Roadmap

← POPSLoader docs · view on GitHub ↗

Last updated: 2026-11-04 (official release 1.4.1)

ROADMAP

This is the forward plan. For current runtime state, the canonical Known-Issues list, the Preservation Contracts, the Behavioral Invariants, and the per-case Hardware Status, see STATE.md — those volatile/shared facts live there and are not re-enumerated here.

Status Snapshot

• Public release is 1.4.1 (2026-11-04). Network-only UDPFS and Ember remain experimental and console-unverified. See STATE.md for current status.
• The 2026-06 (BETA-12-era) deliverables: HDD-resident settings save + in-app HDD .hide (boot-partition RW take-over via EnsureBootPartitionWritable), PAL native 640×512 full-screen render + auto-revert display-change confirm, the bdma_mode.txt marker rename (legacy names still read), the POPSTARTER Memory Card Folder toggle + BDMA⟺folder interlock, the launch-args auto-launch consumer, and the live CI luac syntax gate. See STATE.md > Repo-Verified Runtime State for the detail.
• The BETA-13 session added: gated analog-stick→d-pad fold + frame-counted nav auto-repeat + frame-counted description scroll (all hardware-confirmed), the Boot sound On/Off setting (hardware-confirmed), the layered cover_default.png + cover_missing.png cover placeholder (replacing the removed MISSING.png, −62 KB ELF), the OPL-style Overscan (CRT inset) render-inset adjuster, the HDD save-after-scan fix + Proposal A scan-slot steering, a 6-finding Codex audit pass, POPSTARTER.ELF / SMB-pack release-zip packaging, and per-device POPSTARTER.ELF resolution (PLDR.ResolveLaunchPopstarterPath; removable order: explicit custom/absolute-profile path → the game's own <device>:/POPS/POPSTARTER.ELF → launcher cwd → mc0:/mc1: net; APA order: custom → hdd0:__common/POPS/POPSTARTER.ELF via the same __common partition machinery → cwd → net — an explicit custom path now wins and Profiles act as presets, superseding the earlier device-local-first precedence; commits 28e40bb/9f2477c/26bb06c, not yet hardware-tested). See STATE.md > Repo-Verified Runtime State.
• The 2026-07-09 wave added: Adaptive BDMA (issue #509; per-launch BDMA variant staging with an equipped-check-before-write, launch-cancel on a staging failure, and an adaptive-off restage of the chosen mode), partition-installed POPS games on the HDD (PFS) page (HDDOSD/PSBBN PP./__. one-partition-per-game installs via the new HDD.ListPartitions APA enumerator + an IMAGE0.VCD mount-probe; launch argv0 PP.<partition>.ELF), the boot chime re-encode (22.05 kHz mono, −190.9 KB raw, zero code changes), and HDD-page diagnostics for the open non-HDD-boot list failure (real mount rc surfaced, dir-read-fault toast, status re-probe instead of a dead latch). Commits 6d6a368..3eff270, adversarially reviewed pre-push (8 confirmed findings, all fixed), none hardware-tested. See STATE.md.
• Hardware status, preservation contracts, and known issues are tracked in STATE.md — see STATE.md > Reported Hardware Status, > Preservation Contracts, and > Known Issues. In short: D-10/D-14/D-15, DKWDRV-from-MC, and BOOT.ELF-from-USB (L-07) remain hardware-PASS contracts; U-10 (BOOT.ELF from HDD-booted POPSLoader), DKWDRV-from-custom-HDD-path, and the Class-A HOSDmenu / some-wLE start failures are now RESOLVED (no longer known-broken); the 2026-06 HDD/PAL/BDMA features are implemented / boot on PCSX2 / HDD RW confirmed on hardware (provato) / full flow validating on hardware.
• CAUTION — load-order regression risk. A boot brick (PLDR.HDD methods defined before PLDR.HDD existed) made recent HDD-feature rolling builds un-bootable; fixed d4b04be (2026-06-17). Runtime nil-global / type / load-order errors are invisible to luac -p and CI — only the syntax gate is automated. Re-test HDD-feature builds on PCSX2/hardware before trusting a green CI.

Infrastructure landed post-release:
- .github/workflows/rolling-release.yml — automated rolling-release artifact publication on push to dev. PR events still build and run the gates, but the publish step is if: success() && github.event_name == 'push', so a PR can no longer republish the channel (gated since PR #511, 2026-07-16).
- docs/archive/DOCUMENTATION_FOLLOWUP_AUDIT.md — handoff plan for the post-BETA-10-5 doc cleanup work (now completed; see "Documentation cleanup" under Secondary Work).

SMB (v1) network game browsing is implemented (CI + Rolling green; browse + launch hardware-confirmed on a static IP config, DHCP path fixed in bb62f2be but unconfirmed): the SMB carousel page / GSMBNET scene with network settings, an install-toggle SMB module pack, lazy connect-on-entry, share browse, and POPStarter launch via smb:/POPS/SB.<name>.ELF — see Secondary Work §1 for the detail. ILINK remains an intentionally unimplemented menu entry. HDD (exFAT) is also implemented (BDMA Mode ATA, df2eb9d) and is validating on hardware; the launch-args path routes -page=ata/-mode=ata (and ata0/ataN) to it (opt 3, scene GBDMHDD), while -page=hdd/apa/pfs target the classic PFS page (opt 4) — 3d89631.

Immediate Priorities

POPS UDPFS ? included in 1.4.0: UDPFS Assault 1.0.0 replaces the USB-first mode with network reads and VMC writes; USB is no longer required. Physical-console acceptance remains pending. See setup and limitations.

Ember PR #600: separate Ember library/launcher and Beta 2 global/per-game settings; RiptOPL-matched POPS/ and EMBER/ packaging; exact eight-file APPS/PS1_POPSLOADER/ plus flat PSU with independent verification; digest-verified upstream Ember Watch. Real PS2 Ember game launch, controllers and VMC saves remain hardware-unverified. i.Link is explicitly deferred until RiptOPL proves it on real PS2 hardware and is not a completion/merge requirement for this PR. See the maintained wiki.

1) Cut BETA-13 — DONE (shipped as 1.0.0 on 2026-07-10, then 1.0.1 2026-07-13 and 1.1.0 2026-07-21; kept below as history)

• The BETA-13 input/UX/cover work is in and the headline items are hardware-confirmed: up/down + analog-stick nav land on individual items with smooth continuous scroll (oldman63), boot sound saves and survives reboot (oldman63), cover art loads via the Cover/details folder setting (FifthFox, 2026-06-30), and "everything working fantastically" on the latest rolling (Nuno6573). See STATE.md > Reported Hardware Status.
• Remaining BETA-13 eyeballs before the cut (none currently believed broken — all "awaiting a look"): the Overscan (CRT inset) render-inset on a real CRT; the cover_default.png + cover_missing.png overlay registration against the jewel-case window on both NTSC and PAL; the description right-stick scroll feel; and the HDD Proposal A scan-slot steering (deliberate HW test — see #5 below).
• Cutting BETA-13 needs the tree-adopting merge to master (the proven -s ours + read-tree + verify-empty-diff mechanic; master diverges from the dev branch, so a plain merge is wrong). The mechanic is archived at docs/archive/RELEASE_BETA13.md (with its gate checklist alongside it): preconditions/HW gate, release-prep on the dev branch, then the verified merge + tag + publish sequence. It is written for the BETA-13 cut specifically and names branches that were since renamed, so read it as the pattern rather than as steps to run verbatim — 1.0.0, 1.0.1 and 1.1.0 were all cut this way.

2) µs-as-ms timer-unit sweep — DONE

• Root-cause fact: Timer.getTime() returns microseconds on PS2 (raw clock() tick; CLOCKS_PER_SEC=1e6 — source-verified). Nav auto-repeat and description scroll were already moved to frame-counting (the canonical Enceladus idiom).
• The sweep finished the rest in 9c3f64f + a8e61f3: the PathEditor key-flash + caret blink and the launch-watchdog label are unit-correct (Timer.getTime()/1000); the dead MIN_ACTION_MS action debounce was removed (edge-triggering is the real gate); and the scene-fade / boot-fade / carousel were frame-paced preserving their current feel. os.clock() was deliberately not introduced (frame-counting is the idiom); it stays unused. Only cosmetic/inert sites (busy-overlay throttle, saving marquee, disabled debug log) intentionally remain.

3) Settings UI redesign (Berion mockup) — RESUMED incrementally

• RESUMED 2026-06-22 (maintainer call), built incrementally on dev (then still named BETA-13-PLAY) for tester access (luac/CI green, NOT HW-verified): the OPL-style focused-list now has a focus-following scroll viewport + scrollbar, and collapsible/expandable sections (6d26819 — each section header folds; X toggles, Left collapses, Right expands; session-only collapse state, no new persisted key). Backup branch checkpoint/pre-settings-redesign-2026-06-22.
• Banked next steps (do not forget): (a) the icon menu / overlay from the 2nd 2026-06-22 mockup — Settings / Save Changes / Restore Default / Joypad Configuration / Memory Card Manager / About / Exit; (b) a Joypad Configuration screen; (c) optionally default-collapse the heavier sections (Carousel Devices, Game List) so the page opens compact; (d) persist collapse state if users want it (currently session-only). The 2026-06-22 screenshots supersede the missing Berion PNGs for the scroll/submenu work specifically.
• Blocked on Berion's mockup PNGs landing at C:\Users\natha\Documents\assets\ and being committed to docs/mockups/ for the full per-screen redesign. Without the visual oracle, the deeper Lua port is hard to start.
• The original launch-path hardware blockers (D-10/D-14/D-15/DKWDRV-MC/BOOT.ELF, and now U-10) are settled. However, the 2026-06 HDD-resident settings save, in-app HDD .hide, and PAL-512 features are still validating on hardware (boot on PCSX2; HDD RW confirmed on hardware by provato; full flows not yet broadly hardware-confirmed). Per the maintainer, don't kick off the redesign until those settle, so the redesign builds on a verified settings/persistence base. See STATE.md > Reported Hardware Status.
• Scope: Context menu, Settings (per-category pages superseding the OPL focused-list), Joypad configuration, On-screen keyboard. Boot/splash and game list are out of scope.
• Full implementation prompt and per-screen pixel specs live in docs/archive/GUI_OVERHAUL_PROMPT.md.

4) Layer C full lazy IRX loading

• Precursor (pre-IRX device classification hint) landed in PR #458.
• mmceman deferral — SHIPPED (PR #471, merged; commit d10ec56). MMCE boots load mmceman eagerly (gate at src/main.cpp:503); USB/MC/MX4SIO/HDD boots defer it and lazy-load on demand via PLDR.EnsureMmceReadyOnce() (bin/POPSLDR/system.lua:1206 — System.ensureMmceman + System.reinitPad). MMCE pad-input behavior hardware-confirmed (FifthFox, commit 9f2d550). The vendored mmceman blob was later dropped in favor of ps2sdk's mmceman (no iop/embed/mmceman.irx in-tree).
• ds34bt deferral (Bluetooth pads) — DECLINED (2026-06-22). ds34bt hard-imports usbd and there is no clean "BT pad present" boot signal (pads pair after boot), so any defer gate risks dropping BT input for a small gain.
• usbd deferral — DECLINED (2026-06-22), the high-risk one. ds34usb and ds34bt both hard-import usbd, and the boot-device hint classifies the boot medium not pad transport, so deferring strands USB/BT controller input on every non-USB boot — unrecoverable without a reboot. Keep usbd + ds34usb + ds34bt eager.
• Decision: Layer C is closed at the mmceman win. The projected 30-50% pre-Lua gain in docs/archive/LAUNCH_HYGIENE.md assumed all three deferrals; with ds34bt/usbd off the table the realistic remaining gain is small and not worth the unrecoverable input-loss risk.

5) Display, HDD, and UX verification

• Overscan (CRT inset) — new OPL-style render-inset shipped but not yet eyeballed on a CRT. Screen.setOverscan(permille) / getOverscan() (src/luaScreen.cpp) drive a graphics.cpp transform that wraps every gsKit_prim_* draw site in OVX()/OVY() (identity at permille 0, so inert by default; math identical to OPL rmSetOverscan). The live adjuster is Settings → "Overscan (CRT inset)" (bin/POPSLDR/ui.lua:3649, ±5 step, live preview, discard restores). Confirm the inset on a real CRT.
• HDD Proposal A — the game scan now steers off the live boot pfs slot (b159a43; the boot/settings partition lives on the boot slot, "NEVER reuse"), and the boot RW mount is re-validated on the save path so settings save after a scan (8d1e67a). Still wants a deliberate HW test (Nuno) that game partitions still mount/list when forced off the boot slot. See STATE.md and memory reference-hdd-pfs-slot-model.
• U-06 targets the PAL native 640×512 full-screen render (the menu fills the screen, no letterbox; NTSC is 640×448). On PAL hardware confirm the full-screen fill and the auto-revert display-change confirm prompt (reverts if the new mode isn't confirmed, like OPL). See STATE.md > Reported Hardware Status (U-06 row).
• Re-run U-08 / U-09 on slower/large libraries to judge whether busy overlays communicate activity clearly enough.

6) Coverage and documentation

• Add concrete run logs for: D-13 device switching without runtime locks, S-09 keyboard layout persistence, U-11 boot-device label display.

Resolved Since BETA-10-5 (no longer blocking)

The items previously parked here as "pragmatically accepted / known-broken" are now resolved — see STATE.md > Known Issues (Recently resolved) and > Reported Hardware Status for the canonical record:

• U-10 BOOT.ELF from HDD-booted POPSLoader — RESOLVED via PR #479 (reboot_iop=0). Investigation archived under docs/archive/U10_INVESTIGATION.md.
• DKWDRV from custom HDD path — RESOLVED via PRs #486/#487 (partition-aware path + live pfs-slot scan).
• Class-A start failures (HOSDmenu / some wLaunchELF builds) — RESOLVED (maintainer-confirmed 2026-06-15).
• HDD read-write — achieved via the EnsureBootPartitionWritable boot-partition remount take-over (the launcher owns its boot pfs slot and remounts it RW), not the old ps2hdd-osd.irx → ps2hdd.irx IRX-swap probe. provato confirmed the HDD is RW-writable on real hardware; the old swap-probe item is dropped. HDD-resident settings save + in-app HDD .hide now ship on this path (still validating the full flow on hardware).

Secondary Work

1) Menu paths

• SMB (v1) network game browsing — implemented (CI + Rolling green; browse + launch HARDWARE-CONFIRMED by the maintainer, who has been launching SMB games successfully for some time on a static IP configuration). End to end: an SMB carousel page (scene GSMBNET, main-menu OPT==7) with a network SETTINGS section (server IP, share, user/password, DHCP-or-static IP assignment, port, games path/cwd, link mode — IP addressing only), a BDMA-style "SMB modules" install toggle (copies the 6-IRX in-game SMB streaming pack — poweroff/ps2dev9/ps2ip/ps2smap/smbman/SMSUTILS — into mc:/POPSTARTER and generates IPCONFIG.DAT + SMBCONFIG.DAT from the settings; OFF removes only those SMB files), lazy connect (the network stack and share open only on entering the SMB page or on a settings action — never at boot; Path B = OPL's netman recipe, dev9 shared once across HDD + SMB via g_dev9_loaded), share browse listing VCD games like any other device, POPStarter launch via argv0 selector smb:/POPS/SB.<name>.ELF, disconnect on leaving the page (also tears down a failed connect so no half-open session lingers), a blank-share GETSHARELIST picker (persisted to settings + the in-game SMBCONFIG.DAT, then reconnects), and .DAT backfill on every settings save when the pack is installed. NetBIOS is not supported (deferred — nbns.irx is OPL-custom, not stock ps2sdk; address type must be IP). Commits on BETA-13-PLAY: settings ee4d454, module toggle 121823d, build 0cf7f81, connect/browse 43033dc, launch 68f9ed5, Increment-3 polish 154c872, .DAT backfill 5d0e302, connect-failure session-leak fix f5ac26c, in-UI share picker 1169dbc. Hardware status: the argv0 device prefix (smb:/POPS/SB.<name>.ELF) and the connect handshake are confirmed — the maintainer's static-IP setup has been launching SMB games for some time. Still unconfirmed: the DHCP path and the blank-Share GETSHARELIST DMA. DHCP was broken from the start and hid behind exactly this static-only confirmation: POPStarter has no DHCP of its own, so it needs a literal address in mc?:/POPSTARTER/IPCONFIG.DAT, and SyncSmbDat deleted that file whenever IP assignment was DHCP — the shipped default. A default install therefore browsed fine on the loader's own lease and then died at the handoff (issue #560, elvengf). Fixed in bb62f2be, which writes the leased address instead of deleting the file; guarded by harness T43.
• ILINK — intentionally not implemented; surface "not supported" if entered.
• HDD (exFAT) is implemented as a mass: backend via BDMA Mode ATA (validating on hardware).

2) Art/asset behavior

• Keep current cover behavior stable, and read STATE.md > Cover art plus the EXP71/EXP73 bullets before touching it — the detail is canonical there, not here. In short: the seeker returns exactly ONE candidate (<device>:/ART/<gamefilename>_COV.png; APA/PFS hdd0:__common/POPS/ART/<gamefilename>_COV.png), the selectable folder row was removed in EXP35, and the disc-stripped family name, the POPS/ fallback and the legacy <name>.png are gone. Ignore below: the cover <name>.png (and its matching <name>.txt details sidecar, which rides the same candidate list) is built by BuildCoverCandidates (83e81cf, selectable 7f9b5fa). On removable devices the lookup folder is user-selectable (ART_LOCATION / Settings → Game List → Cover/details folder): POPS/ART (default), the game's own POPS/ folder beside the .VCD, or a top-level <device>:/ART/; the game's own POPS/ folder is always also checked as a fallback so older art never stops showing. HDD/PFS is fixed at hdd0:__common/POPS/ART/<game>.png (from the __common partition). Within each folder the disc-marker-stripped name is tried first (one art/details file serves every disc of a multi-disc game), then the exact per-disc <full-name>.png for back-compat. CoverCache:GetOrLoad loads via Graphics.loadImage/fopen with no open() existence pre-probe (ps2sdk routes open and fopen through the same _open, so the probe was redundant, not a cover fix); don't reintroduce it and don't drop the always-appended POPS/ fallback.
• The no-cover / preview-off cover box now draws the layered cover_default.png base (+ cover_missing.png overlay when a game has no cover); MISSING.png was removed and is no longer a fallback anywhere; default.png stays an optional legacy cover override. The canonical detail (asset layering, the embed-mechanism's 3 coordinated edit sites, the ELF-size delta) lives in STATE.md > Cover art — don't re-enumerate it here.

3) Install/build clarity

• Keep CI package layout and docs synchronized.
• ps2dev/ps2dev image is pinned to v2.0.0 in .github/workflows/compilation.yml and .github/workflows/rolling-release.yml (post-release pin at commit ba8f0d0).
• The embedded-Lua syntax gate (luac5.4 -p on bin/POPSLDR/*.lua + etc/boot.lua) is now LIVE — it used to silently skip because the ps2dev image shipped no luac; the workflows now apk add lua5.4 and hard-fail on a syntax error. It catches SYNTAX only — runtime nil-global / type / load-order errors stay invisible to CI (see the d4b04be load-order boot brick). See STATE.md > CI / release.
• Rolling release workflow publishes two assets — POPSLOADER-rolling-release.zip and the bare POPSLOADER.ELF — to the canonical rolling-release GitHub Release, overwriting them each time. Only a push to dev publishes; PR events build and gate but skip the publish step (gated since PR #511, 2026-07-16).
• Redistributable launcher in the zips. POPSTARTER.ELF (the redistributable POPStarter homebrew launcher — the POPS engine binaries are not redistributable and are not shipped) now ships in both zips: rolling-release.yml puts it at the zip root next to POPSLOADER.ELF and in POPS/, and also ships a POPSTARTER/ pack folder (BDMA/SMB modules) + POPS/PATCH_5.BIN at the rolling root; compilation.yml (the strict-verified PS1_POPSLOADER/ install zip) ships POPSTARTER.ELF in PS1_POPSLOADER/ (next to POPSLOADER.ELF) and POPS/.

4) Settings UI redesign

• 2026-05-19/20 OPL-style focused-list shipped (Settings page rewrite). Hardware verification deferred per the launch-path retest sequence.
• Berion-mockup-driven GUI overhaul is queued (see #5 below). The OPL focused list is intended to be replaced when that overhaul lands, so further iteration on the focused list is paused unless a specific bug appears.

5) Full GUI overhaul (Berion mockups)

• 2026-05-24: graphics-team mockups by Berion and the matching PNG asset set landed (f8fec64). Full implementation prompt and per-screen pixel specs live in docs/archive/GUI_OVERHAUL_PROMPT.md.
• Scope: Context menu, Settings (per-category pages superseding the OPL focused-list), Joypad configuration, On-screen keyboard. Boot/splash and game list are out of scope.
• Prereq: the launch-path hardware verification (DKWDRV-on-HDD, wLaunchELF, U-10) has now settled — all resolved (see STATE.md > Known Issues). The remaining gate is the 2026-06 HDD-resident settings save + in-app HDD .hide + PAL-512 features, which are still validating on hardware; the category-page Settings model in the prompt replaces the OPL focused-list, so coordinate retest sequencing once those settle.
• Mockup HTML/JSX wrapper from Berion's package is referenced by the prompt but not yet committed; either commit the mockup files or use a screenshot/hosted-mockup oracle before starting the Lua port.

6) Documentation cleanup (per docs/archive/DOCUMENTATION_FOLLOWUP_AUDIT.md) — DONE 2026-06-17

• Completed 2026-06-17. The doc set was consolidated: the volatile status facts were merged into the canonical STATE.md (including the former TRUTHSHEET.md), the obsolete/superseded docs were archived under docs/archive/, and the doc set was de-duplicated so each fact lives in exactly one place. The original three-PR plan (source-of-truth sync, agent/handoff cleanup, architecture/component/release polish) is folded into this consolidation.
• Was out of scope (and stayed so): any change to runtime code or to CI/build/release workflows; this was doc-only.

Deferred Ideas

• Additional themes/skins.
• Broader network/backend support building on the SMB (v1) baseline (implemented; browse + launch hardware-confirmed on a static IP config) and once ILINK has a defined baseline.
• More ambitious artwork cache policy after current launch/runtime issues are stable.