4.2.0 memory-optimization → latest-24.10 forward-port gap

Executive summary

The prplWare 4.2.0 RAM-optimization programme lives on proj_prplware-v4.2.0_optimizations branches cut from mainline-24.10 @ 69c998a2fab3. It re-pins 103 components across three feeds to optimization tags, plus integration-level changes in prplos/prplos (feed pins, gmap MIB compile-time flags, CRAM expectations). Since then the release train moved on and latest-24.10 drifted by 1,221 commits. This dashboard measures, per component, the optimization diff, the mainline drift, and their file-level overlap — the raw material for a forward-port effort estimate.

T-shirt rollup (non-ported): 50 S · 27 M · 18 L · 5 XL heuristics ↓

Risk mix by feed

What the optimizations actually do

The optimization branch is not a single technique but a programme: every resident daemon and library was re-pinned to an optimization tag carrying its own changes. Mining the 1,274 commit subjects across all component opt diffs, the recurring patterns are (a commit can match several themes):

• Build/config gating 55 commits
  optional features moved behind compile-time switches so they can be left out
• Deprecated-API swaps 54 commits
  e.g. amxd_object_get_param_value() → amxd_object_param_constcast(), avoiding variant copies
• Linking & shared-library rework 48 commits
  static→shared and dependency slimming in the Ambiorix stack
• Dead code & subsystem removal 47 commits
  unused helpers, log spam and legacy paths dropped
• Data-container tuning 29 commits
  e.g. libamxd “heap memory optimization for flags and parameter names”
• ODL data-model slimming 28 commits
  trimmed object/parameter definitions loaded at daemon start
• Plugin modularization 22 commits
  functionality split into loadable modules
• String/buffer handling 21 commits
  e.g. libamxj “extreme memory overallocation when parsing json string”
• Targeted leak fixes 11 commits
  e.g. mod-amxb-ubus fragmented-message reassembly, libamxb backend manager

At the integration level (prplos/prplos), the optimization branch disables the gmap MIB features INFORMATION, LOCATION, PRIORITY and WAN-ACCESS via CONFIG_SAH_SERVICES_GMAP-MIBS-COMMON_*=n in the touched profiles, re-pins all three feeds, and updates CRAM test expectations to the optimized behaviour — see the forward-port checklist.

Top 15 components by optimization churn

Full gap table — 103 components

† hover for per-component notes · opt_* abbreviates the proj_prplware-v4.2.0_optimizations_* tag prefix · “Opt commits/files/± lines” = base→opt compare in the component repo · “Drift” = commits base→current · “Overlap” = files touched by both diffs · sizes per heuristics.

New in feed-prplos since branch-off — 17 packages, never optimization-reviewed

The forward-port does not land on the base the optimizations were written for: 17 packages were added to feed-prplos after branch-off (feed_amx / feed-prplmesh: none). Two of them — tr181-schedules and mod-scheduler-amxp — ship in the prpl_core profile (81 → 83 packages), i.e. new resident memory consumers that may need their own optimization pass; the swupdate stack rides along in secure_boot_emmc builds.

Forward-port checklist — integration layer

Beyond the 103 component repos, the optimization lives in prplos/prplos integration changes (91 commits, 20 files on the opt branch) that must be re-applied on a branch that has meanwhile moved 1,221 commits:

1. Re-pin the three feeds in the touched profiles to forward-ported optimization tags. Current latest-24.10 pins: feed_amx 762470833040, feed-prplos 0c5d4e4a977d, feed-prplmesh 9ce07a36becd — the feeds moved 13/228/68 commits since branch-off. Touched profiles:
   • profiles/lcm.yml
   • profiles/mtk_filogic.yml
   • profiles/prpl_core.yml
   • profiles/qca_ipq95xx.yml
2. Re-apply the gmap MIB compile-time flags saving resident memory: CONFIG_SAH_SERVICES_GMAP-MIBS-COMMON_INFORMATION/LOCATION/PRIORITY/WAN-ACCESS=n.
3. Update CRAM test expectations to optimized behaviour (11 files, including board build-configs for six targets):
   • .gitlab/tests/cram/arcadyan-mozart/005-build-config.t
   • .gitlab/tests/cram/generic/acceleration-plan-components/014-firewall-levels.t
   • .gitlab/tests/cram/generic/acceleration-plan-components/016-firewall-ipset.t
   • .gitlab/tests/cram/generic/acceleration-plan-components/046-qos-classification.t
   • .gitlab/tests/cram/generic/obuspa.t
   • .gitlab/tests/cram/generic/ubus-datamodel-list.t
   • .gitlab/tests/cram/mxl25641-hdk-6/005-build-config.t
   • .gitlab/tests/cram/prpl-haze/005-build-config.t
   • .gitlab/tests/cram/turris-omnia/005-build-config.t
   • .gitlab/tests/cram/urx851-b0-dk/005-build-config.t
   • .gitlab/tests/cram/wnc-freedom/005-build-config.t
4. Re-align CI plumbing (docker images, release pipeline):
   • .gitlab-ci.yml
   • .gitlab/docker/ib/gitlab.yml
   • .gitlab/docker/ib/lcm/gitlab.yml
   • .gitlab/docker/sdk/gitlab.yml
   • .gitlab/release.yml
5. Stabilize the run stage on the boards the opt branch tuned expectations for (MXL25641-HDK-6, WNC Freedom) before trusting CRAM comparisons.
6. Pick up the in-flight gmap forward-port. dev_ram_optimizations_gmap_latest-24.10 exists (1 commit ahead of latest-24.10, last activity 2026-04-07); its MR was closed unmerged:
   • !2531 Integrating: Various optimizations — closed, updated 2026-04-07
7. Optimization-review the new resident consumers. tr181-schedules and mod-scheduler-amxp entered the prpl_core profile after branch-off and have never been optimization-reviewed — see new components.

Footnotes

Unresolved components (1)

• feed_amx/libsahtrace: opt ref 'sah-next-4.2_v0.1.1' no longer exists upstream (tag deleted or repo moved); current pin identical to base (re-tag only); risk HIGH: metrics unresolved

Unresolvable metrics are pinned to risk HIGH rather than silently dropped; the release tarball (PKG_HASH in the feed) still exists for a manual diff.

Scoring heuristics

Risk — PORTED: the opt ref is fully contained in (or identical to) the current pin. LOW: zero mainline drift (re-tag), or an opt diff ≤ 5 files with no file overlap. HIGH: ≥ 5 overlapping files, overlap ≥ 30% of the opt diff, or truncated/unresolvable data. MED: everything else. Overlap = files changed by both the optimization diff and the mainline drift — a proxy for merge conflicts.

T-shirt size — points for churn (>5 files or >200 lines: 1, >20/1000: 2, >50/3000: 3), drift (>5 commits: 1, >30: 2, >100: 3) and overlap (≥1 file: 1, ≥5: 2, ≥10: 3); S ≤ 1, M ≤ 3, L ≤ 6, XL ≥ 7. Heuristic ranking, not person-days.

Method & data

All numbers are re-derived from GitLab (never from prose docs): the component set is the Makefile-blob diff of each feed between the branch-off and optimization SHAs; refs are parsed from each package Makefile at three feed states; per-component metrics use the GitLab compare API (three-way: base→opt, base→current, current→opt for ported-detection). Truncated compares fall back to exact commit counting. Regenerate with make fetch build — sources in the repo.

uriparser is third-party (GitHub-hosted); its gap is the feed-level packaging diff only.

Removed packages

No packages were removed from the feeds between branch-off and the current pins.