feat(typespec-go): support paging with a relative nextLink

[tadelesh]

Resolves #4994.

What

Adds support for paging with a relative nextLink to the Go emitter, and enables the previously disabled azure/core/page withRelativeNextLink Spector test.

Problem

runtime.FetcherForNextLink builds the follow-up request with runtime.NewRequest(ctx, verb, nextLink), which rejects a non-absolute URL with no Host in request URL. A service that returns a next link relative to its endpoint (e.g. /azure/core/page/with-relative-next-link/page/2, per the Spector scenario) therefore failed on the second page. This is why TestPageClient_NewWithRelativeNextLinkPager was commented out with a TODO: runtime.FetcherForNextLink doesn't support relative next link URLs.

The code model and codegen already carry everything needed (PageableStrategyNextLink + the nextLink field path); only the emitted fetcher was missing the resolution step. Relative-ness is a property of the runtime value, not something TCGC can express statically, so the check has to be in generated code.

Fix

In packages/typespec-go/src/codegen/core/operations.ts, emitPagerDefinition now emits a resolution step before calling runtime.FetcherForNextLink:

nextLink := ""
if page != nil {
    nextLink = *page.NextLink
}
// the service can return a next link that's relative to the endpoint, however
// runtime.FetcherForNextLink requires an absolute URL, so resolve it here.
if nextLink != "" {
    if u, err := url.Parse(nextLink); err == nil && !u.IsAbs() {
        nextLink = runtime.JoinPaths(client.endpoint, nextLink)
    }
}
resp, err := runtime.FetcherForNextLink(ctx, client.internal.Pipeline(), nextLink, ...)

Notes:

• The endpoint expression is client.<endpointParam> for data-plane clients and client.internal.Endpoint() for ARM clients, matching what *CreateRequest already uses.
• runtime.JoinPaths is the same helper every generated *CreateRequest uses to combine the endpoint with a path, so relative links are resolved consistently with first-page requests (this also matches how the Python/JS emitters append a relative next link to the endpoint). It preserves the next link's query string.
• Absolute next links take the u.IsAbs() fast path and are passed through byte-for-byte, so behavior for compliant services is unchanged.
• Skipped when the pager uses a custom next page operation (FetcherForNextLinkOptions.NextReq), since that operation builds the request (and joins the endpoint) itself, and when the client has no single endpoint (the Autorest-only templatedHost compat case).
• For LRO pagers the next link is now hoisted into a local so it can be resolved.

Tests

• Enabled TestPageClient_NewWithRelativeNextLinkPager in test/azure-http-specs/azure/core/azurepagegroup/custom_client_test.go; it passes against the Spector mock server (2 pages, second fetched via the relative link).
• Updated the pageable-lro and tenant-resource unit scenario snapshots.
• Validated: pnpm vitest run (49 unit tests pass); full regeneration of all Spector modules followed by pnpm go-test — all generated modules pass.

[github-actions]

All changed packages have been documented.

• ✅ @azure-tools/typespec-go

Show changes

@azure-tools/typespec-go - feature ✏️

Support paging with a relative nextLink. Pagers now resolve a next link that's relative to the client endpoint before fetching the next page; absolute next links are unchanged.

[github-actions]

Go emitter diff

Baseline gh:f4ad548c39ac22e47f1f98e574b774fb61460fbd vs this PR.

Diff summary: 18 file(s), +204 / -5

Rendered diff: inline on the run summary, or the emitter-diff-go-html artifact.

Informational check (core/eng/emitter-diff); does not block the PR.

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@azure-tools/typespec-go@5079

commit: a8fd511

[tadelesh]

@jhendrixMSFT I remembered that you said we already support it in swagger cases. But I could not find the related code. Current implementation will change all the generated code, which I do not prefer. Any suggestions?

[azure-sdk-automation]

You can try these changes here

🛝 Playground	🌐 Website

[jhendrixMSFT]

We shouldn't be emitting comments.

[jhendrixMSFT]

A few problems here.

• err is discarded when not nil
• we parse just to check IsAbs() then throw that away
  • this happens for all paths and IIRC relative paths are rare

Do we know at codegen time if the path is relative?

[tadelesh]

No, we could not know that. It's kind of a runtime behavior. I remembered that you said we already support it in swagger cases. But I could not find the related code. Current implementation will change all the generated code, which I do not prefer. Any suggestions?