[siracozmen01]

Problem

MUNI_NODISCARD is meant to make every function that returns a status warn when its result is dropped. In C it expanded to [[nodiscard]] only when __STDC_VERSION__ >= 202311L. GCC 14 has the attribute in its C23 mode but reports __STDC_VERSION__ as 202000L there, under -std=c23 and -std=c2x alike. So under GCC the macro was empty:

• A program built with GCC 14 could drop any status from the library without a warning.
• The library's own GCC cells checked none of its internal calls either.

Clang 20 reports 202311L, which is why the clang cells never showed this. maul-locale found the same bug (Rawframe-Project/maul-locale#483) and fixed it in Rawframe-Project/maul-locale#484.

Change

In C, include/maul-unicode/base.h now takes the attribute in any dialect after C17 (__STDC_VERSION__ > 201710L) whose compiler defines __has_c_attribute and reports nodiscard. The __has_c_attribute test is nested under defined(__has_c_attribute), so a compiler without the operator never parses it. C17 and older dialects keep the empty expansion, and the C++ branch is unchanged. A comment explains GCC 14's 202000L. The block is maul-locale's at c346644, byte for byte except the prefix.

test/test_base.c asserts at compile time that the macro expands to something. Every suite builds as C23 with a compiler that has the attribute, so if the macro goes empty again, the build fails.

Verification

• With main's base.h, test_base fails to build under GCC 14 with "static assertion failed: MUNI_NODISCARD is the attribute in C23". With this change it builds.
• A file that drops the result of muniValidateUtf8 gets -Wunused-result from GCC 14 (-std=c23) with this change and nothing with main's header; Clang 20 warns with both.
• The linux-gcc-14 cell builds with the attribute in effect, -Werror included: library, tools, samples, benchmarks and tests. No status is dropped anywhere, and all 13 tests pass.
• The linux-harfbuzz cell (GCC 14, in ci-cpp:2026.10.7) builds the HarfBuzz adapter and its suites the same way, and all 16 tests pass.
• tools/check.sh changed passes: rules, format, clang-tidy in Release and Debug, and the base and results suites under linux-clang-20-asan.

Closes #270

[siracozmen01]

Review at a9a85fd: ready. I read the change and did not run it. CI is pending.

The block in include/maul-unicode/base.h matches maul-locale's include/maul-locale/base.h at c346644 byte for byte apart from the prefix:

• the comment's six lines;
• the C++ branch;
• the #elif with its continuation and the backslash at the same column;
• the nested #if __has_c_attribute(nodiscard) and both #else / #endif pairs.

They are the same line for line, with MUNI_ in place of MLOC_. My review of maul-locale #484 holds for this copy. test/test_base.c takes the same TEXT/EXPANDED static_assert, with the prefix, after its existing includes, and its header comment now names the result attribute. fix without ! is right at 1.x: a program gets a warning where the attribute was always meant to apply. The body has no column-0 name( lines and no em dash.