[Bug] LongestMatchColumnWidthStyleStrategy leaks widths across workbooks #1305

Search before asking

• I searched the issues and found nothing similar.

Fesod version

main@126dde9 (dev-SNAPSHOT)

JDK version

21.0.10 (Microsoft) — project targets Java 8 (CI matrix: 8/11/17/21/25)

Operating system

Windows 10

Steps To Reproduce

Reuse a single LongestMatchColumnWidthStyleStrategy instance for two writers:

import java.io.File;
import java.util.Collections;
import org.apache.fesod.sheet.FesodSheet;
import org.apache.fesod.sheet.write.style.column.LongestMatchColumnWidthStyleStrategy;
import org.apache.poi.ss.usermodel.Sheet;
import org.apache.poi.ss.usermodel.Workbook;
import org.apache.poi.ss.usermodel.WorkbookFactory;

public class ColumnWidthReuseRepro {
    public static void main(String[] args) throws Exception {
        LongestMatchColumnWidthStyleStrategy strategy = new LongestMatchColumnWidthStyleStrategy();

        File first = new File("first.xlsx");
        File second = new File("second.xlsx");

        FesodSheet.write(first).registerWriteHandler(strategy).sheet()
            .doWrite(Collections.singletonList(Collections.singletonList("AAAAAAAAAAAAAAAAAAAAAAAAAAAAAA"))); // 30 chars
        FesodSheet.write(second).registerWriteHandler(strategy).sheet()
            .doWrite(Collections.singletonList(Collections.singletonList("A"))); // 1 char

        System.out.println("first  col0 width = " + columnWidth(first));
        System.out.println("second col0 width = " + columnWidth(second));
    }

    private static int columnWidth(File file) throws Exception {
        try (Workbook wb = WorkbookFactory.create(file)) {
            Sheet sheet = wb.getSheetAt(0);
            return sheet.getColumnWidth(0);
        }
    }
}

Current Behavior

first  col0 width = 7680
second col0 width = 2048

The second file keeps POI's default column width (2048 ≈ 8 characters) instead of the expected 256 (1 character). When each writer gets its own strategy instance, the second file correctly comes out as 256 — so the discrepancy is entirely caused by reusing the instance.

Root cause in the code:

• The cache is keyed only by writeSheetHolder.getSheetNo().
• A sheet number is unique within one workbook, but two separate writers both use sheet number 0.
• A column width is only applied when it is larger than the cached maximum, so the second file's 1 never beats the first file's cached 30 and setColumnWidth is never called.
• Nothing clears the cache when a workbook finishes.

Expected Behavior

Reusing one strategy instance across writers should not carry column widths from a previous workbook into the next. The second file above should end up with a width of 256.

If instance reuse is intentionally unsupported, that limitation should be documented explicitly — silently producing wrong widths is not acceptable.

Proposed Fix

Move the per-sheet maximum column width map onto the WriteSheetHolder instead of keeping it on the strategy instance:

• Add a columnWidthCache field (Map<Integer, Integer>, keyed by column index) to WriteSheetHolder. The strategy reads/writes it via the holder in setColumnWidth and lazily initializes it.
• The strategy becomes stateless: it holds no cache, implements no WorkbookWriteHandler/SheetWriteHandler, and needs no dispose hooks.

Why this fixes the bug: each sheet (and therefore each workbook) gets its own holder with its own columnWidthCache. Reusing a single strategy instance across writers can no longer carry widths from one workbook/sheet into another, because there is no shared state left on the strategy. The cache also dies with the holder, so there is no leak even for an abandoned writer.

This is the column-width counterpart of the converter-isolation work in #740, which solved the same "state shared across reused writer instances" class of problem for converters by giving each holder its own copy.

Regression test

A test comparing separate instances (control) with a shared instance reproduces the problem deterministically; I will include it in the PR.

Related

• fix: ensure converter isolation across ExcelWriter and ExcelReader #740 — converter isolation between ExcelWriter instances (same class of fix, different component). The proposed column-width fix follows the same "state owned by the holder" pattern.
• [Bug] Converter isolation broken - converters leak between ExcelWriter instances #729 — the original converter state-isolation issue (closed by fix: ensure converter isolation across ExcelWriter and ExcelReader #740).

Are you willing to submit a PR?

• I'm willing to submit a PR!