Excel Native-Image Prototype Design

Goal

Determine whether an xlsx reader can run inside a GraalVM native-image binary, using FastExcel (not Apache POI) as the primary reading library.

Research findings

Apache POI: possible but painful

No native-image metadata shipped by POI or in oracle/graalvm-reachability-metadata
XMLBeans (POI's XML layer for xlsx) dynamically generates classes from XSD schemas and loads them via reflection — fundamentally hostile to native-image's closed-world assumption
quarkus-poi proves it works but requires: reflection registration for all XmlObject/StringEnumAbstractBase implementations, resource patterns for formula/schema files, font system property workaround, service loader merging for 5 providers
Multiple open GraalVM issues documenting failures (#1993, #2186, #2187, #3191)

FastExcel reader: strong fit

No POI dependency — only aalto-xml (streaming StAX) + commons-compress
No XMLBeans, no dynamic class generation
Supports: cell values, formulas, format strings, sheet enumeration
Missing: merged regions, named ranges (data exists in xlsx XML but parser skips it)
Both missing features are simple XML elements parseable with raw StAX

Architecture

┌─────────────────────────────────────────────┐
│          excel_sandbox.clj (-main)          │
├──────────────────┬──────────────────────────┤
│  FastExcel Reader │  Supplemental StAX XML  │
│  (values, formulas│  (merged regions,       │
│   formats, sheets)│   named ranges)         │
├──────────────────┴──────────────────────────┤
│  commons-compress ZipFile (shared)          │
├─────────────────────────────────────────────┤
│  aalto-xml (StAX)  │  javax.xml.stream      │
└─────────────────────────────────────────────┘

FastExcel handles cell values, formulas, format strings, sheet enumeration, and streaming row iteration.

Supplemental StAX handles merged regions (<mergeCells> in sheet XML) and named ranges (<definedNames> in workbook.xml) by parsing the same xlsx zip with javax.xml.stream.XMLInputFactory (JDK built-in, zero reflection).

DSL features covered

Feature	Source	Method
Cell values	FastExcel	Cell.getValue/getText/asNumber/asBoolean
Formulas	FastExcel	Cell.getFormula()
Format strings	FastExcel	Cell.getDataFormatString() (requires ReadingOptions)
Sheet names	FastExcel	ReadableWorkbook.getSheets()
Sheet metadata	FastExcel	Iterate rows, count, read first row
Merged regions	Supplemental StAX	Parse `<mergeCell ref="..."/>` from sheet XML
Named ranges	Supplemental StAX	Parse `<definedName>` from workbook.xml

Dependencies

Dep	Version	Why
org.clojure/clojure	1.12.0	Runtime
org.dhatim/fastexcel-reader	0.19.0	Cell values, formulas, formats

Transitive: aalto-xml 1.3.4, commons-compress 1.28.0.

Build pipeline

deps.edn :native-excel alias with replace-deps
AOT-compile entry namespace
Build uberjar with tools.build
Compile with /usr/lib/jvm/java-21-graalvm/bin/native-image
Test with dump/duo-maesn.xlsx

Native-image concerns

FastExcel: StAX streaming + ZipFile — no reflection
aalto-xml: StAX ServiceLoader for XMLInputFactory — may need service provider resource
commons-compress: JDK RandomAccessFile — native-image friendly
Clojure: needs --initialize-at-build-time and standard Clojure reflection config

Separate zip access

FastExcel's OPCPackage is package-private. The supplemental parser opens its own ZipFile on the same file. This is fine — ZipFile is read-only and cheap.