golang-naming

Go favors short, readable names. Capitalization controls visibility — uppercase is exported, lowercase is unexported. All identifiers MUST use MixedCaps, NEVER underscores.

"Clear is better than clever." — Go Proverbs

"Design the architecture, name the components, document the details." — Go Proverbs

To ignore a rule, just add a comment to the code.

All Go identifiers MUST use MixedCaps (or mixedCaps). NEVER use underscores in identifiers — the only exceptions are test function subcases (TestFoo_InvalidInput), generated code, and OS/cgo interop. This is load-bearing, not cosmetic — Go's export mechanism relies on capitalization, and tooling assumes MixedCaps throughout.

// ✓ Good
MaxPacketSize
userCount
parseHTTPResponse

// ✗ Bad — these conventions conflict with Go's export mechanism and tooling expectations
MAX_PACKET_SIZE   // C/Python style
max_packet_size   // snake_case
kMaxBufferSize    // Hungarian notation

Go call sites always include the package name, so repeating it in the identifier wastes the reader's time — http.HTTPClient forces parsing "HTTP" twice. A name MUST NOT repeat information already present in the package name, type name, or surrounding context.

// Good — clean at the call site
http.Client       // not http.HTTPClient
json.Decoder      // not json.JSONDecoder
user.New()        // not user.NewUser()
config.Parse()    // not config.ParseConfig()

// In package sqldb:
type Connection struct{}  // not DBConnection — "db" is already in the package name

// Anti-stutter applies to ALL exported types, not just the primary struct:
// In package dbpool:
type Pool struct{}        // not DBPool
type Status struct{}      // not PoolStatus — callers write dbpool.Status
type Option func(*Pool)   // not PoolOption

These conventions are correct but non-obvious — they are the most common source of naming mistakes:

Constructor naming: When a package exports a single primary type, the constructor is New(), not NewTypeName(). This avoids stuttering — callers write apiclient.New() not apiclient.NewClient(). Use NewTypeName() only when a package has multiple constructible types (like http.NewRequest, http.NewServeMux).

Boolean struct fields: Unexported boolean fields MUST use is/has/can prefix — isConnected, hasPermission, not bare connected or permission. The exported getter keeps the prefix: IsConnected() bool. This reads naturally as a question and distinguishes booleans from other types.

Error strings are fully lowercase — including acronyms. Write "invalid message id" not "invalid message ID", because error strings are often concatenated with other context (fmt.Errorf("parsing token: %w", err)) and mixed case looks wrong mid-sentence. Sentinel errors should include the package name as prefix: errors.New("apiclient: not found").

Enum zero values: Always place an explicit Unknown/Invalid sentinel at iota position 0. A var s Status silently becomes 0 — if that maps to a real state like StatusReady, code can behave as if a status was deliberately chosen when it wasn't.

Subtest names: Table-driven test case names in t.Run() should be fully lowercase descriptive phrases: "valid id", "empty input" — not "valid ID" or "Valid Input".

For complete rules, examples, and rationale, see:

• Packages, Files & Import Aliasing — Package naming (single word, lowercase, no plurals), file naming conventions, import alias patterns (only use on collision to avoid cognitive load), and directory structure.

• Variables, Booleans, Receivers & Acronyms — Scope-based naming (length matches scope: i for 3-line loops, longer names for package-level), single-letter receiver conventions (s for Server), acronym casing (URL not Url, HTTPServer not HttpServer), and boolean naming patterns (isReady, hasPrefix).

• Functions, Methods & Options — Getter/setter patterns (Go omits Get so user.Name() reads naturally), constructor conventions (New or NewTypeName), named returns (for documentation only), format function suffixes (Errorf, Wrapf), and functional options (WithPort, WithLogger).

• Types, Constants & Errors — Interface naming (Reader, Closer suffix with -er), struct naming (nouns, MixedCaps), constants (MixedCaps, not ALL_CAPS), enums (type name prefix like StatusReady), sentinel errors (ErrNotFound variables), error types (PathError suffix), and error message conventions (lowercase, no punctuation).

• Test Naming — Test function naming (TestFunctionName), table-driven test field conventions (input, expected), test helper naming, and subcase naming patterns.

Applying these fixes means renaming existing identifiers — → See samber/cc-skills-golang@golang-gopls skill to do it safely: its rename updates every call site across the workspace and refuses a rename that would break interface satisfaction, which a grep/sed or manual Edit-based rename silently misses.

Many naming convention issues are caught automatically by linters: revive, predeclared, misspell, errname. See samber/cc-skills-golang@golang-lint skill for configuration and usage.

• → See samber/cc-skills-golang@golang-code-style skill for broader formatting and style decisions
• → See samber/cc-skills-golang@golang-structs-interfaces skill for interface naming depth and receiver design
• → See samber/cc-skills-golang@golang-lint skill for automated enforcement (revive, predeclared, misspell, errname)
• → See samber/cc-skills-golang@golang-gopls skill for safe rename when applying a naming fix
• → See samber/cc-skills-golang@golang-refactoring skill for how to apply a rename safely at scale (gopls Rename/Inline, blast-radius mapping, staged PR workflow) once you've decided what to rename identifiers to