feat(jitdump): recover V8 inlined frames from the V8 code log#10
Open
GuillaumeLagrange wants to merge 1 commit into
Conversation
V8's jitdump (--perf-prof) records only the innermost inlined source position per PC, so a sampling profiler collapses an inlined chain root->a->b->c down to a single `root` frame for both TurboFan and Maglev (COD-2821; Maglev needs COD-2820). The full inline tree is emitted, with no node/V8 fork, to the code log's `code-source-info` records (positions + inlining_positions + inlined_functions), which use the same runtime code addresses as the jitdump. - v8_inline.rs: parse that log into a per-code-object inline map (keyed on address + size, since V8 reuses freed code addresses over a run) and reconstruct the inline stack via the SourcePosition::InliningStack walk. - jitdump.rs: resolve the producing process's log as $CODSPEED_V8_LOG/codspeed-v8-<pid>.log from the jitdump header pid, then expand a single jitdump frame into the full inline stack, innermost first, with the code object's own frame appended last so a non-inlined hit resolves exactly as before. Falls back to the single-frame path when no log is found. Refs COD-2821 Generated with AI Agent (Claude Code)
not-matthias
approved these changes
Jun 18, 2026
not-matthias
left a comment
not-matthias
left a comment
Member
There was a problem hiding this comment.
LGTM overall. I think the code readability can be improved a bit by:
- Dropping the references to COD-2821 (I didn't comment for all, so pls search for other references)
- Removing 1-3 letter variables by either renaming them or creating a struct (I think I prefer the latter)
| index: JitDumpIndex, | ||
| /// Inline map from the V8 code log of the process that produced this | ||
| /// jitdump, used to expand TurboFan/Maglev inlined frames that the jitdump | ||
| /// itself collapses (COD-2821). `None` when no log was found. |
Member
There was a problem hiding this comment.
I'd drop the (COD-2821) here, it's already included in the commit/branch.
| //! | ||
| //! V8's jitdump records only the *innermost* inlined source position per PC, so | ||
| //! a sampling profiler collapses an inlined chain `root -> a -> b -> c` down to | ||
| //! just `root` (COD-2821). The full inline tree is instead present in the code |
| /// `(code_offset, script_offset, inlining_id)`, sorted ascending by code_offset. | ||
| positions: Vec<(u32, u32, Option<u32>)>, | ||
| /// `code-source-info` inlining tree: `(inlined_function_id, call_site_script_offset, parent_inlining_id)`. | ||
| inl_pos: Vec<(i32, u32, Option<u32>)>, |
Member
There was a problem hiding this comment.
inlined_function_id can be negative? 🤔
Member
There was a problem hiding this comment.
a bit further down below we even check for negative results, maybe just convert it to u32 like the others
| } | ||
| let (_script_off, mut iid) = cur?; | ||
| let mut frames = Vec::new(); | ||
| // Mirror of V8 SourcePosition::InliningStack. |
Member
There was a problem hiding this comment.
nit: link to the function that's referenced could be helpful
| // Source-position table: an entry applies from its code_offset until the | ||
| // next one, so take the last entry with code_offset <= pc_offset. | ||
| let mut cur = None; | ||
| for &(co, so, iid) in &ci.positions { |
Member
There was a problem hiding this comment.
nit: I see the comment above but 2-3 letter var names are a bit hard to follow when reading. I think creating a struct could make sense here.
| /// there is no genuine inlining at that PC; the code object's own frame is | ||
| /// supplied by the caller (the jitdump symbol), so it is *not* included. | ||
| pub fn lookup(&self, code_addr: u64, code_size: u32, pc_offset: u64) -> Option<Vec<V8Frame>> { | ||
| let ci = self.by_key.get(&(code_addr, code_size))?; |
Member
There was a problem hiding this comment.
nit: by_key is already expressed by .get(key) -> could be renamed to code_inlines then it's also clear what ci is
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Linked: CodSpeedHQ/codspeed-node#83