Gem settings resolution skips Bundler's global config (~/.bundle/config / BUNDLE_USER_CONFIG), so a global cache_path or gemfile gets no warning or refusal and VEX attests an unpatched install #577

[agent] Found by the scheduled Bundler (RubyGems) bug-hunt routine (ledger #316).

Summary

#532 (fixing #483 and #507) made socket-patch read Bundler settings "in Bundler::Settings priority", but it only reads two of the tiers Bundler consults: the local app config ($BUNDLE_APP_CONFIG/config, else .bundle/config) and the environment. Bundler's lookup is temporary → local → env → global → default. The global tier is ~/.bundle/config, or $BUNDLE_USER_CONFIG / $BUNDLE_USER_HOME/config, which is what bundle config set --global … writes. socket-patch never reads it, so the two settings that #532 just fixed regress whenever they are set globally:

• cache_path (stale-install guard, Gem hosted stale-install guard only checks vendor/cache, so a committed cache at a configured cache_path gets no warning, the in-run VEX attests it, and Bundler 2.4 installs the unpatched gem #483's arm): with bundle config set --global cache_path vendor/gems and a committed unpatched vendor/gems/<gem>.gem, hosted scan gives no redirect_gem_stale_install warning, and the same run's --vex attests the gem as not_affected. Bundler then installs the archive from vendor/gems.
• gemfile (Gem manifest resolution lets the BUNDLE_GEMFILE env var override .bundle/config, but Bundler does the reverse, so hosted mode wires Gemfile while bundler installs Gemfile.next unpatched and VEX attests it #507's arm): with bundle config set --global gemfile Gemfile.next, hosted scan rewires Gemfile and exits 0, while Bundler loads Gemfile.next and installs the unpatched gem. With the same value in the local .bundle/config, the scan correctly refuses with redirect_gem_bundle_gemfile_unsupported.

Impact

The patch silently isn't applied, and the VEX document says it is. This is the same outcome #483 and #507 were filed for. Only the place the setting is stored differs, and per-user global config is common on CI images and developer machines.

Repro

Bundler side (real installs, Ruby 3.3.6):

# cache_path: Bundler installs from the globally configured cache dir
mkdir g1 && cd g1 && printf 'source "https://rubygems.org"\ngem "rack", "3.1.8"\n' > Gemfile
export HOME=$PWD/../home && mkdir -p $HOME
bundle _2.4.22_ config set --global cache_path vendor/gems    # lands in $HOME/.bundle/config
mkdir -p vendor/gems && cp /path/to/rack-3.1.8-with-marker.gem vendor/gems/rack-3.1.8.gem
BUNDLE_PATH=$PWD/../bp bundle _2.4.22_ install
grep -c BUGHUNT-MARKER ../bp/ruby/3.3.0/gems/rack-3.1.8/lib/rack.rb   # 1: the cached bytes were installed
# (control: without the global config the marker count is 0; on 4.0.17 the same setup exits on a CHECKSUMS mismatch, which proves it read vendor/gems)

# gemfile: Bundler loads the globally configured manifest
printf 'source "https://rubygems.org"\ngem "rack", "3.1.7"\n' > Gemfile.next
bundle _4.0.17_ config set --global gemfile Gemfile.next
bundle _4.0.17_ install    # "Installing rack 3.1.7", writes Gemfile.next.lock

socket-patch side (a copy of the gem_hosted_stale_archive_at_configured_cache_path_warns_and_is_not_attested case in crates/socket-patch-cli/tests/e2e_redirect_gem_stale_install.rs, with no .bundle/config and the setting written to $HOME/.bundle/config, or to a BUNDLE_USER_CONFIG file, instead):

global cache_path (HOME/.bundle/config):  code=0 redirect_gem_stale_install warnings=0 VEX attests pkg:gem/stale-probe-gem@1.0.0
global cache_path (BUNDLE_USER_CONFIG):   code=0 warnings=0 VEX attests the purl
global gemfile=Gemfile.next:              code=0, Gemfile rewritten into the patch-registry source block, no refusal
local  gemfile=Gemfile.next (control):    code=0, Gemfile untouched, redirect_gem_bundle_gemfile_unsupported

Each line reproduced twice.

Expected vs actual

• Expected: CLI_CONTRACT.md's "Gem stale-install guard" says the cache dir is "resolved in Bundler::Settings priority", and docs/ecosystems.md says a configured manifest other than the project's Gemfile / gems.rb "is refused with redirect_gem_bundle_gemfile_unsupported". Bundler's priority includes the global config below the env var, so a global cache_path should move the probed cache dir, and a global gemfile should be refused.
• Actual: both global settings are ignored. The probe stays on vendor/cache, Gemfile is rewired, and VEX attests.

Matrix

First bad: always. v4.0.0 and earlier ignored the local tier too, and #532 (cbf1f74) added the local tier but not the global one. Tested on main 1169ae6.

Suspect code

• crates/socket-patch-core/src/crawlers/ruby_crawler.rs:984 (read_app_config) is the only config source used by bundler_app_cache_dir_with_env (:1040) and bundler_loaded_manifest_with_env (:962). There's no fallback to the global file after the env var.
• crates/socket-patch-core/src/formats/gem/manifest.rs:27: "The user-level ~/.bundle/config is not consulted." That's an internal note, not in the user-facing contract, which promises Bundler's priority.
• The same gap probably affects the agent crawler's BUNDLE_PATH discovery (discover_bundle_stores_with_env) for bundle config set --global path …. I haven't verified that end to end.