[wan9chi]

Motivation

The file check stops at the first changed file, so 'src/a.ts' modified doesn't say whether reverting that file is enough. The dry-run RFC has -v check every input. With it, the detailed summary lists every change: in dry runs, in real runs, and in --last-details afterwards.

Changes

Without -v, nothing changes. With -v:

$ vp run --dry-run -v read
…
  [1] app#read: $ vtt print-file a.txt b.txt c.txt
      → Cache miss: 'a.txt' modified
                    'b.txt' modified             ← new with -v
                    'c.txt' modified             ← new with -v

• A FileCheck, first file or all, follows -v from the session to the cache lookup.
• The cache.input comparison and the search of the files the task read collect every change instead of stopping. The search stays parallel and unordered. A miss sorts its changes, so the output is stable.
• An input that can't be checked fails the lookup only when nothing else changed, the rule the tracked-env check already follows. Each change is its own result, so a folder entry whose name isn't a valid path doesn't hide the folder's other changes.
• A listed folder reports every entry added or removed.
• When the command, config or a cache.env var changed, only those changes are listed, as before: there's no cached run with the new ones to compare the files with. The RFC's -v section now says so, and drops its real-run example, which listed a file next to the env change.

Cost: with -v, a miss hashes every input instead of stopping at the first changed file.

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches core cache validation and miss reporting on every lookup; behavior change is gated on -v but verbose runs do more filesystem work per miss.

Overview
vp run -v now drives a FileCheck mode through cache lookup: default runs still stop at the first changed input file; with -v, validation walks all explicit glob inputs and inferred/folder inputs and reports every file difference in stable path order.

Cache miss diagnostics use input_changes and multi-entry folder diff logic instead of first_input_change. Partial check failures (e.g. invalid folder entry names) no longer hide other listed changes when something already differs. CHANGELOG, the dry-run RFC, CLI help, and an e2e snapshot fixture document and lock in the new verbose behavior for real runs, --dry-run, and --last-details.

^{Reviewed by Cursor Bugbot for commit 02daff6. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

fspy benchmark

linux

dynamic/launch             change  +0.66%  [ -6.16% ..  +9.26%]  overhead   +57.83%
dynamic/access             change  +1.58%  [ -5.15% ..  +8.04%]  overhead   +21.75%
dynamic/access-relative    change  +0.82%  [ -3.84% ..  +4.45%]  overhead   +72.65%
dynamic/access-contended   change  -0.34%  [-10.85% ..  +4.65%]  overhead   +39.50%
static/launch              change  -0.18%  [ -5.50% ..  +5.15%]  overhead  +195.35%
static/access              change  +0.65%  [-10.60% .. +11.49%]  overhead +1472.11%
static/access-relative     change  -1.69%  [-15.32% .. +10.10%]  overhead +1893.71%
static/access-contended    change  -4.29%  [-11.90% ..  +2.19%]  overhead +1622.39%

macos

dynamic/launch             change  -0.33%  [ -2.56% ..  +2.59%]  overhead  +225.61%
dynamic/access             change  +0.00%  [ -2.71% ..  +1.97%]  overhead    +3.07%
dynamic/access-relative    change  -0.72%  [ -3.12% ..  +2.08%]  overhead  +247.39%
dynamic/access-contended   change  +0.74%  [ -1.78% ..  +3.15%]  overhead    +3.67%

windows

dynamic/launch             change  +0.94%  [ -7.72% ..  +8.18%]  overhead   +25.64%
dynamic/access             change  +0.00%  [ -1.20% ..  +0.82%]  overhead    +1.84%
dynamic/access-relative    change  -0.18%  [ -6.20% ..  +1.68%]  overhead    +2.43%
dynamic/access-contended   change  -0.20%  [ -7.90% ..  +1.86%]  overhead    +1.97%