feat: add Convar<T> for real convar creation/deletion/access

[roflmuffin]

This PR adds Convar<T> which is a strongly typed variant of ConVar which provides better safety (no raw pointer accesses), as well as the ability to create real-convars that exist within the engine (and can be found using find foo in the console) amongst other things.

TODO:

• Add BasePlugin based registration so that convars are automatically unloaded (similar to FakeConvar). Should the .Create static be restricted?

[ubbemir]

This looks awesome. Would be great to also have a list of known ConVars that utilize this to reduce mistake of accidentally using the wrong type with a ConVar. This however would be fitting in another PR I suppose.

[ubbemir]

Magic numbers should not be written by themselves. Maybe make them a constant with a name to clarify?

[KillStr3aK]

these are hash calculated native identifiers

[KillStr3aK]

This looks awesome. Would be great to also have a list of known ConVars that utilize this to reduce mistake of accidentally using the wrong type with a ConVar. This however would be fitting in another PR I suppose.

https://cs2.poggu.me/dumped-data/convar-list

[ubbemir]

This looks awesome. Would be great to also have a list of known ConVars that utilize this to reduce mistake of accidentally using the wrong type with a ConVar. This however would be fitting in another PR I suppose.

https://cs2.poggu.me/dumped-data/convar-list

Yes I know of that. But implement some of these at least into C# as statically typed to make use of the strength of static typing and static code analysis to reduce mistakes and catch erronous usage at compile time.

For example would it be suitable to do something similar to this?

public static class KnownConVars
{
    public static ConVar<bool> SvCheats = new ConVar<bool>("sv_cheats");
    public static ConVar<int> MpWarmupTime = new ConVar<int>("mp_warmuptime");
    public static ConVar<float> MaxSpeed = new ConVar<float>("sv_maxspeed");

    // Example: Add other ConVars here...
}

[Copilot]

Copilot review overview

🟡 Changes recommended

Null native objects can crash the host, deletion can invalidate engine-owned ConVars, and native allocations are leaked.

Review effort: Balanced
Findings: 3

Open (3)

• Delete permits removal of engine-owned ConVars · New
• Null DefaultValue can crash native creation · New
• Null vector or angle values can crash native assignment · New

Resolved since last review (1)

• ConVar change dispatch is not re-entrant

Previously missed (4)

In code that hasn't changed since last review

Unload misses ConVars stored in base or derived fields

managed/​CounterStrikeSharp.API/​Core/​BasePlugin.cs:564

Filtering on an exact declared ConVar<> type skips owned values stored in a ConVarBase field or a subclass of ConVar<T>, so those public plugin fields are never deleted on unload. The loop already checks the runtime value and Owned; select every field assignable to ConVarBase instead.

Returned vector values bypass validation and expose storage

managed/​CounterStrikeSharp.API/​Modules/​Cvars/​ConVarOfT.cs:288

For vector and angle types this returns a NativeObject backed directly by the engine's value pointer. Mutating the returned object's components therefore changes the ConVar without going through SetAs, bypassing bounds enforcement and OnConVarChanged notifications; retaining it also exposes storage after deletion. Return a detached value copy (or an immutable view) so mutations must use the Value setter.

Created ConVar wrappers are leaked

src/​scripting/​natives/​natives_convars.cpp:624

Each creation branch allocates a CConVar<T> with new, but this pointer is only kept in this local variable and is discarded when the native returns. DeleteConVar merely calls Invalidate(), so repeated plugin reloads or create/delete cycles permanently leak every allocated wrapper (and potentially its owned storage). Keep the created object in an ownership registry and destroy it during deletion/shutdown.

Deletion sentinel comment is incorrect

managed/​CounterStrikeSharp.Tests.Native/​RealConVarTests.cs:33

This comment names the wrong deletion sentinel: ConVarBase.Delete sets AccessIndex to ushort.MaxValue, matching the check below, not zero.