Skip to content

DEBLOAT.md

Latest commit

 

History

History
63 lines (47 loc) · 3.31 KB

DEBLOAT.md

File metadata and controls

63 lines (47 loc) · 3.31 KB
Raw

Debloat design (what Slipstream strips, and what it deliberately keeps)

Order of preference: GN arg → default pref → small patch (every patch is rebase debt).

GN args added on top of Thorium's win_AVX2_args.gn

Arg Value Why
enable_rlz false Google install-attribution library; Thorium leaves it on. Highest-value single flag.
enable_mdns false Kills mDNS/DIAL chatter AND is the PRIV-03 local-IP concealment lever (paired with the WebRTC pref, C2 invariant). enable_service_discovery is derived from this (features.gni) — never set it directly.

Dropped at the gn-gen gate (M152), do NOT re-add:

  • enable_background_mode=false / enable_background_contents=false — chrome/browser/background/extensions/BUILD.gn asserts enable_background_mode, and that target is pulled in whenever extensions are enabled. Extensions (CWS) are a hard requirement, so background mode cannot be compiled out at M152. Re-check each milestone; if upstream decouples them, revisit. | is_component_ffmpeg | true | LGPL hygiene (separate ffmpeg.dll). | | bundle_widevine_cdm | false | Runtime component fetch instead — clean redistribution posture. | | safe_browsing_mode | 1 (keep) | See DECISIONS.md. |

Verify each against docs/gn_all_args.txt for the pinned milestone before shipping — the arg surface churns.

Stage-90 patches (see patches/series)

metrics off · crash-upload off (local minidumps kept) · variations fully neutered · component-updater trimmed surgically · optimization-guide fetches off · promo surfaces removed · default-prefs hardening (no prefetch, no search suggest, DoH secure, no UKM).

Deliberately KEPT (do not "debloat" these)

  • Component updater's security components: CRLSet, Chrome Root Store, Origin Trials, cert components — the out-of-band security channel.
  • Certificate transparency log list updates.
  • The entire extension update pipeline and webstorePrivate API.
  • The Chrome/<major>.0.0.0 User-Agent token (CWS gates on it).
  • Windows download quarantine / MotW.
  • chrome://net-internals, chrome://policy (support tooling).

Pillar design (2026-08-30)

The memory/privacy/security feature set is specified in docs/PILLARS.md (verified implementation plan) and docs/pillar-proposals.json (evidence).

Known FAST costs, accepted deliberately (PILLARS.md C6/C7):

  • Purge-on-freeze also purges bfcache pages → back-navigations rebuild caches.
  • Default-search renderer no longer pinned alive → tens of ms on first search after process death (masked by network; search-suggest is off anyway).

Deliberately KEPT despite memory cost (security wins):

  • Strict site isolation, BackupRefPtr/MiraclePtr (build_overrides/partition_alloc.gni), the warm spare renderer, Blink's strong memory cache (pressure-purged instead).

Power-user memory opt-in (documented, never default): add --process-per-site to a shortcut to merge same-site tabs into one process. Caveats: bigger crash blast radius, coarser tab discarding. The shipped default already includes upstream's bounded same-site reuse (kProcessPerSiteUpToMainFrameThreshold: 2 frames / 2 GB cap).

Verification per release

10-minute idle capture must show zero requests to UMA/variations/ optimization-guide/RLZ endpoints; CWS install + update must succeed.