feat(telemetry): crash reports from the game's own window, and performance statistics #983

Open
gab9281 wants to merge 29 commits from gab9281/Longshot:feat/telemetry into main
Member

Type: feat · Risk: medium · Tested on: Linux (CachyOS, KDE Wayland), RTX 3060 Laptop and Radeon 680M · Needs: Windows, any GPU

Why

When the game crashes or runs badly, the team only hears about it if a player finds the log and files an issue by hand, and most don't. This lets a player send a crash report from the game's own crash window with one click, and opt in to performance statistics so the team can see where Longshot stutters, on which hardware. Nothing is sent without the player's click or tick. See #44 (local problem reports, which this sends on from) and #1058 (the SDL message box's cut-off button: that box is now only the fallback, so the cut can still appear there).

What changed

The crash window is now the game's own: it shows the report and sends it itself, so any build can send one, not only a launcher-made one. The game also counts its frames for the opt-in performance summary, which nx2/perf_report.py sends. Both go to Longshot-api at data-services.magicbean.dev. The launcher's side (MagicBeanDev/LongshotLauncher#128) is the statistics tick, the Send crash reports automatically tick, and build reports.

  • Crash window (gpu/vk/vk_crash_window.cpp, vk_ui.h): one window after a crash, with Report automatically and Report manually, which survives an out-of-memory end. It is drawn as the Insert menu (F1) is: the banner, the Libertinus fonts, the gold buttons. Dear ImGui lays it out, and SDL's software renderer draws it through the SDL the crash helper already loads (SDL3 or SDL2), so it never touches Vulkan or the GPU's driver, where the crash may have been. On Wayland it asks SDL for X11 (XWayland), since SDL's Wayland driver has no software framebuffer. SDL's message box is the fallback.
  • Crash report (host/hos_crash_dialog.c): built in C from the run's log: the crash's lines without the game's memory rows, the 150 lines before it and every setting, with the home folder, user name and device names taken out. It is posted to /telemetry/crash with the system's curl, only on Send: https only, to the address make_game.py stamps from nx2/addresses.ini (crash_reports=, crash_issue=), with no redirect followed. nx2/crash_report.py and the launcher's --crash-* commands are gone. The report also carries the renderer and the run's settings (#729), and main's cpu_name() goes under cpu_text().
  • Send crash reports automatically (5398732c): when the launcher's tick has written longshot_crash.ini (send=1) in the game folder, the crash helper posts the report that Report automatically would post, with no window (send_quietly, hos_crash_auto_send). The answer goes to the log, and the report isn't sent again. If it can't be sent (the server is busy or not reached, or the report is refused), the window shows as it does today. Without the file, the window asks first.
  • Performance statistics (gpu/vk/perf_report.cpp, fps_stats.h, vk_fps.h, nx2/perf_report.py): the game counts frame and GPU times, hitches by cause, pacing, renderer events, movies and memory per map. Every 5 minutes and at exit it writes a summary to perf_pending/, and nx2/perf_report.py sends them at the next start with backoff, switches and optional copies. The host gives perf_info to the renderer: MKWVK_API_VERSION 25, after main's playback callbacks (24).
  • Addresses (nx2/addresses.ini, docs): Longshot-api as data-services.magicbean.dev.
  • Docs: docs/performance-statistics.md, and the new switches in docs/development.md.
  • Guide screenshot review (docs/images/guide/): this PR touches ten of the Graphics illustration's reviewed menu sources. Rendered from this branch and from main 02903b04, the fixture differs only in its version caption, so the PNG stays and the reviewed hashes advance (source-review-telemetry.md). Main's own Quality preset help text has drifted since the capture; the review notes it for a recapture.

What is sent

Both go to https://data-services.magicbean.dev with no account, ID or path. Longshot-api refuses a payload that doesn't fit (400, with the reason) and caps each field as the tables say.

Endpoint When Needs Becomes
POST /telemetry/crash When the player presses Send in the crash window The player's click A comment on #81, one per crash, then a count on it
POST /telemetry/perf Every 5 minutes of play and at exit, sent at the next start Send performance statistics ticked Totals on the dashboards; nothing kept per player

Crash report (/telemetry/crash, at most 128 KiB)

Field Type Required Limits What it is
crash array of string yes The crash handler's lines; the first is CRASH: exception <code> (<read|write> ...) at <module+offset>, which with the version names the crash.
version string yes ≤ 40 characters The game's version.
cpu string ≤ 200 characters The CPU, its threads and the RAM.
gpu string ≤ 200 characters The GPU and its driver.
log_tail array of string The log's last lines.
map string ≤ 120 characters The map played when it crashed: the scene and Link's age.
options string ≤ 4000 characters The game's settings, one per line: shown folded under the crash's comment.
os string ≤ 200 characters The operating system and its version.
release boolean From an actual release (the builder's signed source fingerprint); anything else is posted under [not a release].
renderer string ≤ 120 characters The renderer the game ran with.

Example:

{"version": "2.4.1", "release": true, "os": "Arch Linux (Linux 7.2.15)", "renderer": "Vulkan",
 "gpu": "AMD Radeon RX 7800 XT, RADV 26.0", "cpu": "AMD Ryzen 7 7800X3D (16 threads), 32 GB",
 "map": "Kokiri Forest (child)",
 "crash": ["CRASH: exception 0xc0000005 (read 0x8, guest 0x8) at mkw.exe+0x1a2b3c", "CRASH: thread Main, sp 0x1000"],
 "log_tail": ["vk: status: 3.1 of 15.9 GB"], "options": "upscaler=FSR 3\npreset=High"}

Performance summary (/telemetry/perf)

Field Type Required Limits What it is
v number yes always 1 The summary's format.
version string yes ≤ 40 characters The game's version.
cpu string ≤ 120 characters The CPU (the relay works out its vendor).
driver string ≤ 60 characters The GPU driver's version (the relay keeps major.minor).
first boolean The session's first summary.
fps_cap integer or null ≥ 0 0, null or false: no cap.
gpu string ≤ 120 characters The GPU (the relay works out its vendor).
kernel string ≤ 60 characters The kernel's version (the relay keeps major.minor).
map string ≤ 120 characters The map played when the summary was made.
maps object of MapSummary by key at most 16 keys The maps played since the last summary, by name.
memory object vram_used_gb, vram_allowed_gb, ram_used_gb: a sample.
os string ≤ 60 characters The operating system.
preset string ≤ 40 characters The graphics preset.
ram_gb number ≥ 0 The PC's memory, GB (the relay keeps a class).
release boolean A release's version built from the release's own sources.
renderer string ≤ 40 characters The renderer.
session_s number ≥ 0 Seconds since the session started.
upscaler string ≤ 40 characters The upscaler in use.
vram_gb number ≥ 0 The GPU's memory, GB (the relay keeps a class).
Per map (MapSummary)
Field Type Required Limits What it is
draws_left_out number ≥ 0 Draws the renderer left out.
events object keys device_lost, dlss_recreated, movie_upscaler_dropped, program_failed, swapchain_recreated, vram_over_budget Renderer events.
frame_ms array of number Frame counts in 13 buckets: up to 4.17, 6.94, 8.33, 11.1, 16.7, 20, 25, 33.3, 50, 100, 250, 1000 ms, then the rest.
frame_ms_sum number The frame times' sum, ms.
gpu_ms array of number Frame counts in 11 buckets: up to 2, 4, 6, 8.33, 11.1, 16.7, 25, 33.3, 50, 100 ms, then the rest.
gpu_ms_sum number The GPU times' sum, ms.
hitches object keys game_cpu, gpu_bound, memsync, other, pace, pipeline_wait, readback, shader_translate, texture_upload Hitches by what they waited on most.
movie object keys decode_ms, dropped, fallbacks, late, shown Movie playback: pictures shown, dropped and late, decoding ms, decoders that didn't start.
new boolean First summary of this map in the session.
pacing_misses number ≥ 0 Frames over a quarter of the target off the one before.
parts_ms object keys game, gpu_wait, pace, render Where frame time went, ms: the FPS counter's parts.
seconds number ≥ 0 Seconds played on it.

Example:

{"v": 1, "version": "2.4.1", "release": true, "first": true, "session_s": 312,
 "os": "Windows 11", "kernel": "10.0.26100", "renderer": "Vulkan",
 "gpu": "AMD Radeon RX 7800 XT", "driver": "25.10.1", "vram_gb": 16, "cpu": "AMD Ryzen 7 7800X3D", "ram_gb": 32,
 "upscaler": "FSR 3", "preset": "High", "fps_cap": 60, "map": "Kokiri Forest (child)",
 "memory": {"vram_used_gb": 5.2, "vram_allowed_gb": 15.1, "ram_used_gb": 7.9},
 "maps": {"Kokiri Forest (child)": {"new": true, "seconds": 250,
   "frame_ms": [0, 0, 0, 0, 14800, 120, 40, 10, 4, 1, 0, 0, 0], "frame_ms_sum": 250010.5,
   "gpu_ms": [0, 2100, 9800, 2900, 150, 25, 0, 0, 0, 0, 0], "gpu_ms_sum": 81200,
   "parts_ms": {"game": 98000, "render": 52000, "gpu_wait": 21000, "pace": 79000},
   "hitches": {"pipeline_wait": 3, "texture_upload": 1}, "pacing_misses": 12, "draws_left_out": 40,
   "events": {"swapchain_recreated": 1},
   "movie": {"shown": 0, "dropped": 0, "late": 0, "decode_ms": 0, "fallbacks": 0}}}}

The crash window

Drawn offscreen with this branch's code (the same through SDL3 and SDL2):

The crash window

Send to the team

Sent

Measurements

Verdict: No cost found. With the statistics off or on, the start's frame rate stays within main's own spread: 127.0 fps on main, 124.4 off and 122.9 on, while each side's three runs spread wider than those differences. add_frame() costs 16 ns a frame.

Hardware: Linux (CachyOS, kernel 7.3.4), KDE Wayland, RTX 3060 Laptop 6 GB (NVIDIA 620.70.06), Ryzen 7 6800H, 15 GB
Scene: the start (intro and title), 2048x1152, preset Auto (DLSS), no frame generation, uncapped (HOS_FPS_MAX=1000), HOS_AUDIO=0
Method: frames from 20 s on in a 75 s run (RX_VK_FRAME_LOG), 3 runs a side interleaved with main first each round, the median. The PC was busy and swapping, so every side got slower each round
How to reproduce: the same settings and switches with RX_VK_FRAME_LOG set, and longshot_perf.ini holding send=1 (on) or nothing (off); the runs are docs/tested/2026-11-15-perf-stats-gab9281-main-e090.md, -off-e090.md and -on-e090.md

Baseline: this PR's statistics were measured as 9981a39a, on main e090cc48. The branch has been rebased onto main 02903b04 since then, and the frame path did not change in the rebase.

Before: main @ e090cc48 After, tick off: this PR @ 9981a39a After, tick on: this PR @ 9981a39a
fps, median of 3 127.0 (136.7, 127.0, 121.5) 124.4 (129.3, 124.4, 122.1) 122.9 (127.2, 122.8, 122.9)
1% low, fps 50.7 49.9 49.8
frame time, ms 6.46 6.78 6.75
main e090cc48: Median of 3: 127.0 fps (136.7, 127.0, 121.5), 1% low 50.7, frame time 6.46 ms (run perf-stats-main-e090)
off 9981a39a:  Median of 3: 124.4 fps (129.3, 124.4, 122.1), 1% low 49.9, frame time 6.78 ms; inside main's spread (run perf-stats-off-e090)
on 9981a39a:   Median of 3: 122.9 fps (127.2, 122.8, 122.9), 1% low 49.8, frame time 6.75 ms; a summary written each run (run perf-stats-on-e090)

Flags and settings

New switches for testing, all off or at a polite default, and the player's statistics tick, which is off by default.

Flag What it is Before this PR After this PR Effect Helps / doesn't help / untested Recommendation
longshot_crash.ini send=1 The launcher's Send crash reports automatically tick, per game folder not there off After a crash, the report is sent with no window; the window shows only if it can't be sent Tested by tests/test_crash_dialog.py (posted with no window; a busy server brings the window back) Off by default: the player's choice
longshot_perf.ini send=1 The launcher's Send performance statistics tick, per game folder not there off The game writes summaries and the launcher sends them Measured: no cost on or off (above) Off by default: the player's choice
longshot_perf.ini keep_sent=0 Keep a copy of what was sent, turned off not there copies kept No copies in perf_sent/, and those kept are deleted Tested by tests/test_perf_report.py The launcher's default (copies on)
RX_VK_PERF_INTERVAL Seconds of play between two summaries (60 to 840) not there 300 Shorter or longer summaries Tested by tests/test_perf_report.py; no cost to measure Leave unset: only for testing
LONGSHOT_PERF_URL Where the sender posts, for a local relay not there perf_reports in nx2/addresses.ini Points the summaries at a test relay Tested against a local relay Leave unset: only for testing
LONGSHOT_PERF_MAX_SENDS, LONGSHOT_PERF_TIMEOUT, LONGSHOT_PERF_BACKOFF, LONGSHOT_PERF_KEEP_SENT The sender's limits: summaries a start, a request's time, the backoff, the copies not there 12, 10 s, on, the launcher's choice How much the sender sends and how long it waits Tested by tests/test_perf_report.py Leave unset: only for testing
HOS_CRASH_TEST A developer's test: the game crashes on purpose (start, or after some seconds) not there off Opens the crash window without a real crash Used to test the window end to end Off: only for testing
nx2/addresses.ini crash_reports, crash_issue, perf_reports Where reports go, stamped into a packaged build not there data-services.magicbean.dev, issue 81 Without them there is no Send Tested in a packaged build As shipped

Testing

Tested:

  • Linux (CachyOS, KDE Wayland; RTX 3060 Laptop with NVIDIA 620.70.06, Radeon 680M), from a packaged build. A real crash (HOS_CRASH_TEST) opened the themed window through XWayland. Report automatically reached Longshot-api and posted its comment on #81, and the crash was counted in Prometheus.
  • The performance statistics in the game, against the live relay (/telemetry/perf, counted in Prometheus), and the measurements above.
  • tests/test_crash_dialog.py: 19 passed. This includes a run of the helper after a crash with a stand-in for curl: the report posted to the stamp's address, its fields, Show all of it, Send, Open it, offered once, a busy server with Try again, a refusal without it, and no Send without the addresses. With longshot_crash.ini send=1, the report is posted with no window, and a busy server brings the window back with Report automatically still offered.
  • After the rebase on 02903b04: test_fps_stats, test_fps_rate, test_high_fps_fixes, test_perf_report, test_fps_overlay_order, test_crash_dialog, test_crash_record, test_quit_not_crash, test_audio_output_devices, test_doc_anchors, test_menu_env_overrides, test_switches_removed and test_guide_screenshots: passed.
  • python -I -B tests/run_all.py (Linux, 6 at once): 577 passed, 70 failed, 17 skipped. The 70 failures fail the same on main on this PC. The environment has no spirv-val, no sibling Tooling checkout, and the toolchain's packages aren't in the system paths (test_linux_preflight, test_nup_cache and test_packaged_folder, among others).

Not tested: Windows. I have no Windows PC or toolchain here, so the crash window through SDL on Windows, the Win32 paths and the curl that ships with Windows are untested, and I couldn't compile it for Windows either (no llvm-mingw here, and this PR's CI has no Windows build). The Steam Deck is untested too.

Must be tested:

  • Windows, any GPU: start a build with HOS_CRASH_TEST=start, expecting the themed crash window (not the message box) with Report automatically, and a report that reaches #81 when pressed (Test Farm: @Beanstalk windows, HOS_CRASH_TEST=start, press Report automatically)
  • Windows, any GPU: tick Send performance statistics in the launcher, play 6 minutes and start again, expecting perf_sent/ to hold the sent summary (Test Farm: @Beanstalk windows, performance statistics on, 6 minutes of play, then start again)

Risks

What could break: the crash window on Windows, or on a display without X11, where it falls back to SDL's message box. A failed send leaves the report on the clipboard and in the log folder. The statistics' sender runs at start in a process of its own, with a 10 s limit on each request.

Turning it off: HOS_CRASH_WATCH=0 turns the crash helper off. The statistics and the automatic crash reports are off unless the player ticks them. Otherwise, revert the PR.

Compatibility: MKWVK_API_VERSION goes to 25 (perf_info), so the renderer packs and the host must come from the same build, as every API bump requires. Saves, longshot_options.ini and the shader caches don't change. Launchers from before LongshotLauncher#128 still build this version; they just offer no statistics tick, and their --crash-* commands are no longer called.

Issues

See #44
See #1058

Checklist

Everyone

  • One topic.
  • The title is a Conventional Commit (type(scope): description) of the status line's type.
  • Why gives the reason plainly and links the issues that fit (the tracker searched).
  • Measurements has a verdict, or one line saying why there is nothing to measure.
  • Testing names what was tested, what wasn't, and what must still be tested, where and how.
  • Rebased on current main.
  • python -I -B tests/run_all.py passes (any failure that also fails on main is named).
  • No game files, dumps, build output or generated files.
  • third_party/ untouched (a new folder with a PROVENANCE.md instead).
  • The license is OK: GPL-3.0-or-later, nothing incompatible and nothing decompiled.

When it applies

  • AI assistance recorded with Assisted-by: / Co-Authored-By: trailers.
  • Numbers from the same PC, scene and settings, at least 2 interleaved runs each, the median reported.
  • Switches documented (docs/development.md; the README for a menu row) and shipped at their recommended value.
  • Screenshots of the same spot, captioned like the Measurements columns, for a change that shows on screen.
  • A file in docs/tested/ for each new measured run (format in docs/tested-configurations.md).
  • MKWVK_API_VERSION bumped (gpu/vk/vk_api.h's MkwVkHost or MkwVkRenderer changed).
  • A check under tests/ that reads changed source updated in this PR, not weakened.
  • Docs updated.

🤖 Generated with Claude Code

https://claude.ai/code/session_015KgSKmKkUWA2ugWzrw7GN4

**Type:** feat · **Risk:** medium · **Tested on:** Linux (CachyOS, KDE Wayland), RTX 3060 Laptop and Radeon 680M · **Needs:** Windows, any GPU ## Why When the game crashes or runs badly, the team only hears about it if a player finds the log and files an issue by hand, and most don't. This lets a player send a crash report from the game's own crash window with one click, and opt in to performance statistics so the team can see where Longshot stutters, on which hardware. Nothing is sent without the player's click or tick. See #44 (local problem reports, which this sends on from) and #1058 (the SDL message box's cut-off button: that box is now only the fallback, so the cut can still appear there). ## What changed The crash window is now the game's own: it shows the report and sends it itself, so any build can send one, not only a launcher-made one. The game also counts its frames for the opt-in performance summary, which `nx2/perf_report.py` sends. Both go to Longshot-api at `data-services.magicbean.dev`. The launcher's side (MagicBeanDev/LongshotLauncher#128) is the statistics tick, the Send crash reports automatically tick, and build reports. - **Crash window** (`gpu/vk/vk_crash_window.cpp`, `vk_ui.h`): one window after a crash, with Report automatically and Report manually, which survives an out-of-memory end. It is drawn as the Insert menu (F1) is: the banner, the Libertinus fonts, the gold buttons. Dear ImGui lays it out, and SDL's software renderer draws it through the SDL the crash helper already loads (SDL3 or SDL2), so it never touches Vulkan or the GPU's driver, where the crash may have been. On Wayland it asks SDL for X11 (XWayland), since SDL's Wayland driver has no software framebuffer. SDL's message box is the fallback. - **Crash report** (`host/hos_crash_dialog.c`): built in C from the run's log: the crash's lines without the game's memory rows, the 150 lines before it and every setting, with the home folder, user name and device names taken out. It is posted to `/telemetry/crash` with the system's `curl`, only on Send: https only, to the address `make_game.py` stamps from `nx2/addresses.ini` (`crash_reports=`, `crash_issue=`), with no redirect followed. `nx2/crash_report.py` and the launcher's `--crash-*` commands are gone. The report also carries the renderer and the run's settings (#729), and main's `cpu_name()` goes under `cpu_text()`. - **Send crash reports automatically** (5398732c): when the launcher's tick has written `longshot_crash.ini` (`send=1`) in the game folder, the crash helper posts the report that Report automatically would post, with no window (`send_quietly`, `hos_crash_auto_send`). The answer goes to the log, and the report isn't sent again. If it can't be sent (the server is busy or not reached, or the report is refused), the window shows as it does today. Without the file, the window asks first. - **Performance statistics** (`gpu/vk/perf_report.cpp`, `fps_stats.h`, `vk_fps.h`, `nx2/perf_report.py`): the game counts frame and GPU times, hitches by cause, pacing, renderer events, movies and memory per map. Every 5 minutes and at exit it writes a summary to `perf_pending/`, and `nx2/perf_report.py` sends them at the next start with backoff, switches and optional copies. The host gives `perf_info` to the renderer: `MKWVK_API_VERSION` 25, after main's playback callbacks (24). - **Addresses** (`nx2/addresses.ini`, docs): Longshot-api as `data-services.magicbean.dev`. - **Docs**: `docs/performance-statistics.md`, and the new switches in `docs/development.md`. - **Guide screenshot review** (`docs/images/guide/`): this PR touches ten of the Graphics illustration's reviewed menu sources. Rendered from this branch and from main 02903b04, the fixture differs only in its version caption, so the PNG stays and the reviewed hashes advance (`source-review-telemetry.md`). Main's own Quality preset help text has drifted since the capture; the review notes it for a recapture. ### What is sent Both go to `https://data-services.magicbean.dev` with no account, ID or path. Longshot-api refuses a payload that doesn't fit (400, with the reason) and caps each field as the tables say. | Endpoint | When | Needs | Becomes | |---|---|---|---| | `POST /telemetry/crash` | When the player presses Send in the crash window | The player's click | A comment on #81, one per crash, then a count on it | | `POST /telemetry/perf` | Every 5 minutes of play and at exit, sent at the next start | Send performance statistics ticked | Totals on the dashboards; nothing kept per player | #### Crash report (`/telemetry/crash`, at most 128 KiB) | Field | Type | Required | Limits | What it is | |---|---|---|---|---| | `crash` | array of string | yes | | The crash handler's lines; the first is CRASH: exception &lt;code&gt; (&lt;read\|write&gt; ...) at &lt;module+offset&gt;, which with the version names the crash. | | `version` | string | yes | ≤ 40 characters | The game's version. | | `cpu` | string | | ≤ 200 characters | The CPU, its threads and the RAM. | | `gpu` | string | | ≤ 200 characters | The GPU and its driver. | | `log_tail` | array of string | | | The log's last lines. | | `map` | string | | ≤ 120 characters | The map played when it crashed: the scene and Link's age. | | `options` | string | | ≤ 4000 characters | The game's settings, one per line: shown folded under the crash's comment. | | `os` | string | | ≤ 200 characters | The operating system and its version. | | `release` | boolean | | | From an actual release (the builder's signed source fingerprint); anything else is posted under [not a release]. | | `renderer` | string | | ≤ 120 characters | The renderer the game ran with. | **Example:** ```json {"version": "2.4.1", "release": true, "os": "Arch Linux (Linux 7.2.15)", "renderer": "Vulkan", "gpu": "AMD Radeon RX 7800 XT, RADV 26.0", "cpu": "AMD Ryzen 7 7800X3D (16 threads), 32 GB", "map": "Kokiri Forest (child)", "crash": ["CRASH: exception 0xc0000005 (read 0x8, guest 0x8) at mkw.exe+0x1a2b3c", "CRASH: thread Main, sp 0x1000"], "log_tail": ["vk: status: 3.1 of 15.9 GB"], "options": "upscaler=FSR 3\npreset=High"} ``` #### Performance summary (`/telemetry/perf`) | Field | Type | Required | Limits | What it is | |---|---|---|---|---| | `v` | number | yes | always 1 | The summary's format. | | `version` | string | yes | ≤ 40 characters | The game's version. | | `cpu` | string | | ≤ 120 characters | The CPU (the relay works out its vendor). | | `driver` | string | | ≤ 60 characters | The GPU driver's version (the relay keeps major.minor). | | `first` | boolean | | | The session's first summary. | | `fps_cap` | integer or null | | ≥ 0 | 0, null or false: no cap. | | `gpu` | string | | ≤ 120 characters | The GPU (the relay works out its vendor). | | `kernel` | string | | ≤ 60 characters | The kernel's version (the relay keeps major.minor). | | `map` | string | | ≤ 120 characters | The map played when the summary was made. | | `maps` | object of [MapSummary](#per-map-mapsummary) by key | | at most 16 keys | The maps played since the last summary, by name. | | `memory` | object | | | vram_used_gb, vram_allowed_gb, ram_used_gb: a sample. | | `os` | string | | ≤ 60 characters | The operating system. | | `preset` | string | | ≤ 40 characters | The graphics preset. | | `ram_gb` | number | | ≥ 0 | The PC's memory, GB (the relay keeps a class). | | `release` | boolean | | | A release's version built from the release's own sources. | | `renderer` | string | | ≤ 40 characters | The renderer. | | `session_s` | number | | ≥ 0 | Seconds since the session started. | | `upscaler` | string | | ≤ 40 characters | The upscaler in use. | | `vram_gb` | number | | ≥ 0 | The GPU's memory, GB (the relay keeps a class). | ##### Per map (`MapSummary`) | Field | Type | Required | Limits | What it is | |---|---|---|---|---| | `draws_left_out` | number | | ≥ 0 | Draws the renderer left out. | | `events` | object | | keys `device_lost`, `dlss_recreated`, `movie_upscaler_dropped`, `program_failed`, `swapchain_recreated`, `vram_over_budget` | Renderer events. | | `frame_ms` | array of number | | | Frame counts in 13 buckets: up to 4.17, 6.94, 8.33, 11.1, 16.7, 20, 25, 33.3, 50, 100, 250, 1000 ms, then the rest. | | `frame_ms_sum` | number | | | The frame times' sum, ms. | | `gpu_ms` | array of number | | | Frame counts in 11 buckets: up to 2, 4, 6, 8.33, 11.1, 16.7, 25, 33.3, 50, 100 ms, then the rest. | | `gpu_ms_sum` | number | | | The GPU times' sum, ms. | | `hitches` | object | | keys `game_cpu`, `gpu_bound`, `memsync`, `other`, `pace`, `pipeline_wait`, `readback`, `shader_translate`, `texture_upload` | Hitches by what they waited on most. | | `movie` | object | | keys `decode_ms`, `dropped`, `fallbacks`, `late`, `shown` | Movie playback: pictures shown, dropped and late, decoding ms, decoders that didn't start. | | `new` | boolean | | | First summary of this map in the session. | | `pacing_misses` | number | | ≥ 0 | Frames over a quarter of the target off the one before. | | `parts_ms` | object | | keys `game`, `gpu_wait`, `pace`, `render` | Where frame time went, ms: the FPS counter's parts. | | `seconds` | number | | ≥ 0 | Seconds played on it. | **Example:** ```json {"v": 1, "version": "2.4.1", "release": true, "first": true, "session_s": 312, "os": "Windows 11", "kernel": "10.0.26100", "renderer": "Vulkan", "gpu": "AMD Radeon RX 7800 XT", "driver": "25.10.1", "vram_gb": 16, "cpu": "AMD Ryzen 7 7800X3D", "ram_gb": 32, "upscaler": "FSR 3", "preset": "High", "fps_cap": 60, "map": "Kokiri Forest (child)", "memory": {"vram_used_gb": 5.2, "vram_allowed_gb": 15.1, "ram_used_gb": 7.9}, "maps": {"Kokiri Forest (child)": {"new": true, "seconds": 250, "frame_ms": [0, 0, 0, 0, 14800, 120, 40, 10, 4, 1, 0, 0, 0], "frame_ms_sum": 250010.5, "gpu_ms": [0, 2100, 9800, 2900, 150, 25, 0, 0, 0, 0, 0], "gpu_ms_sum": 81200, "parts_ms": {"game": 98000, "render": 52000, "gpu_wait": 21000, "pace": 79000}, "hitches": {"pipeline_wait": 3, "texture_upload": 1}, "pacing_misses": 12, "draws_left_out": 40, "events": {"swapchain_recreated": 1}, "movie": {"shown": 0, "dropped": 0, "late": 0, "decode_ms": 0, "fallbacks": 0}}}} ``` ### The crash window Drawn offscreen with this branch's code (the same through SDL3 and SDL2): ![The crash window](https://git.magicbean.dev/attachments/ceba0333-5d1d-419f-a806-68d846cc4180) ![Send to the team](https://git.magicbean.dev/attachments/e999bd60-5242-4af4-84e3-318203cc3aac) ![Sent](https://git.magicbean.dev/attachments/74941905-7c9d-4ab0-b2eb-4d02eda2ce15) ## Measurements **Verdict:** No cost found. With the statistics off or on, the start's frame rate stays within main's own spread: 127.0 fps on main, 124.4 off and 122.9 on, while each side's three runs spread wider than those differences. `add_frame()` costs 16 ns a frame. Hardware: Linux (CachyOS, kernel 7.3.4), KDE Wayland, RTX 3060 Laptop 6 GB (NVIDIA 620.70.06), Ryzen 7 6800H, 15 GB Scene: the start (intro and title), 2048x1152, preset Auto (DLSS), no frame generation, uncapped (`HOS_FPS_MAX=1000`), `HOS_AUDIO=0` Method: frames from 20 s on in a 75 s run (`RX_VK_FRAME_LOG`), 3 runs a side interleaved with main first each round, the median. The PC was busy and swapping, so every side got slower each round How to reproduce: the same settings and switches with `RX_VK_FRAME_LOG` set, and `longshot_perf.ini` holding `send=1` (on) or nothing (off); the runs are `docs/tested/2026-11-15-perf-stats-gab9281-main-e090.md`, `-off-e090.md` and `-on-e090.md` Baseline: this PR's statistics were measured as 9981a39a, on main e090cc48. The branch has been rebased onto main 02903b04 since then, and the frame path did not change in the rebase. | | Before: main @ e090cc48 | After, tick off: this PR @ 9981a39a | After, tick on: this PR @ 9981a39a | |---|---|---|---| | fps, median of 3 | 127.0 (136.7, 127.0, 121.5) | 124.4 (129.3, 124.4, 122.1) | 122.9 (127.2, 122.8, 122.9) | | 1% low, fps | 50.7 | 49.9 | 49.8 | | frame time, ms | 6.46 | 6.78 | 6.75 | ``` main e090cc48: Median of 3: 127.0 fps (136.7, 127.0, 121.5), 1% low 50.7, frame time 6.46 ms (run perf-stats-main-e090) off 9981a39a: Median of 3: 124.4 fps (129.3, 124.4, 122.1), 1% low 49.9, frame time 6.78 ms; inside main's spread (run perf-stats-off-e090) on 9981a39a: Median of 3: 122.9 fps (127.2, 122.8, 122.9), 1% low 49.8, frame time 6.75 ms; a summary written each run (run perf-stats-on-e090) ``` ### Flags and settings New switches for testing, all off or at a polite default, and the player's statistics tick, which is off by default. | Flag | What it is | Before this PR | After this PR | Effect | Helps / doesn't help / untested | Recommendation | |---|---|---|---|---|---|---| | `longshot_crash.ini` `send=1` | The launcher's Send crash reports automatically tick, per game folder | not there | off | After a crash, the report is sent with no window; the window shows only if it can't be sent | Tested by `tests/test_crash_dialog.py` (posted with no window; a busy server brings the window back) | Off by default: the player's choice | | `longshot_perf.ini` `send=1` | The launcher's Send performance statistics tick, per game folder | not there | off | The game writes summaries and the launcher sends them | Measured: no cost on or off (above) | Off by default: the player's choice | | `longshot_perf.ini` `keep_sent=0` | Keep a copy of what was sent, turned off | not there | copies kept | No copies in `perf_sent/`, and those kept are deleted | Tested by `tests/test_perf_report.py` | The launcher's default (copies on) | | `RX_VK_PERF_INTERVAL` | Seconds of play between two summaries (60 to 840) | not there | 300 | Shorter or longer summaries | Tested by `tests/test_perf_report.py`; no cost to measure | Leave unset: only for testing | | `LONGSHOT_PERF_URL` | Where the sender posts, for a local relay | not there | `perf_reports` in `nx2/addresses.ini` | Points the summaries at a test relay | Tested against a local relay | Leave unset: only for testing | | `LONGSHOT_PERF_MAX_SENDS`, `LONGSHOT_PERF_TIMEOUT`, `LONGSHOT_PERF_BACKOFF`, `LONGSHOT_PERF_KEEP_SENT` | The sender's limits: summaries a start, a request's time, the backoff, the copies | not there | 12, 10 s, on, the launcher's choice | How much the sender sends and how long it waits | Tested by `tests/test_perf_report.py` | Leave unset: only for testing | | `HOS_CRASH_TEST` | A developer's test: the game crashes on purpose (`start`, or after some seconds) | not there | off | Opens the crash window without a real crash | Used to test the window end to end | Off: only for testing | | `nx2/addresses.ini` `crash_reports`, `crash_issue`, `perf_reports` | Where reports go, stamped into a packaged build | not there | `data-services.magicbean.dev`, issue 81 | Without them there is no Send | Tested in a packaged build | As shipped | ## Testing **Tested:** - Linux (CachyOS, KDE Wayland; RTX 3060 Laptop with NVIDIA 620.70.06, Radeon 680M), from a packaged build. A real crash (`HOS_CRASH_TEST`) opened the themed window through XWayland. Report automatically reached Longshot-api and posted its comment on #81, and the crash was counted in Prometheus. - The performance statistics in the game, against the live relay (`/telemetry/perf`, counted in Prometheus), and the measurements above. - `tests/test_crash_dialog.py`: 19 passed. This includes a run of the helper after a crash with a stand-in for curl: the report posted to the stamp's address, its fields, Show all of it, Send, Open it, offered once, a busy server with Try again, a refusal without it, and no Send without the addresses. With `longshot_crash.ini` `send=1`, the report is posted with no window, and a busy server brings the window back with Report automatically still offered. - After the rebase on 02903b04: `test_fps_stats`, `test_fps_rate`, `test_high_fps_fixes`, `test_perf_report`, `test_fps_overlay_order`, `test_crash_dialog`, `test_crash_record`, `test_quit_not_crash`, `test_audio_output_devices`, `test_doc_anchors`, `test_menu_env_overrides`, `test_switches_removed` and `test_guide_screenshots`: passed. - `python -I -B tests/run_all.py` (Linux, 6 at once): 577 passed, 70 failed, 17 skipped. The 70 failures fail the same on main on this PC. The environment has no `spirv-val`, no sibling Tooling checkout, and the toolchain's packages aren't in the system paths (`test_linux_preflight`, `test_nup_cache` and `test_packaged_folder`, among others). **Not tested:** Windows. I have no Windows PC or toolchain here, so the crash window through SDL on Windows, the Win32 paths and the `curl` that ships with Windows are untested, and I couldn't compile it for Windows either (no llvm-mingw here, and this PR's CI has no Windows build). The Steam Deck is untested too. **Must be tested:** - Windows, any GPU: start a build with `HOS_CRASH_TEST=start`, expecting the themed crash window (not the message box) with Report automatically, and a report that reaches #81 when pressed (Test Farm: @Beanstalk windows, HOS_CRASH_TEST=start, press Report automatically) - Windows, any GPU: tick Send performance statistics in the launcher, play 6 minutes and start again, expecting `perf_sent/` to hold the sent summary (Test Farm: @Beanstalk windows, performance statistics on, 6 minutes of play, then start again) ## Risks **What could break:** the crash window on Windows, or on a display without X11, where it falls back to SDL's message box. A failed send leaves the report on the clipboard and in the log folder. The statistics' sender runs at start in a process of its own, with a 10 s limit on each request. **Turning it off:** `HOS_CRASH_WATCH=0` turns the crash helper off. The statistics and the automatic crash reports are off unless the player ticks them. Otherwise, revert the PR. **Compatibility:** `MKWVK_API_VERSION` goes to 25 (`perf_info`), so the renderer packs and the host must come from the same build, as every API bump requires. Saves, `longshot_options.ini` and the shader caches don't change. Launchers from before LongshotLauncher#128 still build this version; they just offer no statistics tick, and their `--crash-*` commands are no longer called. ## Issues See #44 See #1058 ## Checklist ### Everyone - [x] One topic. - [x] The title is a Conventional Commit (`type(scope): description`) of the status line's type. - [x] Why gives the reason plainly and links the issues that fit (the tracker searched). - [x] Measurements has a verdict, or one line saying why there is nothing to measure. - [x] Testing names what was tested, what wasn't, and what must still be tested, where and how. - [x] Rebased on current `main`. - [x] `python -I -B tests/run_all.py` passes (any failure that also fails on `main` is named). - [x] No game files, dumps, build output or generated files. - [x] `third_party/` untouched (a new folder with a `PROVENANCE.md` instead). - [x] The license is OK: GPL-3.0-or-later, nothing incompatible and nothing decompiled. ### When it applies - [x] AI assistance recorded with `Assisted-by:` / `Co-Authored-By:` trailers. - [x] Numbers from the same PC, scene and settings, at least 2 interleaved runs each, the median reported. - [x] Switches documented (docs/development.md; the README for a menu row) and shipped at their recommended value. - [x] Screenshots of the same spot, captioned like the Measurements columns, for a change that shows on screen. - [x] A file in `docs/tested/` for each new measured run (format in `docs/tested-configurations.md`). - [x] `MKWVK_API_VERSION` bumped (`gpu/vk/vk_api.h`'s `MkwVkHost` or `MkwVkRenderer` changed). - [x] A check under `tests/` that reads changed source updated in this PR, not weakened. - [x] Docs updated. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_015KgSKmKkUWA2ugWzrw7GN4
The crash window (#593) gets a first button for a crash in a build the Longshot
Launcher made: Send to the team. It shows the crash report it would send, made by
the launcher with the build's own nx2/crash_report.py (--crash-preview), and sends
that file as it is on Send (--crash-send), to Longshot-api's /telemetry/crash,
which posts it on #81. Sent once: the answer is kept in bug-reports/sent/.

- nx2/crash_report.py reads #75's crash records and the run's log in logs/ by its
  run_id; --preview and --send-file for the launcher. Always send is gone.
- host/hos_crash_dialog.c: the button and its boxes; spawn_exe starts another
  program, with a time limit. The pure parts are in hos_crash_dialog.h.
- nx2/make_game.py writes sender= (LONGSHOT_CRASH_SENDER, the launcher's) and
  source= (the release flag) into longshot_version.txt.
- The renderer logs report: gpu/cpu/os at startup, as the Bug report words them.
- nx2/addresses.ini and addresses.py, the same files as #206.
- tests/test_crash_dialog.py: the Send cases, a real run with a stand-in
  launcher, and three fixes to main's copy (the fixture's argv offset, a
  trailing -x none under clang -Werror, and out.txt left from the run before).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
Found running the Send flow in the game (Linux, RTX 3060 Laptop):

- The report's Renderer was "not known": the rebase dropped the
  report: renderer lines from host/rx_backend.c with the old window's
  calls. Back, with a check that they stay.
- The window still said "Longshot itself sends nothing" next to Send to
  the team. With Send offered it says what Send does and that Report
  itself sends nothing; without it, the text is as on main.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
Out of memory (Linux): the kernel's OOM killer ends the game with SIGKILL,
then systemd stops the game's whole scope (DefaultOOMPolicy=stop), the crash
window's helper with it: the player saw nothing. Seen on this PC tonight
(app-flatpak-dev.magicbean.Longshot-*.scope: Failed with result 'oom-kill').
The helper now:
- reads the game's cgroup's memory.events oom_kill at the start and the end,
  and writes "oom_kill: yes" into the unclean record: the window says the PC
  ran out of memory;
- moves itself into its own scope (StartTransientUnit through busctl) before
  it waits, so stopping the game's scope leaves it running;
- holds SIGTERM back until the record is written;
- lists an out-of-memory end as shown only after its window, so where it
  can't leave the scope (a Flatpak) the next start shows it.
Tested in the game under systemd-run -p MemoryMax=1200M -p OOMPolicy=stop:
the window right after the kill, and with busctl out of reach, at the next
start.

The launcher's window: with sender= in the stamp, the helper first starts
<sender> --crash-window <folder> <window.txt> (what happened, as words the
launcher translates, Report's saved text, the issue link, the log folder;
hos_crash_window_text), and shows its own boxes only when the launcher
answers anything but 0. report() is split so make_report() makes Report's
text and link for both.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
The crash window's two ways to report are named for what they do: Report automatically (the report
shown in full, sent through Longshot-api to #81 only when pressed) and Report manually (main's Report:
the report on the clipboard and the prefilled issue form).

- In a build the launcher made, the window is the launcher's (--crash-window); the game's boxes show
  only when the launcher can't be started or answers 3 (it can't show it), so a crash is never shown
  twice.
- The manual report holds every field Report automatically's card has: a Summary with the version,
  the release build, the system, the renderer, the GPU, the CPU and RAM, the map, the settings'
  count and changes, and the crash in one line, from the same log lines and rules as crash_report.py.
- The issue link fills the form's Channel, What kind of problem, Where in the game and Changes from
  the defaults too.
- crash_report.py --card: the preview's card as data, for the launcher's window.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
Report automatically counted and sent longshot_options.ini from the game folder,
while a launcher-started run (--user) now plays with the user folder's: the
report would have shown an old copy's settings. crash_report.py takes the file
the run's log names (options: the settings file is ..., as pipeline_share.py),
"none" as every setting at its default; count_settings reads the helper's
options argument, as put_options does. test_crash_dialog's options probe now
skips the driver's own arguments, as args does (#729's harness was one off).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKp43znbyyPyDajp3Cm
#626 repairs the check's args and options probes (argv + 2) and its
compiles; the copies here would only conflict with it.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKp43znbyyPyDajp3Cm
With send=1 in the game folder's longshot_perf.ini (the launcher's Send performance statistics tick, off by
default), the renderer adds each frame to its map's totals and writes a summary into perf_pending/ every 5
minutes and when the window closes:

- gpu/vk/perf_report.h, perf_report.cpp: frame-time and GPU-time histograms, the FPS counter's parts, hitches
  by what they waited on, pacing misses, draws left out, rendering events, movie playback, memory; the machine
  (GPU, driver, CPU, RAM, OS, kernel) and the upscaler, preset and cap.
- vk_fps.cpp: the Breakdown's timings while the summary is on; off, one load of perf_measure a frame.
- vk_main.cpp mkwvk_close_asked: the last summary (ExitProcess is _exit on Linux: atexit never runs there).
- host: the map and the movies' totals through MkwVkHost perf_info (API 22).
- tests/test_perf_report.cpp: the summary's functions; add_frame 16 ns at -O1.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
- nx2/perf_report.py: the tick (longshot_perf.ini), and sending perf_pending/ to perf_reports in
  nx2/addresses.ini (Bean-Ops' perf relay) with the version and release flag, oldest first; one the relay
  couldn't take stays for next time. Unticking deletes what is waiting.
- nx2/addresses.ini, addresses.py: the external addresses, out of the code.
- The tick on screen goes to LongshotLauncher (the builders' launchers left main in #171 and #181).
- docs/performance-statistics.md, the README, docs/development.md.
- tests/test_perf_report.py.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
- Windows: the shipped launcher (builder/LongshotBuilder) has the Send performance statistics tick and
  sends what waits, as the Linux and Python builders do.
- Every launcher sends in a process of its own (nx2/perf_report.py start_sending, --send): Play closes the
  launcher at once, and a summary the relay took was left to be sent, and counted, again.
- Unticked while the game runs: the next summary finds the tick off, and the game stops measuring and writes
  nothing more.
- At most 48 summaries wait in perf_pending, counted on disk (a player who never opens the launcher).
- A lost device's summary is written once, not once per thread that sees the loss.
- Swapchain and program-failure counts, and the first frame's waits, start from the game's counts when
  measuring starts, not from 0.
- Movie pictures are counted as the game takes them (submissions are not pictures), so a movie still
  playing at a summary or at exit counts too.
- JSON text is never cut inside a UTF-8 character; the sender reads a broken one as a replacement.
- MapTotals::any covers every movie count; dead code removed.
- The video tests' fixtures get the new counter.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
Before any map: line the summary's map was "not known". It is now
"Preparing shaders" while the start-up screen prepares them, then
"Title and intro". The checks of the launcher pages removed from main
(#171, #181) go: the tick moves to LongshotLauncher.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
- Hitch causes: a draw waiting for a program's translation (draw_stats3)
  counts as shader_translate, not pipeline_wait; program_pick's in-place
  translation is counted once (it is timed inside PROF_PROGRAM too); the
  jobs' translations behind the game no longer count at all.
- The host's map is cleared back at the title and file select, so that
  time is "Title and intro", not the last map's.
- The sender keeps the version the game wrote (the folder may have been
  updated since); a release only when it is the folder's build.
- The game folder from GetModuleFileNameW (vk_paths.h game_dir_path,
  shared with vk_dlss.cpp): any folder name works on Windows.
- atexit only tries the lock: ExitProcess may have ended its holder.
- A map entered just before a summary stays new for the next one.
- A .tmp left by an exit mid-write is removed at the next start, and
  when the tick is taken off.
- The fps_cap label uses the menu's own cap rule (menu_fps_cap).
- perf_frame takes fps_presented's draw waits instead of reading again.
- movie_report's comment is back above it.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
Each summary the relay took is written, exactly as it was posted, to
the game folder's perf_sent/ (the newest 48), so a player or a tester
can read what went to the team. `nx2/perf_report.py <folder>` says how
many are kept. Unticking deletes them with what is waiting.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
- No copy once the tick is off: unticked during a send, the player said no.
- The copy written to a .tmp first, then renamed, so one cut short is never
  taken for what was sent.
- perf_sent/ pruned by when the copies were written, not by their names' time
  (a clock that was wrong once kept the wrong ones), and once after the send.
- The copy made after the summary leaves perf_pending/: nothing between the
  relay's yes and the remove, so a summary isn't sent twice.
- listed(): one listing of a subfolder's summaries; set_sending's docstring
  names perf_sent/.
- Tests: the clock that was wrong, a copy that can't be kept, unticked
  during a send.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
The summaries sent from a laptop held at about 19 fps in Hyrule Field under
the 60 cap counted 4,694 hitches in 5,269 frames: against the cap's 16.7 ms,
every 50 ms frame was one, and the dashboard showed 34,000 hitches an hour.
A hitch is now measured against the longer of the target and the recent
average frame (before this one); slow frames still show in the fps and the
frame-time histograms. The pacing miss keeps the cap's target.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKoUqUvce5UXG89NBgi
- nx2/addresses.ini: perf_reports is https://api.magicbean.gmatte.xyz/telemetry/perf (Longshot-api's telemetry
  service replaced Bean-Ops' perf relay, and answers the new path only).
- nx2/perf_report.py reads the run's first log line in logs/ (main moved the logs there; a folder from before still
  has them at its top), and takes --tick on|off: the launcher's tick, through set_sending.
- gpu/vk/vk_api.h: perf_info is version 23 (main's 22 is the menu's take_text).
- tests/test_video_catchup.py (new on main) stubs nn_video.c's movie_totals, as the other video checks do.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
Main froze the table in docs/tested-configurations.md: the six rows the rebase put
there move to files of their own, and the A/B on main e090cc48 (v3.0.0-rc.4) is
added the same way.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
The sender (nx2/perf_report.py) sends the waiting summaries over one kept
connection (one TLS handshake for them all; through a system proxy, one request
each, as before), at most LONGSHOT_PERF_MAX_SENDS (12) a start, and leaves a busy
relay alone: until its Retry-After, else an hour doubled each time it is busy
again, up to a day (perf_pending/.next; LONGSHOT_PERF_BACKOFF=off asks it at
every start). A kept connection the relay let go is tried once more on a new
one: the summary never arrived, so none is sent twice.

- LONGSHOT_PERF_TIMEOUT (10 s, 2 to 60), LONGSHOT_PERF_URL (https, or http to
  this PC for a local relay), each clamped or defaulted with a word on stderr.
- Copies of what was sent: --tick on --keep-sent off writes keep_sent=0 (the
  launcher's Keep a copy of what was sent, LongshotLauncher#74), which keeps no
  copies and deletes those kept; LONGSHOT_PERF_KEEP_SENT overrides it.
- The game: RX_VK_PERF_INTERVAL, the seconds between summaries (300; 60 to 840,
  a minute under the relay's 900 s for the frame that ends a summary).
- docs/performance-statistics.md's Switches; the interval in development.md.
- tests: one connection, a dropped one, the share a start, Retry-After and the
  doubling, the copies and their switches, LONGSHOT_PERF_URL; the interval's
  rule in tests/test_perf_report.cpp.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmeEEMir1hCDFZ8BmR
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKp43znbyyPyDajp3Cm
fix(telemetry): Longshot-api at data-services.magicbean.dev, its magicbean.dev name
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Failing after 4s
CI / Conventional Commit title (pull_request) Successful in 3s
CI / Workflow lint (actionlint) (pull_request) Successful in 5s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 4s
Third-party parts / As their notes say (pull_request) Successful in 8s
Website / Website on Cloudflare (pull_request_target) Has been skipped
7745956aa6
The crash and performance addresses named Longshot-api's old host,
api.magicbean.gmatte.xyz. It still serves, but the team's name for it is
data-services.magicbean.dev: nx2/addresses.ini, addresses.py's example and
the two docs now use it.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Collaborator

Possibly related, for context:

  • PR #206 feat(perf): opt-in performance statistics, sent from the launcher to the team's Grafana — Earlier open performance-statistics implementation overlaps this combined game-side work.
  • PR #141 feat(crash): report a crash automatically from the crash window, or manually with every field — Earlier open crash-report implementation overlaps this combined game-side work.
Possibly related, for context: - PR #206 feat(perf): opt-in performance statistics, sent from the launcher to the team's Grafana — Earlier open performance-statistics implementation overlaps this combined game-side work. - PR #141 feat(crash): report a crash automatically from the crash window, or manually with every field — Earlier open crash-report implementation overlaps this combined game-side work.
gab9281 force-pushed feat/telemetry from 7745956aa6
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Failing after 4s
CI / Conventional Commit title (pull_request) Successful in 3s
CI / Workflow lint (actionlint) (pull_request) Successful in 5s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 4s
Third-party parts / As their notes say (pull_request) Successful in 8s
Website / Website on Cloudflare (pull_request_target) Has been skipped
to 31e529b70a
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Failing after 4s
CI / Conventional Commit title (pull_request) Successful in 3s
CI / Workflow lint (actionlint) (pull_request) Successful in 5s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 4s
Third-party parts / As their notes say (pull_request) Successful in 8s
2026-11-16 00:39:50 +00:00
Compare
feat(crash): the game's own crash window, themed as the F1 menu, sends its report itself
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 8s
PR description / Description follows the template (pull_request_target) Failing after 9s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 13s
CI / Workflow lint (actionlint) (pull_request) Successful in 15s
Third-party parts / As their notes say (pull_request) Successful in 17s
4fbf1285ed
The crash window was SDL message boxes, with Report automatically handed to the
Longshot Launcher (sender=, --crash-preview/--crash-send) and nx2/crash_report.py.
The crash belongs to the game, and a launcher-made build was the only one that
could send.

- gpu/vk/vk_crash_window.cpp: one window that goes from page to page, drawn with
  vk_ui.h (the Insert menu's banner, box, gold buttons, Libertinus) through Dear
  ImGui and SDL's renderer (SDL3 or SDL2, by name from the SDL the helper loaded),
  never Vulkan. Keys, mouse and scrolling. SDL's message box stays the fallback.
- vk_ui_fonts.h: ui::Fonts apart from vk_imgui.h, so vk_ui.h needs no Vulkan.
- host/hos_crash_dialog.c/.h: the report built in C from the run's log (the crash
  lines as the relay signs them, the 150 lines before, every setting, the home
  folder, user and device names out) and posted with the system's curl (https
  only, the stamp's address, no redirect) only when Send is pressed.
- nx2/make_game.py stamps crash_reports= and crash_issue= from nx2/addresses.ini
  instead of sender=. nx2/crash_report.py and its test are gone.
- tests/test_crash_dialog.py: the new helpers, and a run with a stand-in curl
  (sent, busy, refused, no addresses).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
gab9281 force-pushed feat/telemetry from 4fbf1285ed
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 8s
PR description / Description follows the template (pull_request_target) Failing after 9s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 13s
CI / Workflow lint (actionlint) (pull_request) Successful in 15s
Third-party parts / As their notes say (pull_request) Successful in 17s
to aa40715893
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 6s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 12s
CI / Workflow lint (actionlint) (pull_request) Successful in 13s
Third-party parts / As their notes say (pull_request) Successful in 16s
PR description / Description follows the template (pull_request_target) Failing after 5s
2026-11-16 01:27:06 +00:00
Compare
gab9281 changed title from feat(telemetry): crash reports and performance statistics, to data-services.magicbean.dev to feat(telemetry): crash reports from the game's own window, and performance statistics 2026-11-16 01:27:43 +00:00
fix(perf): names for XeSS and XeSS AA, main's two new upscalers
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 6s
PR description / Description follows the template (pull_request_target) Failing after 8s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 9s
CI / Workflow lint (actionlint) (pull_request) Successful in 10s
Third-party parts / As their notes say (pull_request) Successful in 12s
36f416e1dd
perf_report.cpp's static_assert stopped the build (7 names for 9 upscalers).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
fix(crash): the crash window over X11 (XWayland) on Wayland, where SDL has no software renderer
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 6s
PR description / Description follows the template (pull_request_target) Failing after 8s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 10s
CI / Workflow lint (actionlint) (pull_request) Successful in 10s
Third-party parts / As their notes say (pull_request) Successful in 13s
feebf14afd
Found building this branch and crashing it on purpose on KDE Plasma (Wayland): SDL3's Wayland driver has no window
framebuffer of its own, so with SDL_FRAMEBUFFER_ACCELERATION=0 the software renderer can't be made ("Window framebuffer
support not available") and the helper fell back to SDL's message box. X11's driver has one: the window asks for X11
where DISPLAY is set, Wayland when X11 can't be had, and SDL_VIDEO_DRIVER in the environment still chooses.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
gab9281 force-pushed feat/telemetry from feebf14afd
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 6s
PR description / Description follows the template (pull_request_target) Failing after 8s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 10s
CI / Workflow lint (actionlint) (pull_request) Successful in 10s
Third-party parts / As their notes say (pull_request) Successful in 13s
to 43779925d4
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 6s
PR description / Description follows the template (pull_request_target) Failing after 8s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 11s
CI / Workflow lint (actionlint) (pull_request) Successful in 11s
Third-party parts / As their notes say (pull_request) Successful in 13s
2026-11-16 06:02:35 +00:00
Compare
gab9281 force-pushed feat/telemetry from 43779925d4
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 6s
PR description / Description follows the template (pull_request_target) Failing after 8s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 11s
CI / Workflow lint (actionlint) (pull_request) Successful in 11s
Third-party parts / As their notes say (pull_request) Successful in 13s
to e0b038be6e
Some checks failed
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 6s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 10s
CI / Workflow lint (actionlint) (pull_request) Successful in 12s
Third-party parts / As their notes say (pull_request) Successful in 14s
PR description / Description follows the template (pull_request_target) Successful in 4s
2026-11-16 06:30:16 +00:00
Compare
docs(guides): the Graphics illustration reviewed for the crash window and statistics
All checks were successful
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 6s
PR description / Description follows the template (pull_request_target) Successful in 8s
CI / Workflow lint (actionlint) (pull_request) Successful in 11s
Guide screenshots / Guide images and local UI source review (pull_request) Successful in 12s
Third-party parts / As their notes say (pull_request) Successful in 14s
03203a8c72
The guide-screenshot gate asks for a review when the menu's shared sources change, and this change touches ten of
them. The fixture rendered from this branch and from main 02903b04 differ only in the version caption, so the
upstream PNG and its capture provenance stay; the reviewed hashes and their revision advance, with the review in
source-review-telemetry.md. Main's own Quality preset help text has drifted since the capture: noted there for a
recapture, not done here.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmKkUWA2ugWzrw7GN4
docs(crash): the crash dialog's opening comment says what it does now
All checks were successful
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 7s
CI / Workflow lint (actionlint) (pull_request) Successful in 11s
Guide screenshots / Guide images and local UI source review (pull_request) Successful in 11s
Third-party parts / As their notes say (pull_request) Successful in 14s
PR description / Description follows the template (pull_request_target) Successful in 5s
3ffc75d72d
The file's description still had the message box as the window and said the game sends nothing. It names the crash
window, drawn as the Insert menu is with the message box as its fallback, and Report automatically, which posts to
Longshot-api only when the player presses Send, as host/hos_crash_dialog.h already says.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmKkUWA2ugWzrw7GN4
feat(crash): send crash reports automatically, without the window, when the player chose it
All checks were successful
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 8s
CI / Workflow lint (actionlint) (pull_request) Successful in 10s
Guide screenshots / Guide images and local UI source review (pull_request) Successful in 11s
Third-party parts / As their notes say (pull_request) Successful in 12s
PR description / Description follows the template (pull_request_target) Successful in 5s
5398732c12
A player who ticks the launcher's "Send crash reports automatically" has longshot_crash.ini (send=1) in each game
folder. The crash helper then makes the report Report automatically makes and posts it the same way, with no window,
and logs the answer. A report that can't be sent (server busy or not reached, or refused) shows the window as always,
so nothing is lost; without the file, the window asks as before.

tests/test_crash_dialog.py: the file's rule (hos_crash_auto_send), a report posted with no window, and a busy server
bringing the window back with Report automatically still offered.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_015KgSKmKkUWA2ugWzrw7GN4
HumanBean left a comment

Review: request changes

What this does for a player: after a crash, the game's own themed window can send a crash report to the team in two clicks, or with no window at all when the launcher's "Send crash reports automatically" box is ticked. The game can also count frame times, hitches and movie playback per map for the launcher's opt-in "Send performance statistics". I read the whole diff (code only, nothing run). The design holds up. Every network path is opt-in. The perf side stays off and writes nothing without send=1, and the tick off deletes what is waiting. The crash post is https only, posts to the stamp's address only, and the Sent link can only go to issue 81. The blocking points below are mostly small.

Blocking

  1. The PR conflicts with main, and the API version it takes is gone. Main took MKWVK_API_VERSION 25 for mic_test (#1102). After the rebase, perf_info must come after mic_test as version 26. Six files conflict: gpu/vk/vk_api.h, host/rx_backend.c, gpu/vk/vk_dlss.cpp (two hunks, an include and a log line), tests/test_audio_output_devices.py (three hunks), docs/images/guide/README.md and docs/images/guide/manifest.json (four hunks). The text conflicts are easy, but two need decisions. The version number has to change. The guide manifest's reviewed-source hashes have to be worked out again against main's new Graphics capture (main recaptured it for the microphone test). In the same rebase, keep main's exact checks in tests/test_audio_output_devices.py. This branch loosened them from == 24 to >= 24, and from fields[-6:] to an index search. The right fix is == 26, with perf_info asserted as the last field after mic_test. AGENTS.md asks for the check to be updated, not weakened.

  2. hos_crash_clean can loop forever and write past its buffer (host/hos_crash_dialog.h:940). The device loop calls strstr(s, devices[d]) from the start of the string each time. If the microphone's name (from microphone_device=) is a substring of <device>, the loop never ends, and each pass grows the string by 8 - dn bytes. Examples of such names are dev, vice and device. Even with a normal short name (3 to 7 bytes), every replacement grows the string in place with no check against cap. build_report breaks the "at least 3x its text" rule: it copies a crash line of up to 4095 bytes into line[1208], which leaves it full (host/hos_crash_dialog.c:1307). hos_crash_devices_out's memmove(name + 8, ...) grows the string the same way. This runs after every crash, in the process that has to survive. The fix: search on from after each replacement, and write into a separate output bounded by cap, as hos_crash_redact_line already does.

  3. Stray keys can send a report. Consent is meant to be a click. In the crash window (gpu/vk/vk_crash_window.cpp:408), the first button is selected and Enter or Space presses it. The first button is Report automatically on the first page, and Send on the next (host/hos_crash_dialog.c:1404). Key-repeat events are not filtered, and the window raises itself. A player holding or mashing Space or Enter when the game dies can therefore send the report with two presses, without reading the page. The fix:

    • make Cancel (or Show all of it) the default on the Send page;
    • ignore repeat events, and any key already down when a page appeared (take a key only after its key-up has been seen on this page, or after a short delay).
  4. curl reads the player's .curlrc / _curlrc. The description promises "https only, no redirect followed". A location, proxy, insecure or output line in the player's curl config changes that, and the source-text test ("--location" not in post_fn) can't see it. Make --disable (-q) the first argument (host/hos_crash_dialog.c:1348).

  5. The automatic send leaves no trace of what it sent. send_quietly writes its answer with fprintf(stderr, ...) (host/hos_crash_dialog.c:1383), but the helper's stderr is /dev/null on Linux and nothing on Windows (spawn_exe). The docs' claim that it "notes the answer in the log" ([hos crash] sent on its own) is therefore not true. The payload file is also deleted afterwards. A player who chose sending with no window can't see what went. bug-reports/sent/<record> holds only the status and the link. Keep a copy of the payload there (as perf_sent/ does for summaries), and write the line somewhere that persists.

Suggestions

  • Linux: curl gets the game's environment. spawn_exe passes environ unchanged, LD_LIBRARY_PATH and LD_PRELOAD included. That means overlays (MangoHud, Steam's overlay) and any library folder the game was given. own_scope already strips both for busctl; do the same for curl, or the system curl can load the wrong libssl.
  • A timed-out curl may still send. When curl is still running after 30 s, spawn_exe returns 2 and leaves it running. The page then says "Nothing was sent", which may be false. Kill it, or say "may not have been sent". The window is also not redrawn while curl runs, for up to 30 s. Windows marks a window that stops pumping messages as "Not responding" after about 5 s. A "Sending…" page, or pumping events while waiting, would avoid that.
  • What can still leak. Home, user, Wine's Z:\home, C:\Users, /var/home and audio devices are handled. Some things are not:
    • Paths outside the home folder are sent as they are: a game folder on D:\Games\…, /mnt/<name>-nas/… or \\NAS-NAME\share\…, which also gives away the share's host name. Replacing the game folder itself (s->dir) with <game folder> before the shape rules would catch the most common case.
    • The hid: controller "<name>" lines can carry a Bluetooth alias.
    • A user name of three letters or more is replaced as a word in any case. A short name like max or ret will also replace those words in CRASH: lines, and those lines feed the relay's crash signature. Matching case-sensitively in crash lines would avoid that.
  • Perf: the relay takes at most 16 maps per summary, but the game doesn't cap them. A summary with a 17th map in five minutes (warp songs, quick scene changes) gets a 400. perf_report.py then counts it as refused and deletes it. Write the summary early when maps reaches 16 (gpu/vk/perf_report.cpp:151).
  • Perf measurements. The measurement compares main at 127.0 fps, tick off at 124.4 and tick on at 122.9. It was taken on a PC that was swapping, in the intro and title only, and with the tick on the Breakdown is forced on every frame. The −3% is "inside the spread" only because the spread was 15 fps. One A/B in play (a field or town, capped and uncapped) on a quiet PC would settle it. Not blocking on its own.
  • Windows (untested). The helper now loads all of longshot_vk.dll after a crash, which runs its imports and static initializers, possibly near out of memory. The Windows run should also try a game folder with non-ASCII characters in its path: System32's curl.exe has to open --data-binary @<path>.

The description

  • The out-of-memory work isn't mentioned in What changed. That covers own_scope() calling busctl … StartTransientUnit at every Linux start to leave the game's systemd scope, SIGTERM held back in the helper, and oom_kill: yes records shown at the next start. It came over from #141 and is documented in docs/crash-report-window.md. It is a new process spawned on every start, so it belongs in Why and What changed, and in Risks.
  • "So any build can send one, not only a launcher-made one" doesn't match the code. make_game.py writes crash_reports= only for packaged tools (if tag:), so a build from a checkout has no Report automatically. The README says "In a build the Longshot Launcher made", so make the description and the README agree.
  • HOS_CRASH_TEST is on main already, so its "Before this PR: not there" row is wrong.
  • "nx2/crash_report.py and the launcher's --crash-* commands are gone": neither was ever on main. They came from this branch's earlier commits and from #141. Say "replaces #141's launcher window".
  • One topic: this is crash reports, performance statistics and OOM detection together. That is fine if the team takes it as one "telemetry" change, but the box says one topic.

The launcher contract (from this side), with MagicBeanDev/LongshotLauncher#128

  • longshot_crash.ini. The launcher writes # …\nsend=1\n (or send=0) with \n line endings, before Play, and only when the value changed. hos_crash_auto_send skips the comment, allows blanks and a CR, and takes only send=1. That matches. The game only reads the file and never writes it.
  • longshot_perf.ini. It is written only through nx2/perf_report.py --tick on|off [--keep-sent off], which the launcher calls with exactly that syntax and waits up to 5 s for. --tick off deletes the file, perf_pending/ and perf_sent/. The game reads send= (atoi == 1) at its first present and again before each summary, so unticking during play stops it. --send runs in its own detached process and sends nothing unless send=1. That matches.
  • The removed --crash-* commands. Nothing on either side still calls --crash-window, --crash-preview, --crash-send, sender= or LONGSHOT_CRASH_SENDER.

Overlapping PRs

This PR supersedes both:

  • #206 has the same perf code, older, at API 23 and on the old api host.
  • #141 has the launcher-drawn crash window, which this replaces with the game's own. Its OOM part is carried over here.

Both are by the same author, and both should be closed once this is the path taken. (Their launcher halves, MagicBeanDev/LongshotLauncher#74 and MagicBeanDev/LongshotLauncher#46, look superseded by MagicBeanDev/LongshotLauncher#128 as well.)

Testing

  • Done: the author's Linux runs (CachyOS, KDE Wayland, RTX 3060 Laptop and Radeon 680M) from a packaged build: a real crash through XWayland, a report posted to issue 81, and the perf relay counting summaries. tests/test_crash_dialog.py uses a stand-in curl and tests/test_perf_report.py a relay on 127.0.0.1. Neither test reaches the network.
  • Missing: both Windows runs in Must be tested (the themed window and a report through System32's curl; the statistics ticked, 6 minutes, and perf_sent/ after the next start). No tester has posted a run on the PR. This can't be approved until those runs are posted and the rebase is done.
## Review: request changes What this does for a player: after a crash, the game's own themed window can send a crash report to the team in two clicks, or with no window at all when the launcher's "Send crash reports automatically" box is ticked. The game can also count frame times, hitches and movie playback per map for the launcher's opt-in "Send performance statistics". I read the whole diff (code only, nothing run). The design holds up. Every network path is opt-in. The perf side stays off and writes nothing without `send=1`, and the tick off deletes what is waiting. The crash post is https only, posts to the stamp's address only, and the Sent link can only go to issue 81. The blocking points below are mostly small. ### Blocking 1. **The PR conflicts with main, and the API version it takes is gone.** Main took `MKWVK_API_VERSION` 25 for `mic_test` (https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1102). After the rebase, `perf_info` must come after `mic_test` as version **26**. Six files conflict: `gpu/vk/vk_api.h`, `host/rx_backend.c`, `gpu/vk/vk_dlss.cpp` (two hunks, an include and a log line), `tests/test_audio_output_devices.py` (three hunks), `docs/images/guide/README.md` and `docs/images/guide/manifest.json` (four hunks). The text conflicts are easy, but two need decisions. The version number has to change. The guide manifest's reviewed-source hashes have to be worked out again against main's new Graphics capture (main recaptured it for the microphone test). In the same rebase, keep main's exact checks in `tests/test_audio_output_devices.py`. This branch loosened them from `== 24` to `>= 24`, and from `fields[-6:]` to an index search. The right fix is `== 26`, with `perf_info` asserted as the last field after `mic_test`. AGENTS.md asks for the check to be updated, not weakened. 2. **`hos_crash_clean` can loop forever and write past its buffer** (`host/hos_crash_dialog.h:940`). The device loop calls `strstr(s, devices[d])` from the start of the string each time. If the microphone's name (from `microphone_device=`) is a substring of `<device>`, the loop never ends, and each pass grows the string by `8 - dn` bytes. Examples of such names are `dev`, `vice` and `device`. Even with a normal short name (3 to 7 bytes), every replacement grows the string in place with no check against `cap`. `build_report` breaks the "at least 3x its text" rule: it copies a crash line of up to 4095 bytes into `line[1208]`, which leaves it full (`host/hos_crash_dialog.c:1307`). `hos_crash_devices_out`'s `memmove(name + 8, ...)` grows the string the same way. This runs after every crash, in the process that has to survive. The fix: search on from after each replacement, and write into a separate output bounded by `cap`, as `hos_crash_redact_line` already does. 3. **Stray keys can send a report.** Consent is meant to be a click. In the crash window (`gpu/vk/vk_crash_window.cpp:408`), the first button is selected and Enter or Space presses it. The first button is Report automatically on the first page, and Send on the next (`host/hos_crash_dialog.c:1404`). Key-repeat events are not filtered, and the window raises itself. A player holding or mashing Space or Enter when the game dies can therefore send the report with two presses, without reading the page. The fix: - make Cancel (or Show all of it) the default on the Send page; - ignore repeat events, and any key already down when a page appeared (take a key only after its key-up has been seen on this page, or after a short delay). 4. **curl reads the player's `.curlrc` / `_curlrc`.** The description promises "https only, no redirect followed". A `location`, `proxy`, `insecure` or `output` line in the player's curl config changes that, and the source-text test (`"--location" not in post_fn`) can't see it. Make `--disable` (`-q`) the **first** argument (`host/hos_crash_dialog.c:1348`). 5. **The automatic send leaves no trace of what it sent.** `send_quietly` writes its answer with `fprintf(stderr, ...)` (`host/hos_crash_dialog.c:1383`), but the helper's stderr is `/dev/null` on Linux and nothing on Windows (`spawn_exe`). The docs' claim that it "notes the answer in the log" (`[hos crash] sent on its own`) is therefore not true. The payload file is also deleted afterwards. A player who chose sending with no window can't see what went. `bug-reports/sent/<record>` holds only the status and the link. Keep a copy of the payload there (as `perf_sent/` does for summaries), and write the line somewhere that persists. ### Suggestions - **Linux: curl gets the game's environment.** `spawn_exe` passes `environ` unchanged, `LD_LIBRARY_PATH` and `LD_PRELOAD` included. That means overlays (MangoHud, Steam's overlay) and any library folder the game was given. `own_scope` already strips both for `busctl`; do the same for curl, or the system curl can load the wrong libssl. - **A timed-out curl may still send.** When curl is still running after 30 s, `spawn_exe` returns 2 and leaves it running. The page then says "Nothing was sent", which may be false. Kill it, or say "may not have been sent". The window is also not redrawn while curl runs, for up to 30 s. Windows marks a window that stops pumping messages as "Not responding" after about 5 s. A "Sending…" page, or pumping events while waiting, would avoid that. - **What can still leak.** Home, user, Wine's `Z:\home`, `C:\Users`, `/var/home` and audio devices are handled. Some things are not: - Paths outside the home folder are sent as they are: a game folder on `D:\Games\…`, `/mnt/<name>-nas/…` or `\\NAS-NAME\share\…`, which also gives away the share's host name. Replacing the game folder itself (`s->dir`) with `<game folder>` before the shape rules would catch the most common case. - The `hid: controller "<name>"` lines can carry a Bluetooth alias. - A user name of three letters or more is replaced as a word in any case. A short name like `max` or `ret` will also replace those words in `CRASH:` lines, and those lines feed the relay's crash signature. Matching case-sensitively in `crash` lines would avoid that. - **Perf: the relay takes at most 16 maps per summary, but the game doesn't cap them.** A summary with a 17th map in five minutes (warp songs, quick scene changes) gets a 400. `perf_report.py` then counts it as refused and deletes it. Write the summary early when `maps` reaches 16 (`gpu/vk/perf_report.cpp:151`). - **Perf measurements.** The measurement compares main at 127.0 fps, tick off at 124.4 and tick on at 122.9. It was taken on a PC that was swapping, in the intro and title only, and with the tick on the Breakdown is forced on every frame. The −3% is "inside the spread" only because the spread was 15 fps. One A/B in play (a field or town, capped and uncapped) on a quiet PC would settle it. Not blocking on its own. - **Windows (untested).** The helper now loads all of `longshot_vk.dll` after a crash, which runs its imports and static initializers, possibly near out of memory. The Windows run should also try a game folder with non-ASCII characters in its path: System32's `curl.exe` has to open `--data-binary @<path>`. ### The description - **The out-of-memory work isn't mentioned in What changed.** That covers `own_scope()` calling `busctl … StartTransientUnit` at **every** Linux start to leave the game's systemd scope, SIGTERM held back in the helper, and `oom_kill: yes` records shown at the next start. It came over from https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/141 and is documented in `docs/crash-report-window.md`. It is a new process spawned on every start, so it belongs in Why and What changed, and in Risks. - **"So any build can send one, not only a launcher-made one"** doesn't match the code. `make_game.py` writes `crash_reports=` only for packaged tools (`if tag:`), so a build from a checkout has no Report automatically. The README says "In a build the Longshot Launcher made", so make the description and the README agree. - **`HOS_CRASH_TEST` is on main already**, so its "Before this PR: not there" row is wrong. - **"`nx2/crash_report.py` and the launcher's `--crash-*` commands are gone"**: neither was ever on main. They came from this branch's earlier commits and from https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/141. Say "replaces https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/141's launcher window". - **One topic:** this is crash reports, performance statistics and OOM detection together. That is fine if the team takes it as one "telemetry" change, but the box says one topic. ### The launcher contract (from this side), with https://git.magicbean.dev/MagicBeanDev/LongshotLauncher/pulls/128 - **`longshot_crash.ini`.** The launcher writes `# …\nsend=1\n` (or `send=0`) with `\n` line endings, before Play, and only when the value changed. `hos_crash_auto_send` skips the comment, allows blanks and a CR, and takes only `send=1`. That matches. The game only reads the file and never writes it. - **`longshot_perf.ini`.** It is written only through `nx2/perf_report.py --tick on|off [--keep-sent off]`, which the launcher calls with exactly that syntax and waits up to 5 s for. `--tick off` deletes the file, `perf_pending/` and `perf_sent/`. The game reads `send=` (`atoi == 1`) at its first present and again before each summary, so unticking during play stops it. `--send` runs in its own detached process and sends nothing unless `send=1`. That matches. - **The removed `--crash-*` commands.** Nothing on either side still calls `--crash-window`, `--crash-preview`, `--crash-send`, `sender=` or `LONGSHOT_CRASH_SENDER`. ### Overlapping PRs This PR supersedes both: - https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/206 has the same perf code, older, at API 23 and on the old api host. - https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/141 has the launcher-drawn crash window, which this replaces with the game's own. Its OOM part is carried over here. Both are by the same author, and both should be closed once this is the path taken. (Their launcher halves, https://git.magicbean.dev/MagicBeanDev/LongshotLauncher/pulls/74 and https://git.magicbean.dev/MagicBeanDev/LongshotLauncher/pulls/46, look superseded by https://git.magicbean.dev/MagicBeanDev/LongshotLauncher/pulls/128 as well.) ### Testing - **Done:** the author's Linux runs (CachyOS, KDE Wayland, RTX 3060 Laptop and Radeon 680M) from a packaged build: a real crash through XWayland, a report posted to issue 81, and the perf relay counting summaries. `tests/test_crash_dialog.py` uses a stand-in curl and `tests/test_perf_report.py` a relay on 127.0.0.1. Neither test reaches the network. - **Missing:** both Windows runs in Must be tested (the themed window and a report through System32's curl; the statistics ticked, 6 minutes, and `perf_sent/` after the next start). No tester has posted a run on the PR. This can't be approved until those runs are posted and the rebase is done.
@ -0,0 +148,4 @@
const std::string name = map[0] ? map : shaders_preparing() ? pr::kPreparingShaders : pr::kTitleAndIntro;
if (!current || name != current_map) {
current_map = name;
current = &maps[name];
Owner

The relay takes at most 16 keys in maps, but nothing here caps them. A 17th map within one period gets a 400, and perf_report.py then counts the summary as refused and deletes it. Call summarize(false) early when maps.size() reaches 16.

The relay takes at most 16 keys in `maps`, but nothing here caps them. A 17th map within one period gets a 400, and `perf_report.py` then counts the summary as refused and deletes it. Call `summarize(false)` early when `maps.size()` reaches 16.
@ -249,0 +251,4 @@
"" before a save is loaded) into map, and the movies' totals so far (host/nn_video.c): [0] pictures shown, [1] skipped to
keep up with the sound, [2] shown late (over 50 ms after the one before), [3] decoding time in ns, [4] decoders that did
not start (another one took over). Returns the build's version (LONGSHOT_VERSION) */
const char *(*perf_info)(char *map, int cap, int64_t movie[5]);
Owner

Main now uses version 25 for mic_test (#1102). After the rebase, perf_info goes after mic_test, and MKWVK_API_VERSION goes to 26.

Main now uses version 25 for `mic_test` (https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1102). After the rebase, `perf_info` goes after `mic_test`, and `MKWVK_API_VERSION` goes to 26.
@ -0,0 +405,4 @@
if (in.scancode == 43) sel = (sel + (in.shift ? n - 1 : 1)) % n; // (Tab, Shift+Tab)
if (in.scancode == 79 || in.scancode == 81) sel = std::min(n - 1, sel + 1); // (right, down)
if (in.scancode == 80 || in.scancode == 82) sel = std::max(0, sel - 1); // (left, up)
if (in.scancode == 40 || in.scancode == 88 || in.scancode == 44) result = sel; // (Enter, the keypad's, Space)
Owner

Key-repeat events aren't filtered, and a key already held when the page appears counts. A player holding or mashing Space or Enter as the game crashes presses the selected button, and the window raises itself to the front. Drop repeats (SDL3's repeat byte in the keyboard event, SDL2's at offset 13). Take a key only once its key-up has been seen on this page, or after a short delay.

Key-repeat events aren't filtered, and a key already held when the page appears counts. A player holding or mashing Space or Enter as the game crashes presses the selected button, and the window raises itself to the front. Drop repeats (SDL3's `repeat` byte in the keyboard event, SDL2's at offset 13). Take a key only once its key-up has been seen on this page, or after a short delay.
@ -952,0 +1401,4 @@
free(json.s), free(shown.s);
return 0;
}
const char *const ask[] = {"Send", "Show all of it", "Cancel"};
Owner

Send is the first button, so it is selected and Enter or Space presses it. Together with Report automatically being first on the page before, two stray presses carried over from play send the report. Make Cancel (or Show all of it) the default here.

Send is the first button, so it is selected and Enter or Space presses it. Together with Report automatically being first on the page before, two stray presses carried over from play send the report. Make Cancel (or Show all of it) the default here.
@ -1017,0 +1518,4 @@
game was in (DefaultOOMPolicy=stop) and everything in it, which must not be this helper. Where it can't (no systemd, or
in a Flatpak, which keeps the sandbox's processes in its scope), it stays; an out-of-memory end is then shown at the
next start (hos_crash_show_at_start). At most 2 s, its output nowhere, without the game's library paths */
static void own_scope(void) {
Owner

The PR description doesn't mention this: a busctl --user ... StartTransientUnit at every Linux start (a->mode == 1), plus SIGTERM held in the helper. It came over from #141. Please add it to What changed and Risks, because it is a new process on every start.

The PR description doesn't mention this: a `busctl --user ... StartTransientUnit` at every Linux start (`a->mode == 1`), plus SIGTERM held in the helper. It came over from https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/141. Please add it to What changed and Risks, because it is a new process on every start.
@ -952,0 +1304,4 @@
version, in.release ? "Yes" : "No (a local or modified build)", os, f->renderer[0] ? f->renderer : "not known",
f->gpu[0] ? f->gpu : "not known", f->cpu[0] ? f->cpu : "not known", place, in.settings, in.changed);
for (int i = 0; i < crash.n && i < 12; i++) {
char line[HOS_CRASH_LINE_LIMIT * 4 + 8];
Owner

crash.at[i] can be up to 4095 bytes (one[4096] in crash_lines), so this copy fills line to its end. That breaks hos_crash_clean's rule that the buffer is at least 3x its text, and the device replacement then writes past it. Size it from the text (strlen * 3 + 64, as put_field does), or cut the line to HOS_CRASH_LINE_LIMIT first.

`crash.at[i]` can be up to 4095 bytes (`one[4096]` in `crash_lines`), so this copy fills `line` to its end. That breaks `hos_crash_clean`'s rule that the buffer is at least 3x its text, and the device replacement then writes past it. Size it from the text (`strlen * 3 + 64`, as `put_field` does), or cut the line to `HOS_CRASH_LINE_LIMIT` first.
@ -952,0 +1345,4 @@
remove_utf8(headers), remove_utf8(answer);
snprintf(data, sizeof data, "@%s", payload);
/* (no redirect followed: the report goes to the address in the stamp and nowhere else; at most 30 s) */
const char *const args[] = {"--silent", "--show-error", "--max-time", "25", "--connect-timeout", "10", "--proto", "=https",
Owner

curl reads the player's ~/.curlrc (%APPDATA%\_curlrc on Windows) unless --disable / -q is the first argument. A location, proxy, insecure or output line there breaks the promise of https only and no redirects, and the source-text test can't see it. Put "--disable" first.

curl reads the player's `~/.curlrc` (`%APPDATA%\_curlrc` on Windows) unless `--disable` / `-q` is the **first** argument. A `location`, `proxy`, `insecure` or `output` line there breaks the promise of https only and no redirects, and the source-text test can't see it. Put `"--disable"` first.
@ -952,0 +1348,4 @@
const char *const args[] = {"--silent", "--show-error", "--max-time", "25", "--connect-timeout", "10", "--proto", "=https",
"--header", "Content-Type: application/json", "--user-agent", "longshot-crash-report",
"--data-binary", data, "--dump-header", headers, "--output", answer, s->reports};
const int ran = spawn_exe(curl, args, (int)(sizeof args / sizeof args[0]), 30000, NULL);
Owner

Two things here:

  • If curl is still running after 30 s, spawn_exe returns 2 and leaves it running, and the page then says "Nothing was sent", which may be false. Kill it, or word the page as "may not have been sent".
  • On Linux, curl gets environ as it is, LD_LIBRARY_PATH and LD_PRELOAD included. own_scope strips both for busctl; strip them here too, so the system curl can't load the game's libraries or an overlay's.

The window is also not redrawn for up to 30 s while this waits. On Windows it shows "Not responding" after about 5 s.

Two things here: - If curl is still running after 30 s, `spawn_exe` returns 2 and leaves it running, and the page then says "Nothing was sent", which may be false. Kill it, or word the page as "may not have been sent". - On Linux, curl gets `environ` as it is, `LD_LIBRARY_PATH` and `LD_PRELOAD` included. `own_scope` strips both for busctl; strip them here too, so the system curl can't load the game's libraries or an overlay's. The window is also not redrawn for up to 30 s while this waits. On Windows it shows "Not responding" after about 5 s.
@ -952,0 +1380,4 @@
sent = hos_crash_answer(result, sizeof result, status, body, s->issue, link, sizeof link, &again);
free(body);
if (sent) keep_answer(s, status, link);
fprintf(stderr, "[hos crash] sent on its own (longshot_crash.ini): %s\n", sent ? link : result);
Owner

The helper's stderr is /dev/null on Linux and nothing on Windows (spawn_exe), so this line never reaches a log, although docs/crash-report-window.md says it does. The payload is also deleted just below. For the send with no window, keep a copy of what was sent (in bug-reports/sent/<record> beside the status and link, like perf_sent/), and write this line to a file that persists.

The helper's stderr is `/dev/null` on Linux and nothing on Windows (`spawn_exe`), so this line never reaches a log, although docs/crash-report-window.md says it does. The payload is also deleted just below. For the send with no window, keep a copy of what was sent (in `bug-reports/sent/<record>` beside the status and link, like `perf_sent/`), and write this line to a file that persists.
@ -549,0 +937,4 @@
hos_crash_devices_out(s);
for (int d = 0; d < ndevices; d++) {
const size_t dn = strlen(devices[d]);
for (char *at; dn >= 3 && (at = strstr(s, devices[d]));) {
Owner

This loop calls strstr from the start of s every time and grows the string in place, without checking cap. If the microphone's name is a substring of <device> (dev, vice, device), it never ends and writes past the buffer. With a shorter name (3 to 7 bytes), each hit grows the text by 8 - dn bytes past what the caller made room for. Search on from at + 8, and build into a separate buffer bounded by cap, as hos_crash_redact_line does. hos_crash_devices_out's memmove(name + 8, ...) (line 846) has the same growth.

This loop calls `strstr` from the start of `s` every time and grows the string in place, without checking `cap`. If the microphone's name is a substring of `<device>` (`dev`, `vice`, `device`), it never ends and writes past the buffer. With a shorter name (3 to 7 bytes), each hit grows the text by `8 - dn` bytes past what the caller made room for. Search on from `at + 8`, and build into a separate buffer bounded by `cap`, as `hos_crash_redact_line` does. `hos_crash_devices_out`'s `memmove(name + 8, ...)` (line 846) has the same growth.
@ -26,3 +26,3 @@
assert 'r.value != &output_pick' in menu and 'audio_output_device_id=%s' in menu
assert 'output_forced()' in menu and 'audio_output_device_name' in menu
assert '#define MKWVK_API_VERSION 24' in api
assert int(re.search(r'#define MKWVK_API_VERSION (\d+)', api).group(1)) >= 24 # (24: the playback callbacks; later versions append)
Owner

This loosens the check (== 24 becomes >= 24, and fields[-6:] becomes an index search), which AGENTS.md asks us not to do. Main's version of this test now asserts == 25 exactly, with mic_test last. After the rebase, keep the exact form: == 26, with perf_info asserted as the field after mic_test.

This loosens the check (`== 24` becomes `>= 24`, and `fields[-6:]` becomes an index search), which AGENTS.md asks us not to do. Main's version of this test now asserts `== 25` exactly, with `mic_test` last. After the rebase, keep the exact form: `== 26`, with `perf_info` asserted as the field after `mic_test`.
@ -70,3 +71,3 @@
#include "audio_output_devices.h"
#include "gpu/vk/vk_api.h"
_Static_assert(MKWVK_API_VERSION == 24, "playback callbacks need a new ABI version");
_Static_assert(MKWVK_API_VERSION >= 24, "playback callbacks need a new ABI version");
Owner

Same here: == 26 after the rebase, not >= 24.

Same here: `== 26` after the rebase, not `>= 24`.
All checks were successful
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 8s
CI / Workflow lint (actionlint) (pull_request) Successful in 10s
Guide screenshots / Guide images and local UI source review (pull_request) Successful in 11s
Third-party parts / As their notes say (pull_request) Successful in 12s
PR description / Description follows the template (pull_request_target) Successful in 5s
This pull request has changes conflicting with the target branch.
  • docs/images/guide/README.md
  • docs/images/guide/manifest.json
  • gpu/vk/vk_api.h
  • gpu/vk/vk_dlss.cpp
  • host/rx_backend.c
  • tests/test_audio_output_devices.py
View command line instructions

Manual merge helper

Use this merge commit message when completing the merge manually.

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u 'feat/telemetry:gab9281-feat/telemetry'
git switch 'gab9281-feat/telemetry'
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
3 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot!983
No description provided.