[3.0] Rework naming system by Exanite · Pull Request #2565 · dotnet/Silk.NET

Exanite · 2026-04-10T13:08:14Z

Summary of the PR

This PR combines my 3 previous naming-related PRs into a single PR.
This is because the 3 PRs contain interdependent changes and should be merged as a single unit.
That said, each of the 3 PRs focus on different aspects of the naming system and serve as supplementary documentation for the different changes.

Related issues, Discord discussions, or proposals

PRs contained by this PR:

Discord thread where I noted down my thoughts and decisions: https://discord.com/channels/521092042781229087/1485943448069738608

Further Comments

This PR does not introduce any new changes on top of the changes made in the 3 contained PRs.
However, prefer reviewing the diff through this PR since this is the PR that is intended to be merged.

…ld used by other bindings The default acronym threshold was changed during #29, which was merged as part of dotnet#2503.

Considering we decided to follow Microsoft's Framework Design Guidelines (acronym threshold of 2) for the bindings and rest of the API, might as well be consistent here.

… separation hack (no longer necessary)

…ferenced affixes

…mbers are normalized

This lets us handle prefixing and prettification separately, which notably is important if we add prefixes after prettification. We want to prefix the final name, not the intermediate name in this case.

…igured (cherry picked from commit bc94260)

…ous that it is not the last trimmer

This no longer makes sense to keep and enabling features by baseline version seems fiddly. If we need to toggle features for newer versions, we can explicitly add a boolean config option.

…essor and ReapplyAffixesNameProcessor

Kinda a cop out decision, but it keeps thing simple (and thus maintainable) and implementing it fully seems overkill for what we need.

… for function pointer types are named

This is because we no longer output the separating underscore in ExtractNestedTyping

… of INameTrimmer registrations

…ed-struct-name-affixes

…nvention Note that the goal is to eventually remove the name overrides for the `EFXEAXREVERBPROPERTIESflLateReverbPan` and `-Delegate` cases entirely. This is because these are theoretically possible to handle automatically and the reason it doesn't work is due to an edge case interaction with the name override system. See the "Tasks" section here for more info: dotnet#2555

…ot getting chosen properly

…t comments/variable names to be clearer

… and update test cases to reflect expected behavior

…ffixes yet

I expected this to be an issue *eventually*, but apparently it also affects the SDL output already. This is probably because SDL has multiple overloads of the SDL_main method, which my test case does not test for.

This new snapshot matches the expected output that I wrote in the comments earlier. I just forgot to update the snapshot.

…penAL overrides that are now unnecessary

… update test to indicate that one level of scoping is now supported

…e verbatim value of the referenced name

…e code and config

Exanite

Self code review complete. The review comments I added provide additional context on the changes I made, but the code itself should make sense without them.

Exanite · 2026-04-11T21:13:50Z

sources/OpenAL/OpenAL/al/BufferCallbackSOFT.gen.cs


    public BufferCallbackSOFT(delegate* unmanaged<void*, void*, int, int> ptr) => Pointer = ptr;

-    public BufferCallbackSOFT(BufferCallbackDelegateSOFT proc) =>


Changed to match my decision here: https://discord.com/channels/521092042781229087/1485943448069738608/1487577229121818766

In short, this reordering is because -Delegate is a variant or part of an existing type, similar to how inline array structs are part of their parent struct.

Notably, handle structs are named differently (BufferHandleEXT). This is because the -Handle suffix is a modification to an existing type.

This is a rule that we can consistently follow and also improves discoverability in IDEs since extensions to a type share a common prefix.

Exanite · 2026-04-11T21:17:05Z

sources/SilkTouch/SilkTouch/Mods/Common/ModLoader.cs

        name switch
        {
-            nameof(ChangeNamespace) => typeof(ChangeNamespace),
+            nameof(AddApiProfiles) => typeof(AddApiProfiles),


I alphabetically sorted the names here. This should be easier to read through and cause less diff noise in the future. Even if we forget to sort it when adding new names, sorting it again won't cause as much noise as the first sort.

Exanite · 2026-04-11T21:17:58Z

sources/SilkTouch/SilkTouch/Mods/Common/ModUtils.cs

    /// <param name="type">The type syntax.</param>
    /// <returns>The discriminator string.</returns>
-    public static string DiscrimStr(SyntaxTokenList? toks, TypeSyntax? type) =>
+    public static string GetMethodDiscriminator(SyntaxTokenList? toks, TypeSyntax? type) =>


I renamed this because I always had to read the name twice to figure out what it meant. This should help with skimming the code.

Exanite · 2026-04-11T21:20:17Z

sources/SilkTouch/SilkTouch/Mods/ExtractHandles.cs

+/// <para/>
+/// See <see cref="TransformHandles"/> for applying further transformations.
+/// </summary>
+public class ExtractHandles(ILogger<ExtractHandles> logger) : Mod


This mod is effectively a cut and paste from TransformHandles.
This is part of my move to separate extraction/type creation from type transformation.

Originally, this was because I thought this was needed for better nested struct handling, but it turned out unnecessary. I kept it because it makes the codebase cleaner.

Note I removed the AssumeMissingTypesOpaque config option when moving the code to this mod. This is because the false value for the config option is equivalent to disabling the entire mod.

Exanite · 2026-04-11T21:22:03Z

sources/SilkTouch/SilkTouch/Mods/ExtractNestedTyping.cs

                }

-                var iden = $"{node.Identifier}_{match.Groups[1].Value}";
+                var iden = $"{node.Identifier}{match.Groups[1].Value}";


The underscore here is no longer necessary because the affix metadata lets us distinguish the two parts during prettification.

Exanite · 2026-04-12T16:13:43Z

tests/SilkTouch/SilkTouch/Naming/NameTrimmerTests.cs

-
-namespace Silk.NET.SilkTouch.UnitTests.Naming;
-
-public class NameTrimmerTests : NameTrimmer


Test cases in this class were ported over to the new IdentifySharedPrefixTests class.

Exanite · 2026-04-12T16:14:58Z

tests/SilkTouch/SilkTouch/Naming/NamePrettifierTests.cs

+    /// in favor of calling <see cref="NameUtils.PrefixIfStartsWithNumber"/> separately.
+    /// </summary>
    [Theory]
-    // C# identifiers cannot start with numbers


The tests in this file were updated to match the change I made to Prettify (which now processes name fragments instead of full names).

Exanite · 2026-04-12T16:24:18Z

generator.json

+      "PrettifyNames": {
+        "Affixes": {
+          "SharedPrefix": {
+            "Remove": true


For SharedPrefix, we can either configure it as Remove or as a Discriminator. Both have the same effect until names conflict and no other discriminators are available. Fortunately, we no names conflict in this way right now and I assume this case will be very rare, so I'm defaulting to Remove.

Exanite · 2026-04-12T16:26:08Z

generator.json

-          "ContainerType": "GL"
-        },
+          "ContainerType": "GL",
+          "SeparableTargetEXT": "GL"


This is new override is required since my new name underscorer/splitter handles 2D differently. It outputs 2_D instead of 2D since 2_D is arguably more consistent. I explain why here: https://discord.com/channels/521092042781229087/1485943448069738608/1489457083857506336

It's easier to handle the common case consistently while handling edge cases using overrides.

Exanite · 2026-04-12T16:27:13Z

generator.json

          "ALDEBUGPROCEXT": "DebugProcEXT",
-          "ALDEBUGPROCEXTDelegate": "DebugProcDelegateEXT",
-          "LPALFOLDBACKCALLBACK": "FoldbackCallback",
-          "LPALFOLDBACKCALLBACKDelegate": "FoldbackCallbackDelegate"


Most of these overrides are no longer necessary because of referenced affixes. Overrides to the referenced name automatically propagate to names that reference them.

Exanite added 30 commits March 24, 2026 05:59

Update OpenAL name overrides to match the 2 character acronym thresho…

3ac76f2

…ld used by other bindings The default acronym threshold was changed during #29, which was merged as part of dotnet#2503.

Start of feature/nested-struct-name-affixes branch

634e736

Cleanup leftover return statement

35f0f59

Split TransformHandles into two mods

cd2ccfd

Rename UseDSL to UseDsl

4b119b2

Considering we decided to follow Microsoft's Framework Design Guidelines (acronym threshold of 2) for the bindings and rest of the API, might as well be consistent here.

Cleanup unused property in TransformHandles

f3900b9

Cleanup usages of redundant collections expressions in TransformHandles

851c3ff

Add NameAffixer.AddResolvedNameAffix() for handling compound names

2f018ff

Use AddResolvedNameAffix for extracted function pointer delegate types

50a4115

Use AddResolvedNameAffix for extracted nested structs and remove name…

8388878

… separation hack (no longer necessary)

Rename resolved affix to referenced affix and add parsing code for re…

674c61a

…ferenced affixes

Add extra ConsecutiveNumbers test case documenting that Xs between nu…

d2ae93d

…mbers are normalized

Fix incorrect declarationOrder implementation

d51a68e

Change Prettify to not prefix identifiers starting with numbers

a30ac45

This lets us handle prefixing and prettification separately, which notably is important if we add prefixes after prettification. We want to prefix the final name, not the intermediate name in this case.

Change the list and order of INameTrimmers used to be statically conf…

fb98f40

…igured (cherry picked from commit bc94260)

Replace now outdated comment

68153ff

Move PrettifyNamesTrimmer higher up in the class to make it more obvi…

e0fb52e

…ous that it is not the last trimmer

Remove INameTrimmer.Version

044c84b

This no longer makes sense to keep and enabling features by baseline version seems fiddly. If we need to toggle features for newer versions, we can explicitly add a boolean config option.

Add PrefixIfStartsWithNumberTrimmer

e9b2377

Remove TrimmerBaseline config option

464703c

Rename INameTrimmer to INameProcessor and update rest of codebase

adf4eee

Simplify name affixer name processor names to be StripAffixesNameProc…

f321238

…essor and ReapplyAffixesNameProcessor

Shorten names a bit more for readability

d8e808d

Decide to only support references within the same scope depth

6983172

Kinda a cop out decision, but it keeps thing simple (and thus maintainable) and implementing it fully seems overkill for what we need.

Restrict feature set of AddReferencedNameAffix even more for simplicity

14d9d65

Implement referenced affixes and change how -Delegate types generated…

0228a56

… for function pointer types are named

Update OpenAL name overrides in generator.json

5a4b6cd

This is because we no longer output the separating underscore in ExtractNestedTyping

Update doc comment in ServiceCollectionExtensions to indicate removal…

16ba7ea

… of INameTrimmer registrations

Merge branch 'feature/update-openal-name-overrides' into feature/nest…

3ced75c

…ed-struct-name-affixes

Exanite added 21 commits April 8, 2026 12:59

Implement non-method/method conflict resolution and fix secondaries n…

0858b9a

…ot getting chosen properly

Ensure that the temporary data collections are allocated once and edi…

851bbf3

…t comments/variable names to be clearer

Implement "-Value" suffixing for non-methods conflicting with methods…

8a61406

… and update test cases to reflect expected behavior

Decide it's not worth implementing cross-scope name referencing for a…

4b84780

…ffixes yet

Remove unused RenamedType struct and unused MemberData.Name property

e603c86

Fix -Value suffixing not working properly

9a130c3

I expected this to be an issue *eventually*, but apparently it also affects the SDL output already. This is probably because SDL has multiple overloads of the SDL_main method, which my test case does not test for.

Update test snapshot

c3f0462

This new snapshot matches the expected output that I wrote in the comments earlier. I just forgot to update the snapshot.

Also resolve referenced affixes from the final set of names and the O…

7f5bab0

…penAL overrides that are now unnecessary

Add ReapplyAffixesProcessor.TryResolveName method (refactor)

85daff5

Change ReapplyAffixesProcessor to use one global topological sort and…

d630d1f

… update test to indicate that one level of scoping is now supported

Fix issue where overrides break the topological sort

2574cc5

Reorder tests in file

6f38318

Add test case: SuccessfullyUsesReferencedAffixes_WhenOverridden

dfa7b4e

Rename NameProcessorContext/NameDataVisitor.Scopes to Names

f75ac4b

Work on refactoring name affix processors

e344cc5

Add local OutputName function

c43ac7d

Implement the option to prettify individual affixes

13db50d

Update generator config now that affixes are prettified by default

aac49dd

Don't prettify referenced affixes by default since we usually want th…

51d6b7d

…e verbatim value of the referenced name

Change affixes to not be prettified by default since it simplifies th…

0ba719c

…e code and config

Empty commit

86b8bbd

github-project-automation bot added this to Silk.NET Development Tracker Apr 10, 2026

github-project-automation bot moved this to Todo in Silk.NET Development Tracker Apr 10, 2026

Exanite added 3 commits April 11, 2026 07:58

Merge branch 'develop/3.0' into feature/rework-naming-system

fd4db1f

Update snapshot to match expected stated by comment

a737fe6

Adjust doc comment wording for IdentifySharedPrefixes._passCount

6c6baca

Exanite commented Apr 12, 2026

View reviewed changes

Exanite marked this pull request as ready for review April 12, 2026 16:28

Exanite requested a review from a team as a code owner April 12, 2026 16:28

curin approved these changes Apr 13, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[3.0] Rework naming system#2565

[3.0] Rework naming system#2565
Exanite wants to merge 155 commits intodotnet:develop/3.0from
Exanite:feature/rework-naming-system

Exanite commented Apr 10, 2026

Uh oh!

Exanite left a comment

Uh oh!

Exanite Apr 11, 2026

Uh oh!

Exanite Apr 11, 2026

Uh oh!

Exanite Apr 11, 2026

Uh oh!

Exanite Apr 11, 2026

Uh oh!

Exanite Apr 11, 2026

Uh oh!

Exanite Apr 11, 2026

Uh oh!

Exanite Apr 12, 2026

Uh oh!

Exanite Apr 12, 2026

Uh oh!

Exanite Apr 12, 2026

Uh oh!

Exanite Apr 12, 2026

Uh oh!

Exanite Apr 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants


		public BufferCallbackSOFT(delegate* unmanaged<void, void, int, int> ptr) => Pointer = ptr;

		public BufferCallbackSOFT(BufferCallbackDelegateSOFT proc) =>


		namespace Silk.NET.SilkTouch.UnitTests.Naming;

		public class NameTrimmerTests : NameTrimmer

Uh oh!

Conversation

Exanite commented Apr 10, 2026

Summary of the PR

Related issues, Discord discussions, or proposals

Further Comments

Uh oh!

Exanite left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants