Use Chatbook's reported line number for multi-input evaluations

[rhennigan]

Summary

WolframResearch/Chatbook#1677 (Chatbook 2.7.28) changed WolframLanguageToolEvaluate to evaluate each top-level input of the code separately, the way an interactive kernel session does. Each input gets its own line number:

1 + 1
2 + 2

Out[1]= 2

Out[2]= 4

One tool call can now use up several line numbers, so incrementing the session's line counter once per call puts the next call's Out[n] on the wrong line. Chatbook now reports the line number for the next input as a new "Line" property, and this PR uses it.

Changes (Kernel/Tools/WolframLanguageEvaluator.wl)

• Line counter: chatbookToolEvaluate also requests the "Line" property, continues the session's $line from it, and drops it from the result before returning. Without a valid reported line (older Chatbook, a failed evaluation, or the eval kernel quit), the counter advances by one as before.
• "Local" bookkeeping: session start/save/resume and the paclet load in the eval subkernel now go through a new localKernelEvaluate. On 2.7.28+ it evaluates with "Line" -> None instead of rolling back $Line. This matters because Chatbook now records In/Out history for held input, so the old rollback would overwrite history entries. As a side effect, %, Out[n] and InString[n] work in Local-method sessions now; they used to return internal ByteArrays.
• syncEvalKernelLine: this is now a no-op on 2.7.28+, because Chatbook applies an explicit "Line" option in the Local and Cloud evaluator kernels itself.
• UI notebook (MCP Apps): the input cell holds all of the code, so it's now labeled with the first input's line number, like a notebook input cell. Before, it took the label from the last Out[n]=, so a three-input call starting at line 1 showed In[3]:=. makeEvaluatorUIResult takes that line number as a new third argument.

All of this is gated on the installed Chatbook version, using the same pattern as the existing $CloudSessionMX gate ($linePropertyChatbookVersion = "2.7.28", chatbookLinePropertyQ). With older Chatbook versions (the minimum is still 2.7.0) the behavior is unchanged. I also updated docs/tools.md and the comments that described the old line handling.

Related: WolframResearch/Chatbook#1677, #249. #249 is fixed on the Chatbook side by the same PR, which parses each input in the evaluator kernel.

Test plan

• New mocked unit tests in Tests/EvaluatorSessions.wlt, independent of the installed Chatbook: version gate, continuing at the reported line (string and property-list requests), fallback without a reported line, old-Chatbook path, localKernelEvaluate call shape for both versions, and the syncEvalKernelLine no-op.
• New integration tests, run only with Chatbook 2.7.28+: multi-input line numbers across calls, a session switch, and a resume from disk, under the "Session" method and (when a sandbox kernel can start) the "Local" method, including %/Out/InString history.
• New UI label tests in Tests/WolframLanguageEvaluator-UI.wlt, with cloud deployment mocked.
• Installed Chatbook 2.7.0 (TestReport): EvaluatorSessions.wlt + WolframLanguageEvaluator-UI.wlt, 88/88 pass; the 2.7.28-only tests skip.
• Dev Chatbook 2.7.28 (wolframscript with PacletDirectoryLoad): EvaluatorSessions.wlt 71/71 (including the existing cloud-session integration test), WolframLanguageEvaluator-UI.wlt 20/20, Tools.wlt 110/110.
• End-to-end through the real tool function under "Session", "Local" and "Cloud" with Chatbook 2.7.28: numbering is correct across multi-input calls, session switches, and a resume from disk.
• The same end-to-end run with Chatbook 2.7.0: output is identical to the previous behavior.
• UI path against the real cloud: a two-input call deployed a notebook labeled In[1]:= / Out[2]=, and the next call continued at In[3]:= / Out[3]=.
• CodeInspector: only the two FIXME comments that were already there.

Note: the newest Chatbook on the public paclet server is still 2.7.0, and CI doesn't install a newer one, so the 2.7.28-only integration tests will skip in CI for now.

🤖 Generated with Claude Code

https://claude.ai/code/session_015FJqQoZs86ntnnPqsvkCf6

[Copilot]

🔵 Needs a closer look

The core behavior depends on Chatbook 2.7.28 paths that currently skip in CI.

0 open findings

What changed in this PR

Updates evaluator sessions for Chatbook 2.7.28’s per-input line numbering and history behavior.

Changes:

• Continues session counters from Chatbook’s "Line" property.
• Preserves Local evaluator history during bookkeeping.
• Adds compatibility, integration, and UI-label tests.

File	Description
Kernel/​Tools/​WolframLanguageEvaluator.wl	Implements version-gated line and Local-session handling.
Tests/​EvaluatorSessions.wlt	Tests line counters, history, sessions, and compatibility.
Tests/​WolframLanguageEvaluator-UI.wlt	Tests multi-input notebook labels.
docs/​tools.md	Documents version-dependent line behavior.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.