From 095930ca74af9caee6baf44942d0c59622432cd3 Mon Sep 17 00:00:00 2001 From: JavaZeroo <71128095+JavaZeroo@users.noreply.github.com> Date: Fri, 30 Oct 2026 12:02:32 +0800 Subject: [PATCH] docs(research): record the proven repaint mismatch for #194 The 441-byte chunk consumed at 111x7 carries a ten-row repaint that the three passing siblings consume at 149x10, so the mismatch is measured rather than inferred. --- docs/research/conpty-resize-repaint-timing.md | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/docs/research/conpty-resize-repaint-timing.md b/docs/research/conpty-resize-repaint-timing.md index 901e4e6f..abedb5f9 100644 --- a/docs/research/conpty-resize-repaint-timing.md +++ b/docs/research/conpty-resize-repaint-timing.md @@ -53,3 +53,20 @@ - 复现作业:`.github/workflows/conpty-compat.yml` 的 `windows-2025-actions-repeat`(覆盖 `text_actions|selection_context|shell`); - 诊断变量:`TERMFORGE_CAPTURE_OUTPUT`、`TERMFORGE_OUTPUT_TRACE`、`TERMFORGE_SELECTION_DIAG`; - 相关讨论:[#194](../..#194)、[#156](../..#156)(已关闭,机制记录)、[#38](../..#38)(托管稳定性汇总)。 + +## 已证实(2026-10-30) + +不再只是推断:同一轮的四个样本里,尾部重绘的**消费网格**可以直接读出。 + +失败样本 `text-actions/.../system-1`(terminal 1)的 16 个输出块中: + +``` +t=789406 bytes=210 cols=111 rows=7 repaint=true +t=789468 bytes=441 cols=111 rows=7 repaint=true <- 含 231B 十行形状 + 210B 七行形状 +``` + +同一轮的三个通过样本(local-0、local-1、system-0)里,对应的 231B 十行重绘被消费在 **149×10**;只有失败样本把它消费在 **111×7**。 + +因此"重绘按旧几何(10 行)生成、按新几何(7 行)消费"这一点现在由**字节形状 + 记录的消费网格**共同确定,见 [#194 复核评论](https://github.com/JavaZeroo/termforge/issues/194#issuecomment-6514631287)。 + +仍然未知:精确的 resize 时刻与顺序(该样本早于 `termforge-resize-trace-v1`,轨迹里 0 条 resize 记录),因此"+3 行"(top 75→78)的来源需要下一次带 resize 记录的命中才能补齐。