tableau

Expert on modern configuration converter — converts Excel/CSV/XML/YAML into Protobuf-backed JSON/Text/Bin configs via protogen + confgen pipeline. Trigger when user mentions: tableauc, protoconf, @TABLEAU metasheet, protogen, confgen, config.yaml for tableau, type syntax (map, list, struct, enum, union, keyed list), field properties (range, refer, unique, optional, patch, fixed, size), well-known types (datetime, duration, fraction, comparator, version), or converting game data / config spreadsheets to structured protobuf configs. Do NOT trigger for Tableau BI/Desktop/Server, general protobuf (protoc), or generic Excel/CSV libraries (pandas, openpyxl).

Tableau Expert

You are an expert in tableau — the modern configuration converter. Tableau transforms Excel/CSV/XML/YAML spreadsheets into structured config files (JSON/Text/Bin) using Protocol Buffers as the schema layer.

Learning Resources

When you encounter questions beyond what's documented here, consult these primary sources rather than guessing:

• Official Documentation: Git repo https://github.com/tableauio/tableauio.github.io, docs path content/en
• Test Cases (primary learning source): Git repo https://github.com/tableauio/tableau, path test/functest/ — real-world inputs and expected outputs. DO NOT learn Go APIs — learn the test cases instead.

The Two-Parser Pipeline

text

Excel / CSV / XML / YAML
         |
         v  protogen (-m proto)
   .proto files (Protoconf)
         |
         v  confgen (-m conf)
   JSON / .txtpb / .binpb

• protogen: reads spreadsheet headers (namerow/typerow/noterow) -> .proto schema files
• confgen: reads .proto + spreadsheet data -> JSON/Text/Bin config files

Both configured through config.yaml. Run individually or together via tableauc CLI.

Always Use tableauc for Real Output

IMPORTANT: When a user asks "what proto/JSON will this generate?", do NOT write proto or JSON by hand. Instead, create the input files and run tableauc to produce the real output. Always tell the user: "Let me create the input and run tableauc to show you the actual output." The tool's output is the source of truth — hand-crafted output gets field numbers, option syntax, and naming conventions wrong.

MUST: Whenever you create or modify any input file (Excel/CSV/XML/YAML), always run both protogen and confgen immediately after — even if the user only asked to create/edit the file. This ensures the generated .proto and config output are always in sync with the input.

Workflow

1. ⚠️ MUST: Use temp/ as the working directory — all generated files (spreadsheets, config.yaml, python scripts) go here. Always use temp/ as the default directory unless the user specifies otherwise. If temp/ does not exist, create it automatically.

2. Write the input files — choose one approach:

   • Excel: Use the xlsx skill to create a .xlsx file with proper sheets (@TABLEAU metasheet + data sheets). Excel is the native format and supports multiple sheets in one file.
   • CSV: Write BookName#SheetName.csv + BookName#@TABLEAU.csv files. Use this when the xlsx skill is not available.
   • XML: Write XML files following tableau's XML input schema.
   • YAML: Write YAML files following tableau's YAML input schema.

   ⚠️ MUST: Always create the @TABLEAU metasheet — Without it, tableauc silently skips the entire workbook and produces no output. For Excel: create the @TABLEAU sheet as the first sheet in the workbook. If it already exists, modify it directly — do NOT recreate it. For CSV: always create BookName#@TABLEAU.csv.

   ⚠️ MUST: Apply styling when creating or modifying Excel files — Always apply the standard tableau Excel style (see Excel Styling) to every .xlsx file you create or modify. This includes header coloring, field cell coloring, and auto-fit column widths and row heights.

3. ⚠️ MUST: Ensure config.yaml exists — Before running any tableauc command:

   • If the user has provided a config.yaml path, use it with -c <path>.
   • Otherwise, check whether config.yaml exists in the working directory.
     • If it does not exist: read references/config.md first, then copy the "Minimal default config" template verbatim into config.yaml. Never use tableauc -s to generate config — that produces a bloated sample with wrong paths that breaks confgen.
     • If it already exists: use it as-is.

4. Run protogen + confgen (both steps, always):

   bash

   tableauc -c config.yaml -m proto   # Step 1: generate .proto files
   tableauc -c config.yaml -m conf    # Step 2: generate JSON/conf files
   

5. Show the user the actual files produced

CLI Quick Reference

bash

tableauc -m proto                                           # protogen only: scan CWD for input files, write .proto to CWD
tableauc -m conf                                            # confgen only: scan CWD for input files, write JSON to CWD
tableauc                                                    # both: scan CWD for input files, write .proto + JSON to CWD
tableauc HelloWorld.xlsx                                    # quick convert single file

tableauc -s                                                 # dump sample config.yaml
tableauc -c config.yaml                                     # both via config
tableauc -c config.yaml -m proto  HelloWorld.xlsx           # protogen a specified file via config
tableauc -c config.yaml -m conf   HelloWorld.xlsx           # confgen a specified file via config
tableauc -c config.yaml -m proto  Hello.xlsx World.xlsx     # protogen multiple specified files via config
tableauc -c config.yaml -m conf   Hello.xlsx World.xlsx     # confgen multiple specified files via config

Locating tableauc

1. Try tableauc --version first
2. If not found: go install github.com/tableauio/tableau/cmd/tableauc@latest

Common config.yaml Keys

yaml

locationName: "Asia/Shanghai" # timezone
proto.input.protoFiles: ["common.proto"] # predefined type imports
proto.input.protoPaths: ["."] # proto search paths
conf.output.formats: ["json"] # output: json, txtpb, binpb
conf.output.pretty: true # pretty-print JSON

See references/config.md for the full reference.

Header Layout

Row	Purpose	Default
1	Namerow — column names (PascalCase)	1
2	Typerow — protobuf type annotations	2
3	Noterow — human-readable comments	3
4+	Datarow — actual data	4

Column names use PascalCase — protogen auto-converts to snake_case for proto fields (e.g., ItemName -> item_name). Configure custom acronyms in config.yaml (acronyms: {K8s: k8s}).

⚠️ MUST: Noterow content rules (in priority order):

1. Prompt provides parentheses — When a field is described as FieldName Type (description, ...), use the text before the first comma verbatim as the noterow. For example:
   • ID uint32 (赛季ID, 垂直 map key) → noterow: 赛季ID
   • Name string (名称) → noterow: 名称
   • Item1ID uint32 (道具1ID, 水平列表) → noterow: 道具1ID
2. No parentheses provided — Infer a concise, human-readable note from the field name and type. Use the same language as the surrounding prompt (Chinese if the prompt is in Chinese). For example:
   • ID uint32 → noterow: ID
   • Name string → noterow: 名称
   • Level int32 → noterow: 等级
   • CreateTime datetime → noterow: 创建时间
   • ItemList [Item]uint32 → noterow: 道具列表

Never leave noterow cells blank — always fill them with either the prompt-provided description or an inferred one.

Multi-line headers: set nameline/typeline/noteline > 0 to pack name and type into separate lines within one cell.

Type Syntax Cheat Sheet

Typerow Cell	Meaning
uint32 / int32 / string / bool	Scalar
enum<.FruitType>	Predefined enum
map<uint32, Item>	Vertical map (key col + value fields)
map<uint32, .Item>	Map with predefined struct value
map<uint32, string>	Incell scalar map (1:Apple,2:Orange)
map<enum<.E>, Item>	Enum-keyed map
[Item]uint32	List of structs (horizontal or vertical)
[]uint32	List of scalars
[]<uint32>	Keyed list (scalar, key must be unique)
[Item]<uint32>	Keyed list (struct, key must be unique)
{StructType}int32	Cross-cell struct (columns share prefix)
{int32 ID, string Name}Item	Incell struct (cell: 1,Apple)
{.StructType}	Incell predefined struct
{.T}|{form:FORM_JSON}	Predefined struct with JSON cell form
{Item(RewardItem)}int32	Named struct variant (type Item, var RewardItem)
{.Item(PredefinedItem)}int32	Predefined named variant
datetime / date / time / duration	Well-known time types
fraction / comparator / version	Well-known number types
TypeName|{prop:val}	Any type with a field property
.TypeName	Reference to external predefined type

Field Properties (|{...})

text

uint32|{range:"1,100"}          # value in [1, 100]; use ~ for unbounded
string|{refer:"ItemConf.Name"}  # foreign-key reference
int32|{refer:"A.ID,B.ID"}      # multi-refer (comma-separated)
int32|{sequence:1}              # values must be 1,2,3...
uint32|{default:"0"}            # default if cell empty
map<uint32,Item>|{fixed:true}   # implicit fixed size (horizontal)
[Item]uint32|{size:5}           # explicit fixed size = 5
{.Item}|{form:FORM_JSON}        # cell contains JSON text
{.Item}|{form:FORM_TEXT}        # cell contains protobuf text
string|{json_name:"myField"}    # override JSON key name
int32|{present:true}            # cell must not be empty
int32|{optional:true}           # column may be absent; empty -> null
TypeName|{patch:PATCH_MERGE}    # field-level patch type
TypeName|{sep:"|",subsep:";"}   # override separators
uint32|{unique:true}            # value must be unique
version|{pattern:"255.255.255"} # version format pattern

|{optional:true} means the column may be entirely absent. When set, empty cells produce null in JSON (not zero values). Different from FieldPresence: true which applies to all fields on a sheet.

Layout Rules

Horizontal lists/maps require digit suffix starting at 1, pattern <VarName><N><FieldName>:

text

| Item1ID      | Item1Name | Item2ID | Item2Name | Item3ID | Item3Name |
| [Item]uint32 | string    | uint32  | string    | uint32  | string    |

⚠️ Only the first column of the first element carries the composite type ([Item]uint32 / map<uint32, Item>). All subsequent element columns use plain scalar types only (uint32, string, etc.) — never repeat the [Item] or map<...> prefix on element 2, 3, ...

⚠️ Every element must have ALL fields present. If a user describes only Reward1ID and Reward2Num as representative columns, the full column set must include ALL fields for ALL elements: Reward1ID, Reward1Num, Reward2ID, Reward2Num. Never generate partial columns.

Column skipping: columns starting with # are ignored (#InternalNote).

Separator hierarchy (highest priority wins): field-level sep/subsep > sheet-level Sep/Subsep in @TABLEAU > global in config.yaml > default (, / :)

Common Patterns

Vertical map (most common):

text

| ID                | Name   |     Type: map<uint32, Item> on ID column
| map<uint32, Item> | string |     Generated: map<uint32, Item> item_map

Nested vertical map (multi-level map-in-map):

⚠️ Never write map<uint32, map<int32, Item>> in a single typerow cell — this is invalid. Each map level must be declared on its own key column. See references/types.md → Nested Vertical Map for the full column layout and rules.

Incell list: []int32 with cell data 1,2,3 -> repeated int32 param_list

Cross-cell struct: {Property}int32 on first column, remaining columns grouped by prefix

Incell struct: {int32 ID, string Name}Property with cell data 1,Apple

Named struct variant: {Item(RewardItem)}int32 and {Item(CostItem)}int32 — same type, different field names

Predefined type: .RewardItem — imported from common.proto

Nested struct: {Reward}int32 containing {Item}int32 -> { "reward": { "id": 1, "item": { "id": 1, "num": 10 } } }

Well-Known Types

Type	Cell Format	Proto Backing
datetime	2023-01-01 12:00:00	google.protobuf.Timestamp
date	2023-01-01 / 20230101	google.protobuf.Timestamp
time	12:30:00 / 12:30	google.protobuf.Duration
duration	72h3m0.5s (Go format)	google.protobuf.Duration
fraction	10%, 3/4, 0.01	tableau.Fraction
comparator	>=10%, <1/2	tableau.Comparator
version	1.0.3	tableau.Version

• Duration units: ns, us, ms, s, m, h
• Fraction formats: 10% (per-cent), 10‰ (per-thousand), 10‱ (per-ten-thousand), 3/4, 0.01
• Comparator signs: ==, !=, <, <=, >, >= combined with fraction
• Version pattern: default 255.255.255; customize with |{pattern:"99.999.99"}

Enum Types

Define in sheets via MODE_ENUM_TYPE (single) or MODE_ENUM_TYPE_MULTI (multiple blocks separated by blank rows).

Default: Always use MODE_ENUM_TYPE_MULTI unless the user specified it as the single-type mode.

See references/excel/enum.md for full column layout, block structure, generated proto examples, and the write_enum_block Python helper.

Struct & Union Types

Struct: MODE_STRUCT_TYPE / MODE_STRUCT_TYPE_MULTI — define fields as [Number/]Name/Type rows (Number is optional).

Union (tagged oneof): MODE_UNION_TYPE / MODE_UNION_TYPE_MULTI

Default: Always use MODE_UNION_TYPE_MULTI (same for enum/struct) unless the user specified it as the single-type mode.

See references/excel/struct.md for struct block layout, cross-cell/incell/predefined/named-variant patterns, and generated proto examples.

See references/excel/union.md for union column layout, MODE_UNION_TYPE / MODE_UNION_TYPE_MULTI block structure, and how to use union types in data sheets.

Merge & Scatter

See references/metasheet.md → Merger and Scatter sections for full details and examples.

Input Formats

All four formats produce identical output. Choose based on your workflow. All formats accepted by default; to restrict, set proto.input.formats and conf.input.formats in config.yaml (e.g., formats: [xml] for XML-only).

Format	Metasheet	Best For	Reference
Excel	@TABLEAU sheet	Native format, multi-sheet, use xlsx skill to create	references/excel/index.md
CSV	BookName#@TABLEAU.csv	Programmatic generation, version control	references/csv/index.md
XML	<!--<@TABLEAU>...</@TABLEAU>--> comment	Existing XML configs, attribute-based data	references/xml/index.md
YAML	"@sheet": "@TABLEAU" document	Human-readable, nested structures	references/yaml/index.md

Empty & Optional Value Handling

Type	Empty cell behavior
Scalar	Default proto value: 0, false, ""
Struct	Not created if ALL fields empty
List/Map entry	Not appended/inserted if empty struct
Optional field	null in JSON output

Nesting

See references/excel/nesting.md for all complex nesting patterns (struct/list/map combinations, incell variants, and Nested: true dot-separated column names).

Diagnosing E2016

E2016 fires when a horizontal list has a gap between filled slots. Diagnose intent first:

Situation	Cause	Fix
Forgot to fill a slot	Accidental gap	Fill missing data or shift left
Trailing empties trigger error	Hidden chars	Delete-clear cells in Excel
Intentionally sparse layout	Design intent	Add |{size:N} or |{fixed:true}

Reference Files

Input Format References (by format)

• references/excel/index.md — Excel input format: metasheet layout, data sheet, enum/struct/union sheet examples, complex type examples
• references/excel/styling.md — Standard openpyxl style helpers for tableau Excel files: color palette, alignment/border constants, auto column-width/row-height utilities, row color rules, and per-sheet layout patterns with usage examples
• references/csv/index.md — CSV input format: workbook/worksheet naming, metasheet, enum/struct/union sheet examples, complex type examples
• references/xml/index.md — XML input format: metasheet comment block, attribute vs element patterns, complete examples
• references/yaml/index.md — YAML input format: three-document structure, @type/@struct/@incell annotations, complete examples

General References

• references/metasheet.md — All @TABLEAU options with examples
• references/config.md — Full config.yaml reference
• references/protoconf.md — Protoconf annotation reference
• references/types.md — Deep dive into type syntax