Skip to content

docs(plan): §7 —— 四个缺口的实测回填 - #556

Merged
Sunrisepeak merged 1 commit into
mainfrom
docs/four-gaps-outcome
Sep 4, 2026
Merged

docs(plan): §7 —— 四个缺口的实测回填#556
Sunrisepeak merged 1 commit into
mainfrom
docs/four-gaps-outcome

Conversation

@Sunrisepeak

Copy link
Copy Markdown
Member

四项的实测结果回填。其中两条第一版通过了但什么都没测到:工具分档的判据在已装模拟器的机器上两半都读空;孤儿测试的前两版分别用 timeout(它向进程组发信号,修复前也能打到 ninja)和 200 个并行 TU(32 核主机在窗口内编完)。两者的修法都写下来了。

缺口 1 的答案记成了接口改动而不是一条注记:值放得进 arch_u64,而存储宽度问不出来,所以有了 arch_pte_entry_bytes(),头文件里那句「64 bits on every machine here」也一并改正。

缺口 4 的成因仍未确证,§1.3 保持原样。ninja 那边挖出比 §1.5 更强的事实:同一尺寸的十二个副本有十二个不同哈希 —— 那是就地 patch,所以已安装的二进制根本无法与发布校验和比对

Records what each of the four items measured. Two of them first passed without
testing anything: the tool-tier assertion read empty on a machine where the
emulator was already installed, and the orphan test's first two drafts were
signalled through `timeout` (which reaches the child either way) and then sized
so that a 32-core host finished the build inside the window. Both are written
down with the shape that fixed them.

Gap 1's answer is recorded as an interface change rather than a note: the entry
value fits in `arch_u64` and the STORAGE width was not askable, so
`arch_pte_entry_bytes()` exists and the header's own claim was corrected.

Gap 4's cause remains unestablished; §1.3 stands. The ninja work turned up a
stronger fact than §1.5 had: twelve copies at one size with twelve different
hashes, which is in-place patching, so an installed binary cannot be compared
against a published checksum at all.
@Sunrisepeak
Sunrisepeak merged commit d3773e7 into main Sep 4, 2026
36 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants