Refactor Int32 to use Int32Native pattern matching Int64 architecture by Copilot · Pull Request #12828 · HaxeFoundation/haxe

Copilot · 2026-03-17T06:15:15Z

Reworks haxe.Int32 to follow the same Native delegation pattern established for Int64 in #12826, verifying the system's flexibility for future numeric type unification.

Architecture

haxe/Int32.hx → thin @:op abstract over Int32Native, no platform logic. Documents platform representations (native 32-bit on C++/JVM/HL, clamped on JS/PHP/Python/Lua, Float auto-promoted on Neko).
haxe/numeric/Int32Native.hx → cross-platform abstract(Int) with masking (JS |0, PHP sign-extend, Python modular arithmetic, Lua clampInt32). Also used by Neko, where NekoVM auto-promotes values exceeding 31-bit range to Float while preserving correct 32-bit arithmetic.
haxe/numeric/Int32Direct.hx → shared no-masking implementation for targets where Int is natively 32-bit. Uses ~x+1 for negation to avoid a CPPIA unary-minus overflow bug.
{jvm,hl,cpp}/_std/haxe/numeric/Int32Native.hx → single-line overrides: typedef Int32Native = Int32Direct.
{jvm,hl,cpp}/_std/haxe/numeric/Int64Native.hx → removed unnecessary @:coreApi(check = Off) (private types get the Implied check which silently ignores failures).

Tests

New TestInt32 class: overflow wrap, increment/decrement at boundaries, bitwise ops, shifts, unsigned comparison, mixed-type arithmetic, Float conversion, regression coverage from Fix for #7485 - Int32 negate missing clamp #7491
Regression tests for platform-specific issues: multiplication overflow ([cpp] Inconsistent overflow behavior for Int32 #10780), XOR with dynamic shift on Python ([python] Int32 wrong result for shift left operations #10995), |= not clamping on Python ([python] "<Int32> |= <Int>" not clamping result to 32 bits #5938)
Verified on interp, JS, JVM, Neko, Python, PHP, C++ (including CPPIA)

Example

// Int32.hx is now a pure delegation layer:
@:op(A + B) private static inline function add(a:Int32, b:Int32):Int32
    return Int32Native.add(a, b);

// Int32Direct provides the shared native implementation (C++/JVM/HL):
public static inline function add(a:Int32Direct, b:Int32Direct):Int32Direct
    return cast((a : Int) + (b : Int));

// Int32Native (cross-platform) provides masking for scripting targets:
public static inline function add(a:Int32Native, b:Int32Native):Int32Native
    return clamp((a : Int) + (b : Int));

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.

- Create std/haxe/numeric/Int32Native.hx as cross-platform abstract(Int) with masking - Create JVM, HL, C++, Neko target overrides in _std directories - Refactor std/haxe/Int32.hx to thin @:op layer delegating to Int32Native - Add comprehensive TestInt32 unit test class - All tests passing on interp, JS, JVM, Neko, Python, PHP Co-authored-by: Simn <634365+Simn@users.noreply.github.com>

Co-authored-by: Simn <634365+Simn@users.noreply.github.com>

tests/unit/src/unit/TestInt32.hx

std/neko/_std/haxe/numeric/Int32Native.hx

std/cpp/_std/haxe/numeric/Int32Native.hx

std/neko/_std/haxe/numeric/Int32Native.hx

…docs and @:coreApi - Move #if !cpp to cover entire testArrayIndexWithInt32 function (avoids utest "no assertions") - Remove Neko-specific Int32Native override (cross-platform version works correctly via Float auto-promotion) - Remove doc comments from target-specific Int32Native files (docs belong on main Int32 type) - Remove unnecessary @:coreApi(check = Off) from private types (Implied check handles these) - Add platform representation docs to main Int32.hx Co-authored-by: Simn <634365+Simn@users.noreply.github.com>

Simn · 2026-03-17T08:39:41Z

@copilot

Let's really remove the unnecessary @:coreApi from Int64Native files. I know it's technically more about the parent PR but that's fine.
I'm noticing that we have code duplication for Int32Native on C++, HL and JVM now. Since we are typedeffing it anyway, let's add std/haxe/numeric/Int32Direct.hx with the common implementation and then point all 3 targets to that. I don't insist on the "Direct" suffix so if you have a better name use that instead!

The only failing target is now C++ which we'll have to look into still. We first need to identify what actually changed because this was fine before. Along those lines, I've checked our GitHub issues list for int32 occurrences and paste the links here so we can check if all these cases are covered by the tests:

#10780
#10943
#10995
#5938
#8879

Simn · 2026-03-17T11:09:01Z

@copilot Please continue!

…ove @:coreApi from Int64Native, add issue regression tests Co-authored-by: Simn <634365+Simn@users.noreply.github.com>

Copilot · 2026-03-17T11:43:13Z

@copilot Please continue!

Done in 6dd7641:

@:coreApi(check = Off) removed from all three Int64Native target files (JVM, HL, C++).
std/haxe/numeric/Int32Direct.hx created with the shared no-masking implementation. All three target overrides (cpp/jvm/hl/_std/haxe/numeric/Int32Native.hx) are now a single line each: typedef Int32Native = Int32Direct.
C++ CPPIA bug fixed: the root cause was that CPPIA's unary minus -(x:Int) can return a value wider than 32 bits for INT_MIN, giving 2147483648 instead of wrapping to -2147483648. Int32Direct.neg now uses ~x+1 (the same formula the original Int32 used), which stays within the native 32-bit range on CPPIA.
Regression tests added for the linked issues:
- testMulOverflow_Issue10780: multiplication overflow on C++ (257 * 0x01010101)
- testXorWithDynamicShift_Issue10995: XOR with dynamic shift on Python
- testOrEqualsNotClamping_Issue5938: |= not clamping on Python
Issues [CPP] Int/Int32 inconsistent results for shift left operations #10943 and [cpp] wierd addition results upon int overflow #8879 relate to compiler/optimizer behaviour rather than the library implementation, so no specific tests were added for those.

Initial plan

eb53a67

Copilot AI assigned Copilot and Simn Mar 17, 2026

Copilot AI mentioned this pull request Mar 17, 2026

Unify numeric types API across all targets #12825

Open

Copilot started work on behalf of Simn March 17, 2026 06:15 View session

Copilot AI and others added 2 commits March 17, 2026 07:28

Address code review: add comments for Neko negation and C++ test skip

e5914c2

Co-authored-by: Simn <634365+Simn@users.noreply.github.com>

Copilot AI changed the title ~~[WIP] Unify numeric types API across all targets~~ Refactor Int32 to use Int32Native pattern matching Int64 architecture Mar 17, 2026

Copilot AI requested a review from Simn March 17, 2026 07:30

Copilot finished work on behalf of Simn March 17, 2026 07:30

Simn requested changes Mar 17, 2026

View reviewed changes

Copilot started work on behalf of Simn March 17, 2026 08:02 View session

Copilot AI requested a review from Simn March 17, 2026 08:14

Copilot finished work on behalf of Simn March 17, 2026 08:14

Copilot started work on behalf of Simn March 17, 2026 08:39 View session

Copilot started work on behalf of Simn March 17, 2026 11:09 View session

Copilot AI mentioned this pull request Mar 17, 2026

Improve CI investigation instructions to prevent agents from misreading workflow run status #12833

Merged

Fix CPPIA neg overflow, deduplicate Int32Native with Int32Direct, rem…

6dd7641

…ove @:coreApi from Int64Native, add issue regression tests Co-authored-by: Simn <634365+Simn@users.noreply.github.com>

Copilot finished work on behalf of Simn March 17, 2026 11:47

Simn marked this pull request as ready for review March 17, 2026 12:16

Simn merged commit 338bed4 into haxe-numeric Mar 17, 2026
61 checks passed

Simn deleted the copilot/sub-pr-12825 branch March 17, 2026 12:18

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Refactor Int32 to use Int32Native pattern matching Int64 architecture#12828

Refactor Int32 to use Int32Native pattern matching Int64 architecture#12828
Simn merged 5 commits intohaxe-numericfrom
copilot/sub-pr-12825

Copilot AI commented Mar 17, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Simn commented Mar 17, 2026

Uh oh!

Simn commented Mar 17, 2026

Uh oh!

Copilot AI commented Mar 17, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

Copilot AI commented Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Architecture

Tests

Example

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Simn commented Mar 17, 2026

Uh oh!

Simn commented Mar 17, 2026

Uh oh!

Copilot AI commented Mar 17, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Copilot AI commented Mar 17, 2026 •

edited

Loading