?yieldtrace shows the crash occurs with zero compiled-TB DONE_FLAG resume
re-entries, so it is not the store block-replay path. Combined with
nondeterminism and identity-independence, the corruption is a guest-RAM data
race with concurrent device/timer callbacks. Blanket store-TCI stays default.
E15/E16: runtime store-gating bisection plus repeated-run testing show the
0x4201e210 reader-open corruption is a race (identity-independent, config
alternates crash/pass), not a miscompiled TB. Blanket store-TCI suppresses it
in practice. Wire ?sthash and ?stcw/stcmask/stcval opt-in debug controls.
Default browser path now keeps only 32-bit qemu store TBs on TCI. Normal raw-SD
reader-open passes, while ?nostoretci still reproduces the original
0x4201e210 load fault as a negative control.
Add ?nostoretci and ?tcgdiff2=open browser-test plumbing. ?nostoretci confirms
the original compiled-store corruption still reproduces at normal speed; the
file-open-armed E11 run catches stale register state before the known
0x4201e210 load fault. Production path remains store-TCI and browser raw-SD
stays green.
E11 proved inline compiled stores are correct under live cross-TB accumulation
(153k comparisons, zero divergence) but is too slow to reach the reader
corruption in-browser (guest watchdog fires first). Suspects narrowed to the
store slow-path and store-width interaction; next instrument is a deterministic
slow-path store unit test. Option A (deterministic two-pass) retained as future.
E10 per-TB compiled-vs-TCI differential compared 160k+ store TBs during a
full reader-open with zero divergence, exonerating single-TB store codegen
and confirming the inflate corruption is emergent across TBs. Harness is
opt-in via ?tcgdiff (wasm_diff_enable from app.js preRun); production
store-TCI path is unchanged.
icount throttled TCG and starved the guest watchdog. Measured book open dropped from 146s to 31s. Off by default; ?icount restores it, ?noicount removed in favor of the new default. Harness now reports open-file timing.
Add directory navigation, folder creation, inline deletion, clear controls, and a focused editor view. Include the opt-in FAT16 vvfat experiment and fail-fast browser diagnostics; raw FAT32 remains the default.
Browser SD files are persisted in OPFS (web/sdstore.js) and edited in the
Files tab. At boot the browser builds a deterministic 64 MiB FAT32 image
(web/fat32.js) with nested directories and VFAT long names, then mounts
it as a raw block device, bypassing QEMU's broken populated-vvfat WASM
coroutine path. Guest writes mutate only the ephemeral image.
Exposes a hidden #frame-generation indicator that increments on every
e-ink flush, so automation can wait on real display updates instead of
fixed sleeps. Adds tests/ with a Makefile runner: 'make test' for JS +
FAT32/mtools checks, 'make browser-test' for a headless-Firefox boot of
the official CrossPoint 1.4.1 firmware that navigates to
Test/example.txt. Bumps third_party/qemu to the SPI-SD/CMD13 fixes and
updates README/ROADMAP/ISSUES for the new architecture.
Add a log panel that captures QEMU stdout/stderr, route serial to stdio, and
pass -d guest_errors,unimp so device warnings (e.g. the m25p80 Read-ID gap that
stalls CrossPoint boot) are visible in the browser. Read the framebuffer via
HEAPU8 to match the validated node path.
Accept the ~5 MB release firmware.bin directly: place it at 0x10000 in a
0xFF-erased 16 MB image alongside the bundled ESP32-C3 bootloader (0x0) and
CrossPoint partition table (0x8000). m25p80 requires a full chip-size backing,
so the 16 MB image is still built, but the user no longer pre-merges. Full
16 MB images are still accepted as-is.
evan