auto-delimiter detection: fix IndexError on short files, raise a typed error

[jepegit]

What

find_delimiter_and_start (cellpy/readers/instruments/base.py) failed to load short files with a bare IndexError: list index out of range — an error naming neither the file nor the cause. A short vendor file (a truncated run, a quick sanity export, a small fixture) is perfectly normal.

Root cause

The function split its sample window into a header part and a data part, but computed the data-part start from the nominal window size rather than the lines actually read:

_find_separator(checking_length_whole - checking_length_header, lines, separators)
# → last_part = lines[checking_length_whole - checking_length_header : -1]

When the file was shorter than that offset, last_part was empty and _find_separator ran list({})[0] → IndexError.

Threshold (bisected, not assumed)

The real callers (AutoLoader._auto_formatter, and the same in custom.py) use checking_length_header=100, checking_length_whole=200, putting the split at line 100 — so any file under 102 lines failed. For neware_txt_zero (1 header + D data rows): D=100 → 101 lines crashes, D=101 → 102 lines works, which is exactly the reported "100 fails / 150 works" bracket. (The module default header=30 puts the boundary at 171 rows; both confirmed by bisection.)

Fix

• _find_separator clamps the header window to the actual line count — min(checking_length_header, n-1) — always leaving ≥1 data row, and drops the possibly-truncated last line only when a data row can still be spared. For any file that already worked (n ≥ 102 at header=100), the inspected window lines[100:-1] is byte-identical → zero behavior change on existing files.
• A header line + one data line is now enough (verified to the 2-line floor).
• Genuine failures raise a typed LoaderError naming the file — empty/blank-only files, no consistent delimiter, or an unlocatable header — replacing both the IndexError and the old untyped IOError.

Verification

• End-to-end through the real neware_txt loader: D = 1, 2, 5, 20, 100, 300 all load (previously 1–100 crashed).
• New tests/test_find_delimiter_and_start.py — 35 tests: sizes 1/2/5/20/100/101/150 across four delimiters, preamble handling, the minimal 2-line file, and both typed-error paths.
• No regressions: real neware_uio.csv parity tests and the loader_custom golden (both exercise auto-detection) stay green.

Context

Found while extending the #560 loader-port parity coverage (PR #602); split out to keep that branch scoped. This also removes the reason the synthetic-parity fixtures had to be 300 rows.

🤖 Generated with Claude Code

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.