feat(csharp): generate ToXml()/FromXml()/builders for xml-encoded object types (TwiML)

[devin-ai-integration]

Description

Linear ticket: Refs (none)

C# counterpart of #17734 (Python), #17755 (Java) and #17785 (TypeScript): object types carrying IR encoding.xml metadata (from #17732, e.g. Twilio TwiML) now generate XML serialization, parsing and fluent child builders. Types without XML metadata are untouched; the XML runtime files are only emitted when the IR contains at least one xml-encoded type.

Generated shape (from the csharp-xml-twiml fixture):

public record Dial : IJsonOnDeserialized, IXmlNode
{
    // ...existing JSON properties...
    [JsonIgnore] public Dictionary<string, string> AdditionalAttributes { get; set; } = new();
    [JsonIgnore] public List<XmlElement> AdditionalChildren { get; set; } = new();

    public XElement ToXElement();            // <tw:Dial xmlns:tw="https://www.twilio.com/twiml" ...>
    public string ToXml();
    public static Dial FromXml(string xml);   // validates root name, rejects DTDs
    public static Dial FromXElement(XElement element);

    public Dial AddNumber(Number number);    // `Add` prefix only where the tag collides with a property (Dial.Number text)
    public Dial Client(Client client);
    public Dial AddChild(XmlElement child);  // escape hatch for arbitrary elements
}

Design notes (hard to see from the diff):

• Runtime is System.Xml / System.Xml.Linq only, no new dependencies. XmlUtils is internal (core); IXmlNode and XmlElement are public (public core).
• Unknown attributes/children are preserved in AdditionalAttributes / AdditionalChildren and re-emitted by ToXElement() (typed children first, unknown after); unknown children inside a wrapped list (<Numbers><Extension/></Numbers>) are kept without duplicating known items.
• Child unions dispatch on element local name in both modes: OneOf<Say, Dial, ...> (default) and generated undiscriminated-union classes (use-undiscriminated-unions: true) — the fixture runs both.
• FromXElement uses an object initializer, so the 256-attribute Wide type parses without a giant constructor.
• Enum wire values, Set/List attributes with listSeparator, namespaces/prefixes and xml:-prefixed attributes round-trip. Fixed XmlElement.ToXElement() to route attributes through XmlUtils.SetAttribute so xml:lang doesn't throw.
• IR pin bumped 67.21.0 → 67.25.0 in the four C# packages.

Changes Made

• generators/csharp/base/src/asIs/{IXmlNode,XmlElement,XmlUtils}.Template.cs + test/Xml/XmlElementTests.Template.cs (runtime + shipped unit tests), registered in AsIs.ts; emitted conditionally via GeneratorContext.hasXmlTypes() in model + sdk contexts
• generators/csharp/model/src/object/XmlObjectGenerator.ts: generates the interface impl, ToXElement/ToXml/FromXml/FromXElement, child-union parsers and builder methods; wired from ObjectGenerator only when typeDeclaration.encoding.xml != null
• codegen: System.Xml.Linq.XElement extern + XmlUtils/XmlElement/IXmlNode references
• New csharp-xml-twiml test definition (Response/Say/Dial/Number/Client/Pause/Hangup/Wide/PlainObject: attrs, text, tw: namespace, wrapped Numbers, space-separated lists, empty elements, 256-attr type, non-xml type) with two seed configs; seed output + ir-to-jsonschema snapshots committed
• Changelog: generators/csharp/sdk/changes/unreleased/feat-xml-encoded-types.yml
• Updated README.md generator (if applicable)

Testing

• Unit tests added/updated — XmlElementTests ships into every generated SDK that has xml types (rendering/parsing, namespaces, declarations/comments, malformed XML, DTD rejection, wrong root, scalar/list parsing, wrapper unknown-child preservation)
• Manual testing completed
  • pnpm seed test --generator csharp-sdk --fixture csharp-xml-twiml --skip-scripts → 2/2 passed
  • dotnet test on generated no-custom-config output with a temporary TwiML round-trip suite (attrs/text/escaping, tw: namespace, wrapped lists, list attrs, enums, unknown attrs+children incl. xml:lang, Wide, FromXml(ToXml(x)) equality, error cases): Passed: 186, Failed: 0; use-undiscriminated-unions output: Passed: 186, Failed: 0 (round-trip suite adapted to ResponseChildrenItem)
  • Generated project builds for net462;net8.0;net9.0;netstandard2.0
  • pnpm seed test --generator csharp-sdk --fixture exhaustive → 8/8, zero xml-related output diff (no regression for non-xml types)
  • pnpm turbo run compile --filter @fern-api/fern-csharp-sdk --filter @fern-api/fern-csharp-model → 16/16 tasks

Link to Devin session: https://app.devin.ai/sessions/311fe13ae5434245ada665a9b824809b
Open in Devin Desktop: https://app.devin.ai/desktop/session/311fe13ae5434245ada665a9b824809b?variant=devin

---

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

Devin Review found 3 potential issues.

1 flag not posted on this PR by your GitHub settings — view it in Devin Review. (Configure)

[github-actions]

SDK Generation Benchmark Results

Comparing PR branch against median of 5 nightly run(s) on main (latest: 2026-09-18T04:06:24Z).

Full benchmark table (click to expand)

Generator	Spec	main (generator)	main (E2E)	PR (generator)	Delta
csharp-sdk	square	82s (n=5)	113s (n=5)	57s	-25s (-30.5%)
go-sdk	square	124s (n=5)	295s (n=5)	110s	-14s (-11.3%)
java-sdk	square	187s (n=5)	268s (n=5)	214s	+27s (+14.4%)
php-sdk	square	65s (n=5)	N/A	50s	-15s (-23.1%)
python-sdk	square	142s (n=5)	257s (n=5)	170s	+28s (+19.7%)
ruby-sdk-v2	square	93s (n=5)	112s (n=5)	75s	-18s (-19.4%)
rust-sdk	square	181s (n=5)	204s (n=5)	187s	+6s (+3.3%)
swift-sdk	square	58s (n=5)	435s (n=5)	49s	-9s (-15.5%)
ts-sdk	square	136s (n=5)	139s (n=5)	111s	-25s (-18.4%)

main (generator): generator-only time via --skip-scripts (includes Docker image build, container startup, IR parsing, and code generation — this is the same Docker-based flow customers use via fern generate). main (E2E): full customer-observable time including build/test scripts (nightly baseline, informational). Delta is computed against generator-only baseline.
⚠️ = generation exited with a non-zero exit code (timing may not reflect a successful run).
Baseline from nightly runs on main (latest: 2026-09-18T04:06:24Z). Trigger benchmark-baseline to refresh.
Last updated: 2026-09-18 19:22 UTC

[github-actions]

Docs Generation Benchmark Results

Comparing PR branch against median of 5 nightly run(s) on main (latest: 2026-09-18T04:06:24Z).

Fixture	main	PR	Delta
docs	266.3s (n=5)	253.4s (35 versions)	-12.9s (-4.8%)

Docs generation runs fern generate --docs --preview end-to-end against the benchmark fixture with 35 API versions (each version: markdown processing + OpenAPI-to-IR + FDR upload).
Delta is computed against the nightly baseline on main.
Baseline from nightly run(s) on main (latest: 2026-09-18T04:06:24Z). Trigger benchmark-baseline to refresh.
Last updated: 2026-09-18 19:22 UTC