Nim 2.0.x devel: static generic typeclass type mismatch

[mratsim]

From #370 (comment)

Overview

Command:

nim c  -d:CTT_ASM=false  --tlsEmulation=off  -d:danger  --panics:on -d:noSignalHandler  --mm:arc -d:useMalloc  --verbosity:0 --hints:off --warnings:off  --passC:-fno-semantic-interposition  --passC:-falign-functions=64  --passC:-fmerge-all-constants --threads:on  --noMain --app:staticlib  --nimMainPrefix:ctt_init_  --out:libconstantine.a --outdir:lib  --nimcache:nimcache/libconstantine_static bindings/lib_constantine.nim

It seems like there is a regression in Nim 2.0.x devel branch that breaks type-checking:

No issue in Nim 2.0.2 or 2.04

Investigation

Error

Error: type mismatch
Expression: isOne((coords: [(mres: (limbs: [0'u, 0'u, 0'u, 0'u, 0'u, 0'u])), (mres: (limbs: [
    16517514583386313282'u, 74322656156451461'u, 16683759486841714365'u,
    815493829203396097'u, 204518332920448171'u, 1306242806803223655'u]))]))
  [1] (coords: [(mres: (limbs: [0'u, 0'u, 0'u, 0'u, 0'u, 0'u])), (mres: (limbs: [
    16517514583386313282'u, 74322656156451461'u, 16683759486841714365'u,
    815493829203396097'u, 204518332920448171'u, 1306242806803223655'u]))]): QuadraticExt

Expected one of (first mismatch at [position]):
[1] func isOne(a: BigInt): SecretBool
[1] func isOne(a: ExtensionField): SecretBool
[1] func isOne(a: FF): SecretBool

The input should match:

constantine/constantine/math/extension_fields/towers.nim

Lines 124 to 129 in 976c8bb

	func isOne*(a: ExtensionField): SecretBool =
	## Constant-time check if one
	result = CtTrue
	result = result and a.coords[0].isOne()
	staticFor i, 1, a.coords.len:
	result = result and a.coords[i].isZero()

and ExtensionField is just a generic typeclass

constantine/constantine/math/extension_fields/towers.nim

Lines 34 to 48 in 976c8bb

	type
	NonResidue* = object
	## Non-Residue
	##
	## Placeholder for the appropriate quadratic, cubic or sectic non-residue
	QuadraticExt*[F] = object
	## Quadratic Extension field
	coords*: array[2, F]
	CubicExt*[F] = object
	## Cubic Extension field
	coords*: array[3, F]
	ExtensionField*[F] = QuadraticExt[F] or CubicExt[F]

The only potential buggy part is that it's called in a compile-time context:

constantine/constantine/math/extension_fields/towers.nim

Lines 1442 to 1447 in 976c8bb

	func mulCheckSparse*(a: var QuadraticExt, b: static QuadraticExt) {.inline.} =
	when isOne(b).bool:
	discard
	elif isMinusOne(b).bool:
	a.neg()
	elif isZero(c0(b)).bool and isOne(c1(b)).bool:

[mratsim]

From https://github.com/mratsim/constantine/actions/runs/8629836320/job/23654839440 in PR #368

v2.0.4 with git hash b47747d31844c6bd9af4322efe55e24fefea544c was working fine

[mratsim]

Reproduced locally, the compiler loses the concrete type here:

Some possibly interesting points:

• The b input is static generic
• The b input is instantiated via a dynamicBindSym macro
  

  constantine/constantine/math/constants/zoo_hash_to_curve.nim

  Lines 18 to 23 in 976c8bb

  	{.experimental: "dynamicBindSym".}
  	macro h2cConst*(C: static Curve, mapping: untyped, group: static Subgroup, value: untyped): untyped =
  	## Get a Hash-to-Curve constant
  	## for mapping to a elliptic curve group (G1 or G2)
  	return bindSym($C & "_h2c_" & $mapping & "_" & $group & "_" & $value)

• The isOne proc does not import directly or via its import graph the h2cConst input which might lead to generic sandwich issues

[mratsim]

[mratsim]

Fixed upstream in nim-lang/Nim#23852
And workaround in #434