| Commit message (Collapse) | Author | Files | Lines |
|---|
|
The engine clamps mat_monitorgamma at 2.6, so the 2.8..3.0 end of the slider
did nothing. Read the value back one tick later (ClientCmd is queued) and
interpose SDL_SetWindowGammaRamp to see whether the engine pushes a ramp at
all.
|
|
- Each changelevel trigger now shows its own target map. The trigger whose
target is the map we came from is marked "<- map"; the forward one is
"-> map". No more appending one global previous-map line to every box.
- Draw the volume as a projected 8-corner / 12-edge wireframe with the tag
above its highest visible corner instead of a screen-space rectangle.
- Camera FOV is 40..140 (stock 75) and viewmodel FOV 20..120 (stock 54); 0 is
no longer allowed and old configs using 0 migrate to the stock value.
- Log the mat_monitorgamma readback via ICvar so we can tell whether the
command is rejected or the OS/compositor ignores the gamma ramp.
|
|
- ESP: alert state (IDLE/ALERT/COMBAT/SCRIPT), enemy weapon labels, hostile/
neutral/friendly colour pickers, off-screen direction arrows, and
trigger_changelevel volumes with next/previous map labels.
- Aim: silent aim reuses the Aimbot FOV, vischeck and bone settings; add
optional aim-punch compensation for aimbot, silent aim, triggerbot and
melee (server bullets add m_vecPunchAngle to the usercmd angles).
- View: always-on Camera / display card with camera FOV (writes the client
m_iDefaultFOV netvar), viewmodel FOV and monitor gamma (client cvars).
- Movement: Jump assist card with auto crouch-jump and edge jump (point
ground probe 0.12 s ahead).
- Spread: Engine spread override toggle gates the direct RandomFloat zeroing;
off leaves only the seed search.
- Server: entity count/base/classname helpers, trigger_changelevel map name
and world bounds via IServerUnknown/ICollideable.
- Netvars: m_vecPunchAngle and m_iDefaultFOV.
|
|
The ApplySpread random draws are not guaranteed to run in the player's frame
(the compiler can outline them), so the frame-based shooter check missed and
the shotgun kept its seed-minimised spread. The player's FireBullets
RandomSeed call now arms a short 8-draw window; the next RandomFloat draws
return 0. Keep the frame check as the other path. Remove the diagnostics used
to find this.
|
|
Picking the best of 256 pending seeds still leaves the shotgun's 8/12 pellets
a few degrees apart. CShotManipulator::ApplySpread is inlined into
CBaseEntity::FireBullets and draws x/y through the global
CUniformRandomStream; hook its RandomFloat slot and return 0 when the calling
frame's shooter is the local server player. This collapses every pellet onto
the shot direction while NPC fire and client prediction keep their randomness.
|
|
The server RandomSeed GOT slot was hardcoded to 0xce1348, measured on the
older bin/server.so. The anniversary branch loads hl2_complete/bin/server.so,
whose R_386_JMP_SLOT for RandomSeed is 0xce5348, so the patch hit unrelated
writable data and no spread silently did nothing.
- Read the relocation from the loaded module's ELF (module_jump_slot) instead
of hardcoding, with the two known offsets as fallback.
- Override whenever the RandomSeed caller's `this` is the local server player
(CBaseEntity::FireBullets) instead of matching a hardcoded return-address
range that also moved between builds.
- m_takedamage is a char (FIELD_CHARACTER) in the 2013 SDK; resolve it as such
and read a byte, so the IMMUNE damage tag actually works.
|
|
Props render correctly with all seven DrawModel arguments forwarded; the
per-draw model/client/origin log is no longer needed.
|
|
Draw ARM/VULN/IMM on the left edge of the box with a 10px font and a black
1px shadow, instead of full-size text above the name. The small font is a
second HFont created next to the 16px body font.
CONTEXT: document the DrawModel argument fix and the tag styling.
|
|
DrawModel takes (this, results, info, boneToWorld, flexWeights,
flexDelayedWeights, const Vector &modelOrigin, flags). The hook dropped
modelOrigin, so flags landed in its slot and the real flags were read from
caller stack garbage. studiorender.so 0x83ad0 reads 0x20(%ebp) as the
modelOrigin pointer and 0x24(%ebp) as the flags; passing both restores the
engine's normal draw.
This is the cause of the intermittent (~50%) checkerboard/error-material
props: the draw was being issued with a garbage flag word.
|
|
- Resolve server m_takedamage (DAMAGE_NO/DAMAGE_EVENTS_ONLY = immune).
- Classify gunship/strider/dropship as ARMORED (OnTakeDamage only accepts
DMG_BLAST; only the RPG passes), and headcrab/antlion/manhack as VULN
(gravity gun / physics impact).
- ESP draws ARMORED/VULN/IMMUNE tags above the name (new "Damage tags"
toggle, [esp] dmg).
- Aimbot, silent aim, triggerbot and melee skip immune targets and armoured
targets unless the active weapon is the RPG.
|
|
Drop the temporary DrawModel/SetColor logs now that the prop tint works, and
document the IStudioRender hook layout, the short model-name format used for
prop detection, the SetColor-before-DrawModel ordering, and why prop
materials cannot be tinted directly.
|
|
The engine calls SetColorModulation(1,1,1) immediately before DrawModel, so
the in-draw flag never sees it. Set the prop colour at the start of the prop
DrawModel hook instead (colour only, not alpha), and stop applying
ColorModulate to prop materials so only one mechanism touches props.
|
|
The entity-path rewrite of prop_tint_update dropped studio_props_init(), so
the minimal IStudioRender hooks were never installed (no studio: log lines).
Call it while the prop tint is enabled.
|
|
The forced SetColorModulation/SetAlphaModulation calls before DrawModel are
what turned matched props into the error material and destabilised the
material system. Mirror Amalgam instead: set a "drawing a prop" flag around
the original draw and only substitute the colour when the engine itself calls
SetColorModulation during that draw. Log those calls for the first few draws
to see whether the engine applies modulation inside DrawModel at all.
|
|
Hooking IStudioRender's draw/setter slots made matched props render the error
material and then crashed the material system ("debug/modelstats has bad
reference count 0 when being bound"). Keep the hook code for reference but do
not install it.
Fix the m_clrRender netvar resolution: the walk picked an embedded table entry
at offset 0, so the earlier entity writes stomped entity vtables. Only accept
offsets inside C_BaseEntity (0x20..0x2000), then write the prop colour there
for models/props entities. Restore white when disabled.
|
|
studiohdr_t::name is not "models/props/...". In this build it is the short
model name with Windows separators, e.g. "props_junk\wood_crate002a.mdl"
(the models/ prefix is stripped). Normalise the prefix/separators and match
props/, props\ and props_ (excluding the distant props_skybox set), so the
studio hooks actually recognise prop draws.
|
|
Log the first crate materials the tint pass modifies and the first DrawModel/
DrawModelArray calls with their model names, to see whether the crates are
brush materials that were never reached, materials loaded after the tint was
applied, or a studio draw path that is still missed.
|
|
m_clrRender resolved to offset 0 in this build; writing it stomped entity
vtables instead of tinting anything. Remove that path.
The static-prop logging showed DrawModelStaticProp is never called for these
props, so hook the other studio draw entries as well: DrawModel (29) for
entity models and DrawModelArray (46, the instanced static-prop path). Both
read the model name from DrawModelInfo_t::m_pStudioHdr and force the prop
colour for models/props* draws.
|
|
Static props commonly use brush-style props/... materials
(props/woodcrate001a is LightmappedGeneric), so the models/props-only filter
missed them. Include the props/ prefix.
Also log the first static prop studio draws and a 500-draw summary so we can
see whether the engine draws props through DrawModelStaticProp at all.
|
|
Most of the crates are prop_static, which have no client entity, so writing
m_clrRender only reached the three dynamic ones. Hook the StudioRender
interface (VStudioRender025) on the interface object's vtable: substitute the
prop colour in SetColorModulation/SetAlphaModulation while
DrawModelStaticProp is drawing a models/props model, and set the modulation
explicitly before the original draw in case the engine sets it just before the
call.
Vtable indices verified by disassembling CStudioRenderContext: 27 = three
float stores from the argument pointer, 28 = one float, 30 = the
DrawModelInfo_t-drawing entry with three arguments (the 2013 SDK header lists
these slots five positions earlier in this build).
|
|
Prop model materials (VertexLitGeneric) ignore IMaterial::ColorModulate: the
engine feeds StudioRender the per-entity render colour (m_clrRender) on every
draw, so the material-side tint never showed on crates and other props.
Resolve m_clrRender from DT_BaseEntity and, every 10 ms while World tint and
Props are enabled, write the prop colour into every entity whose model is
under models/props. Restore white once when disabled.
|
|
The procedural texture reaches the sky sampler with the red and blue channels
exchanged (blue sky rendered orange/yellow), while the game's own textures are
correct. Swap R/B in the decoded buffer so the sampled colours match the PNG.
|
|
Move the HDR decision from texture load to apply_sky_texture, after the sky
materials have definitely been enumerated: compensate the pixel buffer
(sRGB -> linear, alpha = 1/8) and re-download the procedural texture before
assigning it. The earlier load-time check ran before the materials were
visible and silently chose the LDR path, so the 8x HDR decode stayed.
Also log the compensation explicitly so the chosen path is visible.
|
|
The material reports shader "Sky_HDR_DX9", not "sky", so the linear/alpha
compensation was never applied and the raw sRGB image was scaled by the HDR
decode's 8x. Match the Sky_HDR prefix so the compensation path is used.
|
|
The HDR "sky" shader samples the compressed texture as rgb *= alpha,
multiplies by 8, then converts linear->sRGB for display. Feeding it the raw
sRGB image produced the blown-out / oversaturated sky.
When the map uses the HDR sky shader, pre-convert the image to linear and set
alpha to 1/8 so the 8x scale cancels and the final linear->sRGB conversion
restores the original colours. For LDR (UnlitGeneric) materials, create the
texture with TEXTUREFLAGS_SRGB so the normal sRGB sampling is used.
The first sky material's shader name is logged for confirmation.
|
|
The last crash happened after "applied", in materialsystem while the renderer
walked a shader-parameter list that contained a null vtable entry. That points
at the Refresh()/SetUndefined experiment: undefining $hdrcompressedtexture
while the sky shader's static combo still expected it (and rebuilding state
mid-level) left a broken material.
Set the texture on $basetexture/$hdrbasetexture/$hdrcompressedtexture (and the
0/1/2 variants) and do neither SetUndefined nor Refresh: the visible-path
material binds its base texture dynamically, which is what made the earlier
black sky (empty 798x798 VTF) appear black at all. Log the first sky
material's shader name so we know whether this map uses the LDR (UnlitGeneric)
or HDR (sky) path.
|
|
- The material system rejects textures whose dimensions are not multiples of 4
("Image dimensions must be multiple of 4!"), so scale the decoded image to
the next power of two (bilinear). 798x798 -> 1024x1024. The invalid VTF also
crashed the renderer after the apply.
- Move skybox_update() out of the VGUI paint hook into the FRAME_RENDER_START
FrameStageNotify stage. ITexture::Download only regenerates bits when the
material system considers the frame render-safe; during paint it returned
early, the regenerator never ran (no pixels), and a later upload of the
empty VTF crashed.
- Log the regenerator entry so the log shows whether the upload path ran.
|
|
The HDR sky materials use the "sky" shader with $hdrcompressedtexture (DXT5).
Assigning an RGBA8 procedural texture there sampled as compressed data and
rendered black. Point $hdrbasetexture at the image, undefine
$hdrcompressedtexture (and its 0/1/2 variants) so the shader falls back, and
call IMaterial::Refresh (37) because the sky shader binds textures in
SHADER_INIT, which is otherwise cached.
GUI: draw_checkbox and the card expand chevron ignored the content clip, so
cards scrolled out of view leaked header widgets outside the window.
Log the first regenerator call (size/pitch/first pixel) for diagnosis.
|
|
client.so, engine.so and vguimatsurface.so all export their own
png_create_read_struct. Loading the system libpng16 made its internal calls
bind to the game's older libpng, whose png_struct layout differs, so
png_safe_warning read a NULL error_ptr and crashed.
Replace libpng with a small PNG reader: zlib's uncompress (dlopen'ed) plus the
five PNG row filters, supporting 8-bit non-interlaced RGB/RGBA/gray/palette
images with tRNS alpha for palettes. Output verified byte-identical to libpng
for the skybox.png in use.
|
|
The crash was CreateTextureFromBits: this build's IMaterialSystem interface
stops at GetDisplayDeviceName, and the slots after it are CMaterialSystem's
own virtuals, so index 134 was an unrelated function.
Use CreateProceduralTexture (81, verified by disassembling the interface
vtable) together with an ITextureRegenerator (SetTextureRegenerator 12,
Download 13) that fills mip 0 via IVTFTexture::ImageData (39; IVTFTexture has
a two-slot virtual destructor, so Width=15, Height=16, RowSizeInBytes=19).
|
|
sv_skyname did not take effect at runtime in this build, so the Skybox card now
replaces the skybox materials' textures with <config dir>/skybox.png.
The PNG is decoded through libpng16's simplified API (dlopen'ed, so no new link
dependency) and uploaded with IMaterialSystem::CreateTextureFromBits as an
RGBA8888 texture. IMaterialVar::SetTextureValue assigns it to $basetexture and
the HDR variants for every material in the SkyBox textures group (all six
faces get the same image).
Original textures are recorded per material/var (texture pointer plus string
fallback) and restored when the feature is off. Records survive level changes
so a reused sky material is still restorable; the override is re-applied to a
new map automatically. The card shows the image path and has a Reload image
button; a missing/unreadable PNG raises an error toast.
|
|
New Visuals -> Skybox card with a runtime sv_skyname override using the same
mechanism as the stock skybox_swapper entity. The choice list is the HL2
skybox set extracted from the game VPKs.
The map's own sky is captured from loaded materials (SkyBox group, parsing
skybox/<name>... face names) before the first override, so turning the
feature off or picking "Map default" restores it; a level change restores it
implicitly. Sky changes invalidate the world tint so new sky materials are
tinted as well.
GUI: add a CK_CHOICE control (value box with < / > zones).
|
|
Read the server m_NPCState and mark NPC_STATE_SCRIPT actors neutral so
cutscene NPCs and scripted pets are not aimbot targets and draw in the
neutral colour. Add Lamarr (Kleiner's pet headcrab) to the neutral model
list as requested.
Also log state changes in the NPC debug line and resolve/log m_NPCState in
the server datamap summary.
|
|
- World tint now also colours prop materials (models/props*) with their own
colour, covering static props and prop_physics while leaving NPC, player and
weapon materials untouched.
- New Fog card: fog_override colour/start/end via ClientCmd, optional skybox
fog, and a Remove fog mode. Restores the map fog (fog_override 0) when
disabled and re-applies after level changes.
- Config sections [worldtint] prop/colour keys and [fog].
|
|
Replace the six RGB sliders on the World tint card with a CK_COLOR control:
each row shows a live swatch and expands an inline picker with a
saturation/value square, a hue bar, and hex/RGB preview. The picker grows the
card in place so it scrolls/clips like any other row, keeps HSV state while
open (grey does not lose its hue), writes the shared 0-255 RGB triple used by
the config, and saves on mouse release.
Costs ~1.2k small rects per frame only while a picker is open; no texture
interface needed.
|
|
Port Amalgam's modulation approach: obtain IMaterialSystem from the already
loaded materialsystem.so (VMaterialSystem080) and walk materials with
FirstMaterial/NextMaterial/GetMaterial. ColorModulate the "World textures"
group (sky names excluded) and the "SkyBox textures"/sky-named group.
The walk runs only when the enabled groups or colours change, restores white
when disabled, and re-applies after level changes. Before touching materials,
GetNumMaterials() and the first material's name/group are sanity-checked so a
wrong vtable slot logs and aborts instead of walking garbage.
GUI: "World tint" card with world/sky toggles and RGB sliders; config section
[worldtint].
|
|
|
|
m_Relationship is DEFINE_UTLVECTOR (FIELD_CUSTOM), not FIELD_EMBEDDED, so the
datamap scan always returned -1 and every NPC fell back to the model
heuristic. Resolve it as FIELD_CUSTOM with an embedded fallback.
Floor turrets never add relationship rows when friendly: CNPC_FloorTurret
overrides Classify() to CLASS_PLAYER_ALLY for citizen-built
(SF_FLOOR_TURRET_CITIZEN) and Alyx-hacked turrets. Read m_spawnflags and
m_bHackedByAlyx from the server instead.
Widen the datamap offset window (0x4..0x4000) for derived-class fields and
log the new offsets plus turret state in the NPC debug line.
|
|
|
|
- Read the server NPC's m_Relationship CUtlVector (entity row for the local
player first, then the CLASS_PLAYER row): D_LI friendly, D_NU neutral,
D_HT/D_FR hostile.
- Use it for ESP colours and aimbot/silent-aim/melee target filtering, with
the existing model heuristic as a fallback.
- Treat bugbait-controlled antlions as friendly when their server
m_hFollowTarget is set.
- Resolve m_Relationship/m_hFollowTarget datamap offsets and log them.
|
|
The fullbright_update/fullbright_invalidate definitions were lost when the
old text-menu section was removed, leaving undefined symbols that made the
dynamic loader reject libhl2hake.so. Restore both definitions and add
-Wl,--no-undefined so missing definitions fail the build instead of the game.
|
|
stage markers
|
|
- Poking VEngineCvar004/ConVar vtables directly crashed before the main menu
(ConVar uses a secondary IConVar vtable). Use the stable IVEngineClient
ClientCmd slot 7 for mat_fullbright instead.
- Reject out-of-range m_hOwnerEntity/m_hOwner offsets: the resolver found a
bogus negative owner offset (-136296061), which could make item-ESP weapon
ownership read wild memory.
|
|
- New Visuals card with the usual Active + Bind; default off. Unbound+Active
is always-on, a bind makes it hold-to-brighten.
- Resolve VEngineCvar004 and set mat_fullbright 1/0 through FindVar (slot 12)
and ConVar::SetValue(int) (slot 11), bypassing the console cheat gate.
Fall back to ClientCmd if the cvar interface is unavailable.
- Reapply after map loads/respawns and persist [fullbright] active/bind.
- Log on/off; show an error toast if the cvar cannot be applied.
|
|
|
|
- Validate the GUI font handles with GetTextSize and recreate them when the
local player entity or UI scale changes. Small scaled sizes fall back to the
proven 16 px size, and the failure log/toast is emitted once instead of every
frame. This fixes sessions where the menu rendered with no text after a
level transition.
- Rework visibility: trace with MASK_SHOT/TRACE_EVERYTHING while ignoring the
local player and target, so static props, crates and other NPCs block.
Non-solid glass and breakable-glass materials are skipped; solid non-break
glass (bulletproof) blocks. Fix CONTENTS_WINDOW 0x2 (was 0x4) in the masks.
- Pass self/target to is_visible from aimbot/silent/melee callers.
|
|
The client m_hActiveWeapon is a networked EHANDLE: low 11 bits are the
entity index and the rest is the serial. Masking with 0xFFFF included serial
bits, so on later maps the weapon index became garbage (e.g. 49184/12319)
and melee killbot could not recognize the crowbar. Use 0x7FF for client and
server active-weapon handles, weapon owner checks and item-ESP weapon skips.
Document the layout in CONTEXT.
|
|
- Map transitions often leave a non-crowbar weapon active, so the killbot
silently does nothing. Log every active-weapon change as a normal
`melee: active ... crowbar=0/1` line and toast once suggesting the crowbar.
- Recover the crowbar model from the server entity when the client weapon
model index is missing/0 after a map change.
- Clear the bone cache and aim/silent locks when the local player entity
changes (map load/respawn) so stale entity/model data cannot poison melee
target selection.
|
|
- Silent aim and triggerbot set IN_ATTACK after CInput::CreateMove, so the
minimal-spread seed was never applied to their shots. Re-apply
spread_control_apply after setting the attack button, then update the CRC.
- Crossbow bolts and RPG rockets are not hitscan. Lead the selected hitbox by
target velocity * travel time; the crossbow bolt also gets its 0.05 gravity
drop (30 u/s^2). Apply the same correction to the smooth aimbot.
- Document both paths in CONTEXT.
|
|
- config_load now accepts max_dist=0 instead of clamping it to 250, so the
"unlimited" setting survives a restart.
- The Max distance slider shows "All" at 0 and its tooltip documents 0 =
unlimited; the categories log prints max=All.
- Skip unclassified entities before projecting them / computing distance.
- Keep the removed prop/crate debug logging and the user's rule that the
distance slider applies to every tag.
|
