per changed file

file                                              reads  edits  tests
src/runtime/bake/dev_server/incremental_graph.rs      1      2      0
src/runtime/bake/dev_server/mod.rs                    1      2      0
src/runtime/bake/dev_server/source_map_store.rs       2      4      0
src/runtime/bake/mod.rs                               1      1      0
test/bake/dev/server-sourcemap.test.ts                5     11      0

self-review · no surviving concerns

26 concerns were raised and did not survive verification.

[robobun]

Status: reproduced with the released build by running test/bake/dev/server-sourcemap.test.ts as updated here (4 tests fail with the old positions, e.g. at throwError (pages/error-page.tsx:6:1) for a throw on line 7); all 5 pass with this branch's debug build. Client-side maps (test/bake/dev/sourcemap.test.ts, test/bake/dev/html.test.ts) unchanged and passing.

[coderabbitai]

Walkthrough

The change centralizes HMR runtime prefix selection by chunk kind and side. Bundle generation and source-map assembly use the shared selection. Dev-server tests now verify exact source-mapped stack frames.

Changes

HMR prefix contract and integration

Layer / File(s)	Summary
Centralized HMR prefix contract src/runtime/bake/mod.rs, src/runtime/bake/dev_server/mod.rs	Adds HmrRuntime::from_static and ChunkKind::prefix for initial, client HMR, and server HMR responses.
Bundle and source-map integration src/runtime/bake/dev_server/incremental_graph.rs, src/runtime/bake/dev_server/source_map_store.rs	Uses ChunkKind::prefix(side) for bundle prefixes and source-map line offsets.
Exact source-map stack assertions test/bake/dev/server-sourcemap.test.ts	Adds shared output helpers and exact frame assertions for async errors, HMR, nested imports, and repeated reloads.

Suggested reviewers: jarred-sumner, dylan-conway

🚥 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 and concisely describes the primary change: correcting server-side stack-frame source mapping in the bake dev server.
Description check	✅ Passed	The description covers the problem, cause, fix, verification steps, test results, and known limitations using the repository template sections.

---

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

[claude]

LGTM — centralizing the chunk prefix in ChunkKind::prefix(side) fixes the off-by-one on server maps and makes the emitter and mapper agree by construction.

What was reviewed:

• Verified ZStr::from_static/as_bytes() exclude the NUL, so from_static(b"({\\0") yields line_count = 0 and the client literal yields 1.
• Confirmed client paths are unchanged: render_json(.., Side::Client) at DevServer.rs:1774 and render_mappings (client-only /_bun/report_error path) still see the same line counts as before.
• Checked render_json(.., ChunkKind::HmrChunk, Side::Server) at DevServer.rs:4291 is the server call site that now gets line_count = 0.
• Test assertions are strictly stronger (full at fn (…file:line:col) regex vs. substring includes); no existing coverage weakened.

Extended reasoning...

Overview

Fixes server-side stack-frame remapping in the bake dev server. source_map_store::Entry::join_vlq was ignoring its side parameter and always offsetting mappings by the client HMR-chunk prefix's newline count (1), while take_js_bundle_to_list_server emits a prefix (({) with 0 newlines — so every server frame remapped one generated line too high. The fix introduces ChunkKind::prefix(side) -> HmrRuntime as the single source for both the emitted prefix and its line count, plus a const HmrRuntime::from_static that derives line_count from the literal. take_js_bundle_to_list{,_server} and join_vlq all pull from it.

Five files touched: the three bake dev-server Rust files plus bake/mod.rs (the from_static helper) and the test file, whose expectations are corrected and tightened.

Security risks

None. This is dev-server source-map line-count arithmetic; no user input, auth, or serialization boundaries are involved.

Level of scrutiny

Medium. The change is small and mechanical, but it rewrites the assertions in an existing test file — REVIEW.md flags that as something to check carefully. I verified each rewritten expectation against the fixture source (line 7 = throw new Error, line 2 = the call, etc.) and confirmed the old expectations were the buggy output, not the correct behavior. The client path is provably unchanged: for (HmrChunk, Client) the new helper returns the same literal with line_count = 1, and for (InitialResponse, Client) it returns get_hmr_runtime(Side::Client) exactly as before. render_mappings still hardcodes Side::Client, which is fine since its only caller (get_parsed_source_map → error_report_request.rs) handles client-reported errors.

Other factors

• robobun confirms the updated test fails on the released build with the old positions and passes on this branch, satisfying the "test fails for the right reason" bar.
• test/bake/dev/sourcemap.test.ts and html.test.ts (client maps through the same function) were run and pass unchanged.
• The (InitialResponse, Server) arm now returns get_hmr_runtime(Side::Server) where the old code used the client runtime's line count regardless; the comment at take_js_bundle_to_list_server notes server InitialResponse is unreachable, so this is a consistency-only change.
• The PR description flags a known future conflict with #37396 (one column expectation changes from 7:13 to 7:9) and a separate unaddressed issue (server patch maps overwriting each other under bake://server.patch.js) — both correctly scoped out.