c99-guide

C99 Coding Guidelines

Essentials

• Memory management - Free all heap allocations, avoid leaks, see references/memory-management.md
• Modern C99 - Use const, designated initializers (ZII), small functions, see references/designated-initializers.md, references/const-correctness.md
• Fixed-width types - Name widths with <stdint.h> for sized/serialized data; size_t for counts and indices, see references/fixed-width-types.md
• Type safety - Favor inline functions over macros where practical, see references/inline-functions.md
• Quality - Pin strict C99, treat warnings as errors, keep an ASan/UBSan debug build, run static analysis, see references/build-and-warnings.md

Best practices

• Error handling - Use return codes, check all fallible operations, see references/error-handling.md
• Value-oriented APIs - Return small results by value or in a {ok, value} result struct; reserve out-params for large or multiple results, see references/value-types.md
• Strings - Borrow length-carrying views, write through bounded builders over caller memory — not strlen/strcat/strtok rescans, see references/string-views.md
• Input validation - Check bounds, NULL pointers, division by zero
• Readability - Small functions, clear naming, comments for non-obvious logic
• Paradigm - Functional style (pure functions, explicit context) → fp-guide; object/data modeling → oop-guide

Gotchas

• Signed integer overflow is undefined behavior — even INT_MAX + 1 lets the compiler eliminate "impossible" code paths
• Strict aliasing means casting int* to float* is UB unless via union or memcpy — silently miscompiled at higher optimization levels
• malloc returns memory that's max-aligned; custom arenas must preserve alignment for _Bool/double/SIMD types
• Designated initializers (.field = x) zero-fill unmentioned members — a missing field becomes silent zero, not a compile error
• long is 32-bit on Windows and 64-bit on most 64-bit Unix targets — a struct of int/long serializes and hashes differently per platform; use <stdint.h> exact-width types for any stored or shared data
• Comparing a signed int against an unsigned size_t converts the int — a negative value becomes a huge size_t and the bounds check passes; keep counts/indices in size_t end to end
• Return-by-value is for small PODs; returning a large struct just copies it — large or caller-owned results still take a pointer
• -std=c99 usually still means the GNU dialect — set C_EXTENSIONS OFF for true ISO C99, then #define _XOPEN_SOURCE 700 or POSIX calls (readlink, strnlen, ssize_t) become implicit-declaration errors
• -Wextra's -Wmissing-field-initializers/-Wmissing-braces fire on intentional zero-init — suppress those two, keep the rest of -Werror
• A bare snprintf into a fixed buffer trips -Wformat-truncation; check its return (>= size ⇒ truncated) to handle the case and clear the warning

Progressive disclosure

• Read references/build-and-warnings.md - Load when configuring the C standard, feature-test macros, warning flags, or sanitizers
• Read references/fixed-width-types.md - Load when choosing integer types for struct fields, serialized data, counts, or indices
• Read references/string-views.md - Load when handling strings without repeated terminator scans or hidden allocations
• Read references/value-types.md - Load when designing function signatures: returning results by value vs. out-parameters
• Read references/memory-management.md - Load when allocating memory or managing resource lifetimes
• Read references/designated-initializers.md - Load when initializing structs or arrays with specific values
• Read references/inline-functions.md - Load when replacing macros or writing small utility functions
• Read references/compound-literals.md - Load when creating temporary values without named variables
• Read references/const-correctness.md - Load when marking immutable data or understanding pointer const
• Read references/error-handling.md - Load when implementing error codes or handling failures