Fix KeyError in insert_all when pk column is missing from a single record

[ATOM00blue]

Fixes #732

The bug

insert_all() with a pk= column that isn't present in the records raised a KeyError — but only when inserting exactly one row. Every other row count succeeded:

import sqlite_utils

for n in [0, 1, 2, 3, 10]:
    db = sqlite_utils.Database(memory=True)
    db.conn.execute("CREATE TABLE t (a TEXT, b INT, PRIMARY KEY (a, b))")
    rows = [{"a": f"x{i}", "b": i} for i in range(n)]
    try:
        db.table("t").insert_all(rows, pk="not_a_column", alter=True)
        print(f"n={n}: ok")
    except KeyError as e:
        print(f"n={n}: KeyError {e}")

n=0: ok
n=1: KeyError 'not_a_column'
n=2: ok
n=3: ok
n=10: ok

Root cause

The single-record branch that populates last_pk reads the pk value straight out of the just-inserted row (row[pk]). When the named pk column doesn't actually exist on the table, that lookup throws. The multi-record path doesn't go through this branch, hence the inconsistency.

Fix

Check that the named pk column(s) are present on the row before reading them, and fall back to the rowid otherwise — the same fallback already used when no pk is given. Single-row inserts now behave like every other row count.

Testing

Added test_insert_all_pk_not_in_records, parametrized over 0/1/2/3/10 rows, which fails on main (KeyError for the single-row case) and passes with this change. Full test suite, black, flake8 and mypy all pass locally.

---

📚 Documentation preview 📚: https://sqlite-utils--745.org.readthedocs.build/en/745/

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR fixes inconsistent insert_all() behavior when a pk= column is specified but does not exist in the table/records, aligning single-row inserts with multi-row behavior (no KeyError; fall back to rowid).

Changes:

• Add regression test covering 0..N inserts when pk column is missing.
• Update insert_all() logic for computing last_pk to tolerate missing PK columns and fall back to last_rowid.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
tests/test_create.py	Adds parametrized regression coverage for missing pk column behavior across row counts
sqlite_utils/db.py	Adjusts last_pk derivation to handle missing PK columns and fall back to rowid

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.