spike: 005 export all tables of a SQLite file to CSV

[iheitlager]

Goal

First taste of V1's value ("open any .sqlite file, extract the data") as a working end-to-end tool, ahead of the real steps landing. Extends spike 002's reader (#4/#6) from one hardcoded table to: enumerate all tables via sqlite_master, walk each, and export every table as CSV.

Branch: spike/005_csv_export

Behavior

sqlite-rs-export path/to/mydata.db
# writes, next to the input file:
#   path/to/<table>_mydata.csv        (one per user table)

• Output naming: <table>_<file>.csv (file = input basename without extension), written in the same folder as the input
• All user tables from sqlite_master (skip sqlite_* internal tables; skip virtual tables with a warning — graceful unknowns per Tier 0)
• Header row = column names from the minimal DDL read; rows in rowid order

What this spike de-risks (beyond spike 002)

Area	Step it informs
Multi-table enumeration + root-page resolution from sqlite_master	step 7 (minimal DDL reader)
Multi-page b-trees (interior nodes) on real data	step 4
Overflow chains, if the chosen fixture has large values	step 4
CSV value rendering: NULL representation, quoting/escaping, blob encoding (hex?), float formatting	step 9 (output contract) — spike 002 finding 3 gets a concrete decision

Experiment (Atomic)

Throwaway crate under tests/spike/005_csv_export/, reusing spike 002's decoding sketch:

1. Fixture: a db with 3+ tables, one multi-page (1000+ rows), one with all value types, one with a large TEXT/BLOB (overflow)
2. Enumerate sqlite_master, resolve root pages + column names
3. Walk each table b-tree recursively (interior → leaf), decode records
4. Write CSVs; oracle-diff each against sqlite3 -csv "SELECT * FROM <t>"

Falsification criteria

Spike FAILS (valuably) if: interior-node traversal or overflow reassembly has undocumented surprises; or CSV parity with sqlite3 -csv is unreachable without semantic knowledge we don't have at Tier 0 (e.g. affinity-dependent rendering).

Out of scope

WAL (#7), locking (#8), UTF-16, WITHOUT ROWID, index b-trees. Loud asserts on all of these.

Timebox

1-2 days. Findings feed steps 4/7/9 ticket specs; the CSV rendering decisions go into the step 9 output contract.

Exit criteria

• All tables of the fixture exported, sqlite3 -csv parity per table
• Multi-page traversal + overflow exercised
• CSV rendering decisions documented (NULL, quoting, blobs, floats)
• findings.md + make spike-005

---

🤖 Generated with Claude Code