Crash abort 1.4.0 RC in manage fonts

[paulandrewbowers-cloud]

Device

X4

Affected Version

1.4.0 rc

Build Variant

tiny

EPUB Optimization

No / Not Applicable

Cache State

No / Not Applicable

Describe the Bug

Multiple crash aborts . Manage fonts getting font list. Manage fonts update all. Update individually looks ok

Steps to Reproduce

Settings, manage fonts.

Expected Behavior

No crash

crash_report.txt

Relevant Log Output/Screenshots

[pillarsdotnet]

Confirming this also happens on X3, firmware-tiny-v1.4.0-83de648-RC.bin (tag v1.4.0-rc, commit 83de6488). Same abort() panic the instant Settings > Fonts > Manage Fonts finishes connecting to WiFi and starts loading the font list.

The trigger on my side appears to be having a lot of SD-card fonts installed — 24 families × 8 point sizes (192 .cpfont files) under /.fonts. This lines up with @paulandrewbowers-cloud's observation that "getting font list" and "Update all" crash while "Update individually looks ok": the individual-download path frees the family list before doing its work, so it peaks at lower memory.

Root cause (my reading of the source at 83de6488)

FontDownloadActivity::onEnter() calls sdFontSystem.releaseLoadedFont(renderer), which only unloads the active rasterized glyph font. It does not free the SD-card font registry. The codebase already has SdCardFontSystem::releaseForNetwork(), which unloads the active font and clears the registry (registry_.clear()), and it's used before network work in KOReaderSyncActivity and KOReaderAuthActivity — but not here.

So during Manage Fonts, memory peaks with all of these live at once:

1. the full in-RAM font registry (192 file entries, each with a path std::string),
2. the parsed manifest (ArduinoJson JsonDocument for the 24-family fonts.json), and
3. the families_ vector (24 ManifestFamily, each with several std::strings + 8 ManifestFiles).

Heap is already very low at that moment — the WiFi log line shows heap=49844 maxAlloc=45044 (largest allocatable block ≈ 45 KB). An allocation in the manifest-build loop fails, and because exceptions are disabled a non-nothrow operator new calls abort() instead of failing gracefully (the failure mode called out in AGENTS.md rule #9).

The stack dump in crash_report.txt supports this. It contains half-built manifest strings (Merriweather, Sans-serif, Latin, Cyrillic, NotoSansExtended_10, ...xt_9) and the word 0x00085398 (= 545688), which is the exact byte size of NotoSansExtended_10.cpfont on my SD card — i.e. the abort fired mid-loop while accumulating family.totalSize / pushing ManifestFile entries.

Note 2d117df0 ("free font catalog before SD font downloads; fixes #236") frees families_ before the per-file download, but not before the manifest parse/list build, which is where this crash happens.

Suggested fix

In FontDownloadActivity::onEnter(), call sdFontSystem.releaseForNetwork(renderer) instead of releaseLoadedFont(renderer), matching the existing KOReader pattern, so the registry is freed before the manifest is fetched and parsed. It would also help to build the manifest list with nothrow allocations and surface a "Not enough memory" message rather than aborting.

crash_report.txt (X3)

CrossInk version: v1.4.0-RC-tiny
CrossInk variant: tiny

Panic reason: abort() was called at PC 0x421c7f87 on core 0

Last logs:
[671915] [INF] [WIFI] Auto-connect candidate: ssid=Bobbalisa saved=1
[671916] [INF] [WIFI] Connecting to ssid=Bobbalisa auto=1 saved=1 encrypted=1 passProvided=1 heap=49844 maxAlloc=45044
[672018] [INF] [WIFI] WiFi.begin returned status=6/DISCONNECTED
[672030] [INF] [WIFI] Connection poll: elapsed=114ms status=6/DISCONNECTED rssi=0
[672131] [INF] [WIFI] STA event: connected to AP
[672133] [INF] [WIFI] Connection poll: elapsed=217ms status=0/IDLE rssi=0
[673651] [INF] [WIFI] STA event: got IP 10.0.0.222
[673658] [INF] [WIFI] Connection poll: elapsed=1742ms status=3/CONNECTED rssi=-56
[673659] [INF] [WIFI] Connected to ssid=Bobbalisa ip=10.0.0.222 rssi=-56
[276] [INF] [BOOT] Reset diagnostic: reset=4(PANIC) sleepWake=0(UNDEFINED)
[277] [INF] [HW] Using cached device type: X3
[279] [INF] [GYR] QMI8658 IMU found at 0x6B
[279] [INF] [GYR] QMI8658 gyro initialized and put to sleep
[280] [INF] [CLK] DS3231 RTC found
[280] [INF] [MAIN] Hardware detect: X3
[281] [INF] [BOOT] Post-GPIO diagnostic: device=X3 usb=0 silentReboot=0 silentTarget=0


Stack memory:
0x3FCB2900: 0x3C1FDBFC 0x00000000 0x3FCB2918 0x4038DED4 0x00000000 0x00000000 0x3FCA0030 0x3FC94328 
0x3FCB2920: 0x3FCB292C 0x3FC94344 0x3FCB2918 0x63313234 0x37386637 0x00000000 0x726F6261 0x20292874 
0x3FCB2940: 0x20736177 0x6C6C6163 0x61206465 0x43502074 0x34783020 0x37633132 0x20373866 0x63206E6F 
0x3FCB2960: 0x2065726F 0x00000030 0x3FCA0000 0x0BFDD90A 0x3FCB2B38 0x00000000 0x00000064 0x421C7F8A 
0x3FCB2980: 0x00001800 0x00000000 0x00000064 0x421C7FD6 0x3FCB2B38 0x00000000 0x00000064 0x421C75CA 
0x3FCB29A0: 0x3FCB2B38 0x00000000 0x0000001F 0x421C8224 0x00000001 0x3FCB2AA8 0x0000001E 0x4201E3C4 
0x3FCB29C0: 0x00000004 0x00000002 0x3FCA9CC0 0x3C1FE000 0x00000001 0x3C21C000 0x3FCD6458 0x00000000 
0x3FCB29E0: 0x00000001 0x0000001A 0x3FCB2B38 0x4201E5B6 0x0000001A 0x00000000 0x00000000 0x3FCD6844 
0x3FCB2A00: 0x00000001 0x3FCD6844 0x0000001A 0x3FCB2A60 0x00000001 0x00000004 0x3FCC5318 0x420D93C4 
0x3FCB2A20: 0x00000000 0x3FCB2AA8 0x05555555 0x3FCC5424 0x3C202ED4 0xAAAAAAAB 0x50001F28 0x50001D00 
0x3FCB2A40: 0xFFFFFFFF 0x3FCB30D4 0x3FCD1E94 0x3FCB2AA8 0x3FCD5FCC 0x3FCB2AA8 0x3FCD6418 0x3FCB2AA8 
0x3FCB2A60: 0x3C224DAC 0x00000000 0x00000000 0x3FCD5F8C 0x04BD0476 0x3FCB2AA8 0x00000000 0xFFFFFFFF 
0x3FCB2A80: 0x3FCB2AA8 0x3FCD6458 0x0493048C 0x3FCB2AA8 0x3FCB2BEF 0x3FCD6458 0x3FCB2AA8 0x3C202ED4 
0x3FCB2AA0: 0x00000004 0x00000001 0x3FC90628 0x00000000 0x3FCD893C 0x00800080 0x3FCD1E6C 0x00800080 
0x3FCB2AC0: 0x3FCD22B0 0x00800080 0x3FCD272C 0x00800080 0x3FCD2F88 0x3FCD5B48 0x000E000E 0x4201FFFF 
0x3FCB2AE0: 0x00050000 0xFFFF0020 0x00000000 0x00000000 0x00000000 0x00000000 0x00000000 0x00000000 
0x3FCB2B00: 0x00000000 0x00000000 0x3FCD0C2C 0x0000004A 0x0000004A 0x00000001 0x3FCB2B80 0x42089ADC 
0x3FCB2B20: 0x3FCD10A4 0x00000013 0x00000013 0x3FCD0C64 0x3FCA9000 0x00000001 0x3FCB2B40 0x00000000 
0x3FCB2B40: 0x00000000 0x632E395F 0x6E6F6670 0x00000074 0x00000000 0x00000000 0x73696C00 0x50001CF8 
0x3FCB2B60: 0x00000010 0x0000001E 0x61657769 0x72656874 0x00737500 0x3FCB2B7C 0x00000000 0x72726500 
0x3FCB2B80: 0x61657769 0x72656874 0x00737500 0x3FCB2B94 0x0000000A 0x736E6153 0x7265732D 0x3F006669 
0x3FCB2BA0: 0x00000000 0x50001EAC 0x00000034 0x00000034 0x43202C6E 0x6C697279 0x0063696C 0x50001F38 
0x3FCB2BC0: 0x50001F98 0x50001F98 0x50001EF8 0x50001F1C 0x50001F1C 0x00085398 0x00000000 0x6F746F4E 
0x3FCB2BE0: 0x736E6153 0x65747845 0x6465646E 0x0030315F 0x74780000 0x0000395F 0x30373531 0x33393831 
0x3FCB2C00: 0x3F003132 0xFFFFFFFF 0x00000001 0x00000001 0x3FCA9000 0x3FCA9000 0x00000000 0x600C2000 
0x3FCB2C20: 0x00000001 0x0000000A 0x00000001 0x00000000 0x3FCB32BC 0x00000001 0xFFFFFFFF 0x3FCB2CF0 
0x3FCB2C40: 0x3FCB2CA0 0x00000000 0x00000000 0x00000000 0x00000000 0x3FCA9000 0x3FCA9000 0x0BFDD90A 
0x3FCB2C60: 0x00000000 0x00000000 0x3FCAA864 0x00000000 0x00000000 0x00000000 0x3FCB2CA0 0x3FCB2CF0 
0x3FCB2C80: 0xFFFFFFFF 0x00000001 0x3FCA2D00 0x00000003 0x3FCA9000 0x00000002 0x3FCC5318 0x420D976A 
0x3FCB2CA0: 0x3FCA9000 0x00000002 0x3FCB2CD4 0x4208A801 0x3FCA9000 0x00000002 0x3FCA2CE8 0x4208B786 
0x3FCB2CC0: 0x00001000 0x00000000 0x00000000 0x00000000 0x00000001 0x00000000 0x3FCB2CF0 0x3FCC5318 
0x3FCB2CE0: 0x00000000 0x420D6066 0x420D979A 0x40387900 0x3FCB2D01 0x3FCB2CFC 0x00000009 0x62626F42

[pillarsdotnet]

hA hardware-verified fix is ready on a fork branch. I couldn't open a PR (this repo limits PRs to collaborators), so the branch is here for you to pull or cherry-pick:

• Branch: pillarsdotnet/CrossInk@development
• Commit: 23aba785 — fix: prevent Manage Fonts OOM crash with many SD fonts installed
• One-click compare (opens a PR form on your side): development...pillarsdotnet:CrossInk:fix/344

Verified: flashed to an X3 (tiny) with 24 SD-card font families installed — Manage Fonts now loads the Font Browser instead of aborting.

Two changes, both in FontDownloadActivity: (1) onEnter() uses releaseForNetwork() instead of releaseLoadedFont() so the font registry is freed before the network work; (2) fetchAndParseManifest() parses in two passes so the parsed manifest JSON and the registry are never resident at once — that coexistence was the abort. Details in my earlier comment above.

[pillarsdotnet]

For visibility: I opened an upstream PR for the same crash against crosspoint-reader:develop — crosspoint-reader#2553.

Same mechanism as here (bad_alloc → __terminate during font-manifest parse under low heap). Upstream had no font/registry release path, so that PR adds SdCardFontRegistry::clear() + SdCardFontSystem::releaseForNetwork() and then applies the same fix used on this fork (release the active font + registry before the network work, and parse the manifest in two passes so the JSON and the registry aren't resident at once).

Verified on an X3 with a 3-way hardware A/B (unfixed → aborts; two-pass alone → still aborts because the registry is loaded at boot and never freed; release + two-pass → loads the font browser).

The CrossInk-side fix for this issue remains on pillarsdotnet/CrossInk@development (linked in my earlier comment).

[pillarsdotnet]

@uxjulia

It looks like you've merged my changes already; thanks!