Skip to content

Refactor reader packages to XUdt-style EncodeBody/DecodeBody. #39

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
fanfeilong merged 1 commit into from
Sep 24, 2026
Merged

Refactor reader packages to XUdt-style EncodeBody/DecodeBody. #39

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
113 changes: 76 additions & 37 deletions docs/streaming-packages.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,58 +1,97 @@
# Streaming package / progressive render
# Package / PackageParser (XUdt-style)

ExcelTk's binary and OpenXML readers are built around **packages**: complete local
format units that a parser can emit as soon as enough input has arrived.
ExcelTk frames Office-defined protocol units — not UI progressive-render DTOs.

## Mental model
## Roles

| Role | Responsibility |
| --- | --- |
| **Package** | One Office unit (BIFF record type / SpreadsheetML element). Owns `EncodeBody` / `DecodeBody` of Office fields. |
| **PackageParser** | `PushData` / `Pump` / `TryFrameOne` state machine: recognize complete unit → empty package by type → `DecodeBody` → queue; `HavePackage` / `PopPackage` / `HaveUnParsedData` / `Reset`. Does not invent field layout. |
| **Reader** | Workbook semantics (SST, styles, DataSet) consuming popped packages / cursor records. |

Wire layouts are dictated by Office Excel — do not change them for streaming convenience.

## Encode / Decode model

Same shape as XUdt:

| Layer | Binary (`BinaryPackage`) | XML (`XmlPackage`) |
| --- | --- | --- |
| Header framing | Base writes/reads 2-byte type + 2-byte body size | Element start tag / framer owns fragment bytes |
| Body | Virtual `EncodeBody` / `DecodeBody` on each concrete package | `EncodeBody(XmlWriter)` / `Decode(XmlReader)` (bytes route via `IPackage.DecodeBody`) |
| Required size | `GetRequiredEncodeBodyBufferLength()` | Encode-to-`MemoryStream` via `EncodeBody(byte[])` helper |

Base only frames the header (or element wrapper). **Every concrete Package implements body encode/decode.** `OnBytesReady` is gone — field parsing lives in `DecodeBody`.

## Binary (.xls)

```
network / file Stream
│
OLE/XlsStream (sector assembly) ← NOT a Package
│ workbook byte[]
▼
PackageParser ── recognizes a finished local unit ──► Package
│ │
│ (continue) ▼
│ Renderer / DataSet builder
BinaryPackageParser.PushData
│ CreateEmpty(BIFFRECORDTYPE) → concrete XlsBiff* Package
▼
BinaryPackage (base in Reader.Binary)
├── EncodePackage = header + EncodeBody
├── DecodePackage / IPackage.DecodeBody = header → DecodeBody(body)
├── XlsBiffBOF, XlsBiffRKCell, XlsBiffLabelSSTCell, …
└── BinaryPackage itself (unknown types: raw body passthrough)
```

Examples of a “finished local unit”:
**Each `XlsBiff*` record class in `Reader/Binary/` is a Package** (one Office BIFF command type = one class).

| Format | Package | Completeness boundary |
| --- | --- | --- |
| BIFF (.xls) | `XlsBiffRecord` (: `BinaryPackage`) | one record header (`type`+`size`) + body |
| OpenXML sheet | `XmlCellPackage` / `XmlRowPackage` | finished `</c>` / `</row>` |
| OpenXML | `XmlMergePackage` / `XmlHyperlinkPackage` | finished empty element |
| Is a Package | Is NOT a Package |
| --- | --- |
| `BinaryPackage` + every `XlsBiff*` record type | `XlsStream`, `XlsHeader`, `XlsFat`, `XlsRootDirectory` (OLE) |
| | `XlsWorksheet`, `XlsWorkbookGlobals`, `ExcelBinaryReader` (workbook semantics) |
| | `BiffWorkbookCursor` (Seek/ReadAt index helper) |

A progressive renderer can subscribe to `IPackageParser<T>.Parse()` (or
`IPullPackageParser.TryRead`) and paint each package without waiting for EOF.
- BIFF header = 2-byte type + 2-byte body length; body from Office header.
- Factory: `BinaryPackage.CreateEmpty(type)` → typed subclass → `DecodePackage` / `AttachSharedBytes` → `DecodeBody` fills Office fields.
- Every concrete `XlsBiff*` owns **field-level** `DecodeBody` (buffer → members) and `EncodeBody` (members → buffer). No raw-body cheat on known types.
- `BinaryPackage` base `CaptureRawBody` remains only for **unknown** record types.
- Opaque known types (`CONTINUE`, `QUICKTIP`) store `byte[] m_payload` as the Office body member — still field-level.
- **SST:** members `m_count`, `m_uniqueCount`, `m_stringData` (bytes after the 8-byte header). CONTINUE-spanned strings stay Reader-side (`Append` + `ReadStrings`).
- **HyperLink:** range, GUID, flags, optional description/frame blocks, and URL are members; Encode rebuilds that structure.
- Formula string results may look ahead to the next STRING package on the shared buffer.
- `XlsBiffBlankCell` is a mid-base for shared row/col/xf; subclasses call `base.DecodeBody` then extend (except types that rewrite the full body).

## What this codebase does today
## OpenXML (.xlsx)

- **Binary**: `BinaryPackageParser` slices **one BIFF record** into an owned byte
buffer per emit. OLE sector assembly still materializes the workbook stream
(compound-file seeks); record emission itself is one-package-at-a-time.
- **OpenXML**: `ZipWorker` keeps the ZIP open and **streams entry inflate**
(`GetInputStream`) instead of extracting the whole archive to disk.
`XmlPackageParser` emits row/cell/merge/hyperlink packages from `XmlReader`.
- **Batch path**: markdown/json/tex still build a `DataSet` by consuming the
package stream (convenient, not required for progressive UI).
```
worksheet XML stream (file / NetworkStream / PushData buffer)
│
▼
XmlPackageParser (XmlReader tokenize)
│ complete SpreadsheetML element
▼
XmlPackage.CreateEmpty(localName) → Decode(reader) → queue
│
EncodeBody(XmlWriter) writes the same Office element shape
```

## Non-seekable network ZIP
Office element packages only:

Fully forward-only OPC (single `ZipInputStream` pass over a non-seekable network
body) is a follow-up: sheet parts are not contiguous with workbook.xml. Seekable
streams (files, buffered downloads) are supported via `ZipFile` entry streaming.
| Element | Package |
| --- | --- |
| `dimension` | `XmlDimensionPackage` |
| `row` | `XmlRowPackage` (attrs + child `c` cell packages) |
| `c` | `XmlCellPackage` (attrs `r`/`t`/`s`; children `v`/`t`/`f`) |
| `mergeCell` | `XmlMergePackage` |
| `hyperlink` | `XmlHyperlinkPackage` |

## TCP demo
`worksheet` / `sheetData` / `mergeCells` / `hyperlinks` are **parser state** — not packages.

Run a loopback server/client that streams sheet XML and BIFF packages:
`ExcelOpenXmlReader.ReadSheetGlobals` obtains `XmlDimensionPackage` via `XmlPackageParser` when present; sheets without `dimension` still fall back to counting row/cell packages for size.

## TCP demo

```bash
dotnet run --project src/Exceltk/Exceltk.csproj -- -t tcpstream \
-xlsx src/test/test1.xlsx -biff src/test/test8.xls
```

(or from `src/`: `dotnet run --project Exceltk/Exceltk.csproj -- -t tcpstream`)

The receiver prints packages as they complete (`[xml-stream] row#…`, `[biff-stream] #…`)
and exits with `TCP stream test OK` / `Done!` on success. Wired into `src/test.sh`.
- **BIFF**: server sends raw workbook stream chunks; client `BinaryPackageParser.PushData` pops concrete `XlsBiff*` packages.
- **XML**: server sends worksheet XML chunks; client `XmlReader` + `XmlPackageParser.TryFrameOne` / `PopPackage`.
Loading
Loading