wolfxl 2.0.0rc2

Fast, openpyxl-compatible Excel I/O with Rust backend and built-in formula engine (67 functions, financial, date, time, text, lookup, conditional stats)

WolfXL

Openpyxl-compatible Excel automation with a Rust backend.
Read, write, and surgically modify workbooks, including charts, images, encryption, structural ops, and pivot-table construction.

---

Openpyxl-style imports. One import change.

- from openpyxl import load_workbook, Workbook
- from openpyxl.styles import Font, PatternFill, Alignment, Border
+ from wolfxl import load_workbook, Workbook, Font, PatternFill, Alignment, Border

Most workbook automation keeps the same shape: ws["A1"].value, Font(bold=True), and wb.save() all work the way openpyxl users expect.

For codebases where changing every openpyxl import is impractical, install the runtime alias once at process startup:

import wolfxl; wolfxl.install_as_openpyxl()

import openpyxl
from openpyxl.styles import Font

---

_{Dated WolfXL 2.0 release-artifact evidence is available in ExcelBench: 2026-04-28 wheel-backed rerun, 18/18 green features, and performance snapshots.}

On the dated 2026-06-04 supported-scope benchmark (openpyxl 3.1.5 vs WolfXL 2.0): ~8x faster on large writes and ~14-16x faster on large reads at 200,000-row scale, with the same openpyxl API. Per-workload measured numbers are in Batch APIs below; smaller grids show smaller multipliers.

Current Evidence

• WolfXL package surface is at 2.0.0.
• The current strict SOTA audit reports the supported-scope gate is currently green: current fidelity evidence passes, and fresh clean-source OpenPyXL plus Rust/Rust-backed benchmark reruns are source-relevant.
• WolfXL runs OpenPyXL's own published test suite unchanged: the upstream tests import openpyxl, which install_as_openpyxl aliases to wolfxl, so they exercise WolfXL's real code paths. It passes across the supported scope with zero triaged behavior gaps, and this openpyxl-compatibility check runs in CI on every change (the parity job in .github/workflows/ci.yml, driven by scripts/run_openpyxl_corpus.py against the vendored upstream corpus).
• The broader all-future-surface SOTA claim is still not ready; the audit keeps that gated until future package/install-route, open-ended render, click-level interaction, and future-surface proof boundaries are resolved.
• The latest local evidence snapshot is docs/performance/baselines/2026-06-05-current-sota-claim-audit.md.
• For launch copy, use the Launch Claim Brief: it turns the proof files into safe public wording and keeps the broader all-future-surface claim gated.
• Historical wheel-backed ExcelBench evidence is available in the 2026-04-28 release snapshot fidelity report.
• The paired dated evidence includes the release snapshot dashboard and the matching perf snapshot; refresh these before using them as next-release proof.
• Use the Public Evidence Status page to see which claims are current, historical, or still gated.

Install

pip install wolfxl

Quick Start

from wolfxl import load_workbook, Workbook, Font, PatternFill

# Write a styled spreadsheet
wb = Workbook()
ws = wb.active
ws["A1"].value = "Product"
ws["A1"].font = Font(bold=True, color="FFFFFF")
ws["A1"].fill = PatternFill(fill_type="solid", fgColor="336699")
ws["A2"].value = "Widget"
ws["B2"].value = 9.99
wb.save("report.xlsx")

# Read it back — styles included
wb = load_workbook("report.xlsx")
ws = wb[wb.sheetnames[0]]
for row in ws.iter_rows(values_only=False):
    for cell in row:
        print(cell.coordinate, cell.value, cell.font.bold)
wb.close()

Pivot tables through modify mode

import wolfxl
from wolfxl.chart import Reference
from wolfxl.pivot import PivotCache, PivotTable

wb = wolfxl.load_workbook("source-data.xlsx", modify=True)
ws = wb.active
src = Reference(ws, min_col=1, min_row=1, max_col=4, max_row=100)
cache = wb.add_pivot_cache(PivotCache(source=src))
pt = PivotTable(
    cache=cache, location="F2",
    rows=["region"], cols=["quarter"], data=[("revenue", "sum")],
)
ws.add_pivot_table(pt)
wb.save("pivot.xlsx")

WolfXL constructs pivot tables with pre-aggregated records — pivots open in Excel, LibreOffice, and openpyxl with data populated, through the modify-mode patcher (load_workbook(..., modify=True)), with no refresh-on-open required. In the project comparison below, openpyxl preserves pivots on round-trip but does not provide this constructor, and XlsxWriter does not support pivots.

Link a chart to the pivot:

from wolfxl.chart import BarChart

chart = BarChart()
chart.pivot_source = pt          # emits <c:pivotSource> + per-series <c:fmtId>
ws.add_chart(chart, "F18")

Existing pivots can also be edited in modify mode: source ranges, row/column/page field placement, page-field selection, and data-field aggregation all route through PivotTableHandle. Layout edits stamp refreshOnLoad="1" so Excel refreshes derived pivot cache records on open; source-range edits with the same shape stay byte-stable.

Three Modes

Mode	Usage	Engine	What it does
Read	load_workbook(path)	NativeXlsxBook	Parse XLSX with values, formulas, styles, metadata, drawings, and tables
Write	Workbook()	rust_xlsxwriter	Create new XLSX files from scratch
Streaming write	Workbook(write_only=True)	Per-sheet temp file + <sheetData> splice	Append-only, bounded-memory ETL exports — peak RSS dominated by SST + styles, not by row data
Modify	load_workbook(path, modify=True)	XlsxPatcher	Surgical ZIP patch — only changed cells are rewritten

Modify mode preserves everything it doesn't touch: charts, macros, images, pivot tables, VBA.

Supported Features

Features marked Preserved are kept verbatim on modify-mode round-trip (open, edit other cells, save). The table below lists the currently documented support surface; caveats for pivots, external links, and adjacent non-openpyxl file formats live in Limitations.

Category	Features
Data	Cell values (string, number, date, bool), formulas, comments, hyperlinks
Styling	Font (bold, italic, underline, color, size), fills, borders, number formats, alignment; Color(theme=...) and Color(indexed=...) accepted
Structure	Multiple sheets, merged cells, defined names (read + write), freeze panes, row heights, column widths, document properties
Tables / Validation / CF	ws.tables, ws.add_table, ws.data_validations, ws.conditional_formatting (read + write in Workbook() mode)
Charts	16 chart families — BarChart, LineChart, PieChart, DoughnutChart, AreaChart, ScatterChart, BubbleChart, RadarChart, BarChart3D, LineChart3D, PieChart3D, AreaChart3D, SurfaceChart, SurfaceChart3D, StockChart, ProjectedPieChart
Pivots	PivotCache, PivotTable, RowField / ColumnField / DataField / PageField; slicers, calculated fields/items, GroupItems; pivot-chart linkage via chart.pivot_source = pt; deep-clone of pivot-bearing sheets
Images	Image (PNG / JPEG / GIF / BMP); one-cell, two-cell, absolute anchors; modify-mode add_image
Encryption	Read + write Agile (AES-256 / SHA-512) via wolfxl[encrypted]
Iteration	iter_rows, iter_cols, rows, columns, values, range slicing (ws["A1:B2"], ws["A:B"], ws[1:3])
Utils	get_column_letter, column_index_from_string, coordinate_to_tuple, range_boundaries, absolute_coordinate, quote_sheetname, range_to_tuple, rows_from_range, cols_from_range, get_column_interval, dataframe_to_rows, is_date_format
Preserved / linked content	Macros (VBA) — round-trip cleanly through modify mode with raw xl/vbaProject.bin bytes available via Workbook.vba_archive for inspection (no authoring API); external workbook links (wb._external_links exposes target / sheet_names / cached values and supports append/remove/update-target authoring); embedded objects also round-trip

openpyxl compatibility status

Modules that import from openpyxl generally work against wolfxl. Unsupported classes raise NotImplementedError with a clear hint at the construction site - no silent no-ops.

Class / API	Status
Font, PatternFill, Border, Side, Alignment, Color	Full support
Comment, Hyperlink	Read + write (write mode); modify-mode setters T1.5
DataValidation, Table, TableStyleInfo, TableColumn	Read + write (write mode); modify-mode setters T1.5
CellIsRule, FormulaRule, ColorScaleRule, DataBarRule, IconSetRule	Read + write (write mode); modify-mode setters T1.5
DefinedName, DocumentProperties	Read + write (write mode); modify-mode setters T1.5
NamedStyle, Protection, GradientFill, DifferentialStyle	Constructor / dataclass support
BarChart, LineChart, PieChart, Reference, Series (from wolfxl.chart)	Full support (1.6+) — 16 chart families incl. 3D / Stock / Surface / ProjectedPie
Image (from wolfxl.drawing.image)	Full support (1.5+) — PNG / JPEG / GIF / BMP, all anchor types
PivotTable, PivotCache (from wolfxl.pivot)	Supported (2.0+) — modify-mode construction, chart linkage, and deep-clone within the documented pivot caveats
AutoFilter	Read + write support
ws.insert_rows, ws.delete_rows	Full support (modify mode, 1.1+) — RFC-030
ws.insert_cols, ws.delete_cols	Full support (modify mode, 1.1+) — RFC-031
ws.move_range	Full support (modify mode, 1.1+) — RFC-034
wb.move_sheet	Full support (modify mode, 1.1+) — RFC-036
wb.copy_worksheet	Full support (write + modify mode, 1.1+) — RFC-035

Performance at Scale

Dated WolfXL 2.0 release-artifact evidence is now available for the OpenPyXL and required Rust/Rust-backed benchmark grids. Use those reports for claim-grade numbers, not older draft 10×-100× marketing copy or placeholder benchmark rows. On the dated 2026-06-04 supported-scope benchmark, the measured large-workbook results are roughly 8x faster writes and 14-16x faster reads versus openpyxl at 200,000-row scale; smaller grids show smaller multipliers. The broader all-future-surface SOTA claim remains gated.

For historical context and reproducible benchmark commands, see Benchmark Results. For the current claim status, see Public Evidence Status.

The implementation goal remains stable throughput as files grow, with linear pivot construction over source-row count.

How WolfXL Compares

Every Rust-backed Python Excel project picks a different slice of the problem. In the supported-scope comparison tracked by the current audit, WolfXL covers all four practical workflow areas shown below: formatting, modify mode, openpyxl API compatibility, and pivot-table construction. This is a capability map, not a claim that every future Excel surface is already exhausted.

The measured Rust/Rust-backed benchmark gate is separate from this table. It currently covers rust_xlsxwriter 0.95.0, crates.io package xlsxwriter 0.6.1 with repo alias xlsxwriter-rs, calamine 0.35.0, umya-spreadsheet 3.0.0, fastexcel 0.20.2, and python-calamine 0.6.2; see Public Evidence Status for the dated, supported-scope result and caveats.

Library	Read	Write	Modify	Styling	openpyxl API	Pivots
fastexcel	Yes	—	—	—	—	—
python-calamine	Yes	—	—	—	—	—
FastXLSX	Yes	Yes	—	—	—	—
rustpy-xlsxwriter	—	Yes	—	Partial	—	—
openpyxl	Yes	Yes	Yes (full DOM)	Yes	Native	Round-trip only*
XlsxWriter	—	Yes	—	Yes	—	—
WolfXL	Yes	Yes	Yes (surgical)	Yes	Yes	Yes (construction + linkage)

• Styling = reads and writes fonts, fills, borders, alignment, number formats
• Modify = open an existing file, change cells, save back — without rebuilding from scratch
• openpyxl API = same load_workbook, Workbook, Cell, Font, PatternFill objects
• Pivots = construct a pivot table from Python with pre-aggregated records (the file opens in Excel / LibreOffice / openpyxl with data populated, no refresh-on-open)

*openpyxl preserves pivot tables on round-trip but does not provide a Python-side constructor that emits the pivotCacheRecords snapshot. WolfXL's public ecosystem claim here should stay tied to the current Public Evidence Status boundary.

See Limitations before relying on OLAP or external pivot caches, broad pivot visual styling, or linked-workbook external-link refresh behavior.

WolfXL's public .xlsx and .xlsb readers are native. .xlsb is read-only but exposes values, cached formula results, and cell styles; legacy .xls remains on the Calamine-backed value path.

Batch APIs for Maximum Speed

For write-heavy workloads, use append() or write_rows() instead of cell-by-cell access. These APIs buffer rows as raw Python lists and flush them to Rust in a single call at save time, bypassing per-cell FFI overhead entirely.

from wolfxl import Workbook

wb = Workbook()
ws = wb.active

# append() — fast sequential writes (3.7x faster than cell-by-cell)
ws.append(["Name", "Amount", "Date"])
for row in data:
    ws.append(row)

# write_rows() — fast writes at arbitrary positions
ws.write_rows(header_grid, start_row=1, start_col=1)
ws.write_rows(data_grid, start_row=5, start_col=1)

wb.save("output.xlsx")

For reads, iter_rows(values_only=True) uses a fast bulk path that reads all values in a single Rust call. The per-workload numbers below come from the dated 2026-06-04 supported-scope benchmark (openpyxl 3.1.5 vs WolfXL 2.0); smaller grids show smaller multipliers.

wb = load_workbook("data.xlsx")
ws = wb[wb.sheetnames[0]]
for row in ws.iter_rows(values_only=True):
    process(row)  # row is a tuple of plain Python values

For ingestion, dataframe, or review workflows that need values plus formatting signals, use cell_records(). It returns compact dictionaries without creating one Python Cell object per coordinate:

records = ws.cell_records(
    include_format=True,
    include_formula_blanks=False,
    include_coordinate=False,
)

for record in records:
    print(record["row"], record["column"], record["value"], record.get("number_format"))

API	vs openpyxl (2026-06-04 dated rerun)	How
ws.append(row)	7.2x faster write (200k rows)	Buffers rows, single Rust call at save
ws.write_rows(grid)	7.9x faster write (200k rows)	Same mechanism, arbitrary start position
ws.iter_rows(values_only=True)	16.1x faster read (200k rows)	Single Rust call, no Cell objects
ws.cell_records()	Fast styled sparse read	Single Rust call, values plus compact format metadata
ws.cell(r, c, value=v)	3.4x faster write	Per-cell FFI (compatible but slower)

Formula Engine

WolfXL includes a built-in formula evaluator with 67 functions across 7 categories. Calculate formulas without external dependencies - no need for formulas or xlcalc.

from wolfxl import Workbook
from wolfxl.calc import calculate

wb = Workbook()
ws = wb.active
ws["A1"] = 100
ws["A2"] = 200
ws["A3"] = "=SUM(A1:A2)"
ws["B1"] = "=PMT(0.05/12, 360, -300000)"  # monthly mortgage payment

results = calculate(wb)
print(results["Sheet!A3"])  # 300
print(results["Sheet!B1"])  # 1610.46...

# Recalculate after changes
ws["A1"] = 500
results = calculate(wb)
print(results["Sheet!A3"])  # 700

Category	Functions
Math (10)	SUM, ABS, ROUND, ROUNDUP, ROUNDDOWN, INT, MOD, POWER, SQRT, SIGN
Logic (5)	IF, AND, OR, NOT, IFERROR
Lookup (7)	VLOOKUP, HLOOKUP, INDEX, MATCH, OFFSET, CHOOSE, XLOOKUP
Statistical (13)	AVERAGE, AVERAGEIF, AVERAGEIFS, COUNT, COUNTA, COUNTIF, COUNTIFS, MIN, MINIFS, MAX, MAXIFS, SUMIF, SUMIFS
Financial (7)	PV, FV, PMT, NPV, IRR, SLN, DB
Text (13)	LEFT, RIGHT, MID, LEN, CONCATENATE, UPPER, LOWER, TRIM, SUBSTITUTE, TEXT, REPT, EXACT, FIND
Date / Time (12)	TODAY, DATE, YEAR, MONTH, DAY, EDATE, EOMONTH, DAYS, NOW, HOUR, MINUTE, SECOND

Named ranges are resolved automatically. Error values (#N/A, #VALUE!, #DIV/0!, #REF!, #NUM!, #NAME?) propagate through formula chains like real Excel. Install pip install wolfxl[calc] for extended formula coverage via the formulas library fallback.

Case Study: SynthGL

SynthGL switched from openpyxl to WolfXL for their GL journal exports (14-column financial data, 1K-50K rows). Results: 4x faster writes, 9x faster reads at scale. 50K-row exports dropped from 7.6s to 1.3s. Read the full case study.

Case Study: Finance Template Modify Mode

For a representative finance workflow, see Template-Driven Finance Workbook Updates. It benchmarks the exact workload where WolfXL is most differentiated: open an existing workbook with formulas, charts, comments, validations, and large detail sheets, touch three assumption cells, and save.

Case Study: Styled Report Generation

For a fresh-workbook export path, see Styled Report Generation. The current local snapshot on a 10k-row reporting workbook came in at 0.553s for openpyxl vs 0.380s for WolfXL, about 1.46x faster.

Case Study: Workbook-Preserving ETL

For the "update one data block but keep the workbook" workflow, see Workbook-Preserving ETL Update. The current local snapshot for replacing a 2,000-row block inside a 20,000-row reporting workbook came in at 1.064s for openpyxl vs 0.496s for WolfXL, about 2.15x faster.

Case Study: Pivot Construction

For a capability-first example, see Pivot Construction From Python. It demonstrates workbook-scoped pivot cache creation, pivot-table emit, and pivot-linked chart output through WolfXL modify mode. The current local snapshot constructed the full artifact on a 10,000-row source sheet in 0.229s median with emitted pivot cache, pivot records, pivot table, and chart <c:pivotSource> parts all validated.

How It Works

WolfXL is a thin Python layer over compiled Rust engines, connected via PyO3. The Python side uses lazy cell proxies: large workbooks do not materialize every cell up front, so values and styles are fetched from Rust only when you access them. Runtime still depends on the workbook and access pattern; validate speed claims against a measured workload. On save, dirty cells are flushed in one batch, avoiding per-cell FFI overhead.

License

MIT