fix(cli-generator): apply result extraction to --page-all output

[devin-ai-integration]

Description

Stack 2/5 ("Benchling CLI fixes", split out of #17620). Stacked on #17621.

--page-all emitted raw page envelopes ({"items": [...], "nextToken": "..."}) instead of the extracted items a plain list call prints, so --format csv|table were unusable for paginated output.

Changes Made

• executor.rs: item extraction is now applied per page, before output, by a PageItemsPath decided once on the first page and reused for every later page:

  enum PageItemsPath { Undecided, Body(path), Value, Property(key), None }
  decide(body, output_val, declared_results):
    declared `x-fern-pagination.results`  -> Body(path)   # resolved against the full body, before x-fern-sdk-return-value
    output_val is array                   -> Value        # already extracted
    output_val has exactly 1 array prop   -> Property(k)  # ignores `_`-prefixed / nextPageToken / kind
    otherwise (0 or ≥2 candidates)        -> None         # ambiguous: page left untouched
  items(page): array -> items; absent/null -> [] ; other -> leave page as-is
  

  • only engaged when the op has x-fern-pagination or the body has the heuristic next-token property (is_paginated_response); a single-page non-list response under --page-all keeps its shape
  • runs on the pager/TTY path and the piped/captured path alike, so both emit the same items (the pager path streams one array per page, as it already did for csv/table/jsonl)
  • captured pages are then just concatenated (merge_captured_pages)
  • --no-extract skips all of this and returns full page envelopes
• discovery.rs: PaginationConfig::results_path() accessor.
• Changelog fix-page-all-extraction.yml; regenerated seed/cli/cli-basic-auth.

Before/after against a 2-page Benchling mock (benchling dna-sequence list --page-all --format csv):

# before
items,nextToken
"[{""id"":""seq_1""},{""id"":""seq_2""}]",p2
"[{""id"":""seq_3""}]",
# after
id,name
seq_1,a
seq_2,b
seq_3,c

Testing

• Unit tests: PageItemsPath decide/items (body-relative declared path + return-value, first-page key lock vs. later sibling arrays, null/absent pages, ambiguity), handle_json_response end-to-end for return-value+results, --no-extract, and non-paginated responses; cargo test --lib in generators/cli/sdk passes (1974)
• pnpm seed test --generator cli --fixture cli-basic-auth --skip-scripts --local 2/2
• Manual testing against the mock: piped json/csv, script-TTY --no-pager, PAGER=cat, --no-extract, and plain single-page list

Link to Devin session: https://app.devin.ai/sessions/d3e6c0a2eab24903ae7765c8a4c54032
Open in Devin Desktop: https://app.devin.ai/desktop/session/d3e6c0a2eab24903ae7765c8a4c54032?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

[nitpickybot]

AI Review Summary

Reasonable fix for --page-all output shape; merge logic is defensive and falls back to raw pages instead of dropping data. Two concerns: the new unit test appears to be placed outside mod tests, and the heuristic key selection can latch onto the wrong array key when the first page's real results are empty but another array field is populated.

• 🟡 2 warning(s)
• 🔵 1 suggestion(s)

---

To request another review, comment /ai-review on this pull request.

[nitpickybot]

🔵 suggestion

The heuristic picks the key from the first page that yields a match; if page 1 has {"items": [], "warnings": ["x"]} the heuristic could latch onto warnings and then flatten warnings across all pages, silently discarding real items. Consider preferring sole_shared_array_key (consistent across all pages) before falling back to the single-page heuristic, or at least verifying the chosen key is an array on every page — which the later loop does, but it can't detect that the wrong array was chosen.

[devin-ai-integration]

Devin Review found 1 potential issue.