Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

JIT: the EIGS_JIT_HOT 'native-byte share' metric weights chunks by frame entries, ignoring back-edges and OSR — reads 100% on a program that is ~0% native by that measure

Open
  • Is something wrong?
#1,179 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
52/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
c
Domain
compilers, performance

Research direction

Reproduce the EIGS_JIT_HOT output with tests/bench_dmg_shape.eigs, then trace the aggregate’s use of exec_count, advance, back_edge_count, and OSR coverage. Establish whether the metric can defensibly represent executed native and total bytes, including loop back-edges and OSR thunks, and verify that the resulting report no longer gives the documented misleading share or that the metric is removed.

Written by the indexing model from the issue text.

Description

area:jit found-by:critic kind:gate-defect

The native-byte share aggregate at the foot of the EIGS_JIT_HOT dump weights every chunk by exec_count — frame entries — when the quantity it claims to report is executed bytecode bytes:

bytes_native += r->exec_count * (uint64_t)r->advance;
bytes_total  += r->exec_count * (uint64_t)r->code_len;

A chunk entered once that then runs a hot loop for 565k back-edges contributes 1 * code_len. On tests/bench_dmg_shape.eigs that is the whole program:

<module>    1  ?   0.0%    0   636   0.0%  565536  yes    61    56  <end>
native-byte share (top 9 hot): 100.0%   (all 9 chunks): 100.0%

The 100% is an artifact — the module's 565k loop iterations count as 636 bytes against 17.6M from the opN callees, so the one chunk that dominates the run is rounded out of the metric.

Two errors compound:

  1. Back-edges are not counted. back_edge_count is tracked per chunk and ignored by the aggregate.
  2. OSR coverage is not counted. The advance term uses jit_advance only. A chunk whose work is entirely inside an OSR thunk (osr=yes, oadv=61 above) scores 0% native while running native.

So the number that should answer "would extending the JIT prefix pay?" understates loop-heavy programs and overstates call-heavy ones. On DMG it reads 46.0% — and perf says 9.9% of samples are actually in JIT'd code, so it is not usable as a proxy for time either.

A defensible version weights each chunk's native/total bytes by an estimate of bytes actually executed (entries for straight-line prefixes, back-edges for loop bodies, and the OSR thunk's advance where an OSR slot is live), or the metric is dropped in favour of the perf attribution it is trying to approximate.

Context: the dump itself only started producing output again in #1176 — this metric has never been read against a real workload.

Dominant language
C
Stars
3
Forks
7
Avg merge
3h 56m
Merged PRs (30d)
102

Getting set up

Open in Codespaces

Starts the project's dev container in your browser, under your own GitHub account.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from InauguralSystems/EigenScript

All issues in InauguralSystems/EigenScript

Similar issues

More C issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.