Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

EigenOS link break: the freestanding allowlist is a ONE-SIDED contract — pthread_once/pthread_self are declared HAL roots that EigenOS never implemented

クローズ
  • 何か問題がありますか？
#1,181 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
48/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
c
領域
build-system, operating-systems

調査の方向性

Start with EigenOS/src/hal.c and the pthread roots in tools/freestanding_allowlist.txt, then inspect trace.c:1392/1400 and builtins.c:2299/2315 to understand the existing callers and guards. Add the two missing HAL roots, implement the reverse allowlist-versus-nm check, and correct the allowlist comment. Done means the freestanding link succeeds, the safety behavior remains enabled, and the check fails when a declared HAL root is absent.

索引モデルが issue の本文から書いたものです。

説明

area:embed blocks-release found-by:cold-read kind:gate-defect

EigenOS fails to link the runtime: undefined pthread_once and pthread_self.

The obvious reading is "builtins.c and trace.c call pthread unconditionally, so add EIGENSCRIPT_FREESTANDING guards". Measured, that is the wrong fix — and it would trade away a loud-failure guard.

What is actually true

tools/freestanding_allowlist.txt deliberately lists both symbols as HAL roots that EigenOS is expected to provide:

# --- scheduler roots (pthread over spawn/yield/park) ---
pthread_create
pthread_join
pthread_once          <-- line 44
...
# Thread IDENTITY, not scheduling: #1142's replay tape records which OS
# thread opened it (g_replay_owner_tid) so a take from any other thread
# fails loud instead of silently mixing taped and live values. EigenOS's
# cooperative scheduler already owns the pthread roots above; the identity
# pair is the same HAL story — a task handle the kernel can compare.
pthread_self          <-- line 66

So make freestanding-check is green correctly. The runtime is using only declared roots. The gate is not blind.

The break is that EigenOS does not provide two of them. EigenOS/src/hal.c implements 14:

pthread_create pthread_join pthread_mutex_init/lock/unlock/destroy pthread_cond_init/wait/timedwait/signal/broadcast/destroy pthread_condattr_init/destroy/setclock

and is missing exactly pthread_once and pthread_self — which is exactly the pair that fails to link.

(Instrument calibrated before reading absence as fact: the same grep finds eigs_embed in 3 files and malloc in 9, and finds pthread_create/pthread_mutex_lock in 1 file each.)

Why guarding them out is the wrong direction

  • trace.c:1392/1400 is #1142's replay-owner check: a tape take from a thread other than the one that opened it fails loud instead of silently mixing taped and live values. Compiling that out on EigenOS removes the loud failure and leaves the silent-wrong behaviour it was built to prevent — the exact trade the fail-soft→loud reform (#971) forbids.
  • builtins.c:2299/2315 is one-time PRNG seeding. Note the inner function already carries #if EIGENSCRIPT_FREESTANDING for getpid(), so freestanding was considered there and the pthread_once wrapper around it was missed — the half-converted shape (outer guard left behind while the inner one was done).

Both are a few lines on a cooperative scheduler: pthread_self() returns the current task handle; pthread_once(once, fn) is a flag plus a call. That matches what the allowlist already promises, and keeps the safety guard.

The finding underneath, which is the durable one

The allowlist is a one-sided contract. The gate checks "the runtime uses only allowlisted roots" and nothing checks "EigenOS provides every allowlisted root." So a root can be ADDED to the contract without its counterparty ever being told.

That is what happened: #1142 added pthread_self and wrote, as its rationale, that "EigenOS's cooperative scheduler already owns the pthread roots above." Measured today, it owns 14 of 16 — and not those two. The sentence was reasoned, not checked.

The missing direction is cheap and mechanical: compare the allowlist against nm of EigenOS's HAL objects and fail on any allowlisted root the HAL does not define. That check would have gone red the moment #1142 landed, in the repo that made the promise, instead of surfacing as a link error in the consumer weeks later. It is the same both-directions rule the gate already applies to its own undefined-symbol set — just extended across the repo boundary.

Suggested shape

  1. Add pthread_once and pthread_self to EigenOS/src/hal.c (restores the link, keeps #1142's guard).
  2. Add the reverse check so the contract cannot gain an unimplemented clause again.
  3. Fix the allowlist comment, which currently asserts something false about the counterparty.
主要言語
C
スター
3
フォーク
7
平均マージ
11時間 30分
マージ済み PR（30日）
269

環境構築

Codespaces で開く

このプロジェクトの開発コンテナを、あなたの GitHub アカウントでブラウザ上に起動します。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

InauguralSystems/EigenScript のほかの issue

InauguralSystems/EigenScript の issue をすべて見る

似ている issue

C の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。