[chsword]

做了什么

新增 ExcelStreamToObject：逐行读出 .xlsx 的一张工作表，不把整个工作簿建进内存。上一版让导出不再受行数牵制，这一版补上读的一侧。

using var file = File.OpenRead("orders.xlsx");
foreach (var order in ExcelHelper.ExcelStreamToObject<Order>(file))
    Handle(order);

本机实测（五列，含文本、金额与日期，net8.0）：

行数	整份读入	流式导入
5 万	393 MB / 3.1 秒	103 MB / 1.1 秒
20 万	1265 MB / 8.4 秒	194 MB / 2.9 秒

两条路读出的数据逐字段一致（实测中以金额求和校验，两者相同）。快出一截是因为不必先为每一格建一个对象。流式导入的内存并非常数：共享字符串表会整个留在内存里，省掉的是「每一格一个对象」那一份。

怎么做的

读的是工作表那份 XML（XSSFReader 取出各部件，XmlReader 逐行解析），共享字符串表与样式表仍走 NPOI 的公开入口。一行解出来即交给调用方，随后便可回收。

导入的类型转换原本直接写在 ICell 之上，而逐行读出拿到的是原始 XML，并没有单元格对象。故把两者都归到内部的 CellData（一格读到的东西）与 IImportRow（一行）上：转换那一套因而只有一份，两条路的行为不会各自漂移。这一步是纯重构，改完既有的 392 个用例原样通过。

三处不同

• 公式格读的是文件中存着的上一次计算结果，而非当场求值。Excel 存盘时会写下这个结果，故 Excel 保存过的文件照常读出；本库导出的文件里公式没有这个结果（要等 Excel 打开时才算出），此时该格读作空白。整份读入那条路仍当场求值，行为不变。
• 只有 .xlsx 能够逐行读出，.xls 的数据并非顺序存放，传入时照旧整份读入，结果一致。
• 工作表在一开始就定位，表名写错当场抛出 Excel2ObjectException，不必等到取第一行。

行为变更

均对既有入口同样生效，属读取行为的收敛：字典形式的导入改为惰性给出；标题行中同名的标题以最左一列为准（此前 ToDictionary 会抛 ArgumentException）；非文本的标题不再中断导入；日期序列号超出 Excel 日历时上报的异常改为本库抛出并写明该序列号。

测试

新增 12 个用例（StreamingImportTest）：与整份读入逐字段比对、共享字符串与行内字符串两种写法、空白与缺列、按表名定位、表名写错当场报错、TitleSkipLine、公式有/无缓存值、读取失败的上报位置（D4）、惰性取用、.xls 入口、字典形式。

全量 404 个用例在 net8.0 与 net10.0 通过，库在 7 个目标框架上 0 警告 0 错误。

版本 2.11.0，文档见 docs/versions/v2.11.0.md。

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

Unresolved critical XML parsing and multiple moderate correctness and resource-cleanup issues remain.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Adds lazy, row-by-row .xlsx importing through ExcelStreamToObject, while unifying existing conversion logic.

Changes:

• Added XML-based streaming worksheet parsing.
• Refactored shared cell, row, and error handling.
• Added tests, documentation, and version 2.11.0 updates.

File summaries

File	Summary	Findings
README.md	Chinese feature and release documentation	Nit: incorrect additional v2.10.0 link.
README_EN.md	English feature and release documentation	Nit: incorrect additional v2.10.0 link.
docs/versions/v2.11.0.md	Version documentation and API examples	No findings.
docs/README.md	Documentation index update	No findings.
Chsword.Excel2Object/Internal/XlsxRowReader.cs	Streaming XLSX XML parser	Critical namespace parsing issue; moderate empty-row, row-number, and uncached-formula handling issues.
Chsword.Excel2Object/Internal/ImportRow.cs	Shared import-row abstraction	No findings.
Chsword.Excel2Object/Internal/ImportContext.cs	Import context and error positioning	No findings.
Chsword.Excel2Object/Internal/CellData.cs	Shared cell representation	No findings.
Chsword.Excel2Object/ExcelImporter.cs	Streaming API and conversion refactor	Moderate resource cleanup, option callback, duplicate-header, and non-seekable .xls issues; nit on nullable cast.
Chsword.Excel2Object/ExcelHelper.cs	Static streaming API facade	No findings.
Chsword.Excel2Object/Chsword.Excel2Object.csproj	Version update to 2.11.0	No findings.
Chsword.Excel2Object.Tests/StreamingImportTest.cs	Streaming import test coverage	No findings.

Review details

Suppressed comments (3)

Chsword.Excel2Object/ExcelImporter.cs:204

• Duplicate header titles are still not resolved to the leftmost column for model imports: ContainsKey(cell.Key) only checks the current column, so both occurrences are added and PopulateModelFromRow assigns the property twice, leaving the rightmost value. De-duplicate by the matched property/title, as the dictionary path does, so the first matching column is retained.

                if (prop.Key != null && !dictColumns.ContainsKey(cell.Key))

Chsword.Excel2Object/ExcelImporter.cs:118

• For a seekable .xls stream this branch falls back to the existing importer, but for a non-seekable .xls stream LooksLikeXlsx returns true and OPCPackage.Open rejects it as an .xlsx. Thus the documented .xls fallback only works for seekable streams even though this public API accepts any Stream; either preserve the prefix while detecting the format or document and validate this restriction.

        if (!input.CanSeek) return true;

        var position = input.Position;
        var head = new byte[2];
        var read = input.Read(head, 0, 2);
        input.Position = position;
        return read == 2 && head[0] == (byte) 'P' && head[1] == (byte) 'K';

Chsword.Excel2Object/ExcelImporter.cs:133

• This null-forgiving cast suppresses nullable analysis in newly added code. The type test makes the runtime cast safe, but the repository's nullable guidance requires avoiding ! for this pattern; return a typed lazy projection (for example, Cast<TModel>()) instead of hiding the warning.

        if (typeof(TModel) == typeof(Dictionary<string, object>))
            return (InternalExcelToDictionary(rows, context) as IEnumerable<TModel>)!;

• Files reviewed: 12/12 changed files
• Comments generated: 8
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.