[michvllni]

Replaces source/libs/inih/ini.{c,h} with upstream benhoyt/inih 577ae2d (2026-01-30), byte for byte, and raises INI_MAX_LINE from 1024 to 2048 in BUILD_FLAGS. A pasted Token= is an OAuth access token and Google's run past a kilobyte.

Keeping the two files identical to upstream means the next resync is a two-file copy with no patch to re-apply, so inih's compile-time options stay in the Makefile. source/libs/inih/INIReader/ is untouched — it is a local fork, and GetValues() (which getConfiguredSyncPaths() depends on) has no upstream equivalent.

Why this update is worth taking

It fixes an error misattribution this project actually hits. Given a token longer than the buffer, the old parser cut the line and parsed its tail as the next line, so ParseError() named the line after the offending one — debugConfigSummary() would report syntax error on line 3, that line was ignored while pointing at a perfectly valid Path=. Upstream now discards the overflow and reports the real line number.

Validation

Everything below was run against the vendored copy, not upstream's tree.

• Clean 3DS build from scratch: zero warnings, all artifacts produced (.3dsx, .3ds, .cia, zip). ini.c and INIReader.cpp are also clean under devkitARM at -Wall -Wextra -Wpedantic.
• Host test suite: 115 passed, 0 failed. (These never touched inih, hence the extra work below.)
• Upstream's own test matrix: all 15 configurations match their baselines — multiline, heap, realloc, custom allocator, INI_HANDLER_LINENO=1, and the rest.
• Old-vs-new differential over a 17-file corpus of README-shaped configs (1.1 KB token, inline comments, multiline continuations, BOM, CRLF, unterminated sections, over-long section/key names, no trailing newline). Only two changes, both improvements: the line numbering above, and inline comments now stripped from continuation lines (inert here — a continuation value would already be broken as a filesystem path).
• Invariants re-checked per AGENTS.md: INI_HANDLER_LINENO still defaults to 0, so INIReader::ValueHandler's four-argument form is still correct; ini_parse/ini_parse_string signatures are unchanged apart from the no-op INI_API; the renamed statics (rstrip → ini_rstrip and friends) were all file-local, so nothing outside ini.c referenced them.

One upstream quirk, left unpatched

A line whose content is exactly INI_MAX_LINE - 1 bytes and ends in a newline parses completely, value intact, yet is still flagged as a syntax error. Reproduced standalone against ini.c alone at INI_MAX_LINE=20, so it is upstream's off-by-one rather than a bad copy.

At 2048 this needs a 2047-byte config line; main() only bails on ParseError() < 0 so a sync still runs, and the symptom is a debug-only message claiming a line was ignored when it was not. Not worth breaking the byte-identical invariant for — better filed upstream.

Notes, no action needed

The new assert()s on NULL arguments are live (NDEBUG is unset, __assert_func links in) but unreachable through INIReader — ini_parse only calls through after a successful fopen. And since INI_USE_STACK=1, the 1024 → 2048 bump grew ini_parse_stream's frame to 2196 bytes; it is called from main(), so there is ample stack.

🤖 Generated with Claude Code

adresses #14