[libc][test] spew of `-Wuninitialized` with gcc-14

[nickdesaulniers]

I'm seeing a large spew of:

In constructor ‘__llvm_libc_20_0_0_git::testing::FormatSectionMatcher::FormatSectionMatcher(__llvm_libc_20_0_0_git::printf_core::FormatSection)’,
    inlined from ‘virtual void LlvmLibcPrintfParserTest_DoublePercentIsAllowedInvalidIndex::Run()’ at /android0/llvm-project/libc/test/src/stdio/printf_core/parser_test.cpp:626:3:
/android0/llvm-project/libc/test/UnitTest/PrintfMatcher.h:25:9: error: ‘expected0.__llvm_libc_20_0_0_git::printf_core::FormatSection::conv_val_ptr’ is used uninitialized [-Werror=uninitialized]
   25 |       : expected(expectedValue) {}
      |         ^~~~~~~~~~~~~~~~~~~~~~~
/android0/llvm-project/libc/test/src/stdio/printf_core/parser_test.cpp: In member function ‘virtual void LlvmLibcPrintfParserTest_DoublePercentIsAllowedInvalidIndex::Run()’:
/android0/llvm-project/libc/test/src/stdio/printf_core/parser_test.cpp:621:46: note: ‘expected0.__llvm_libc_20_0_0_git::printf_core::FormatSection::conv_val_ptr’ was declared here
  621 |   LIBC_NAMESPACE::printf_core::FormatSection expected0;
      |                                              ^~~~~~~~~

when building the unit tests with gcc-14.

Perhaps EXPECT_PFORMAT_EQ is a macro that constructs a FormatSectionMatcher, which copy constructs the FormatSection expected (highlighted in the diagnostic), but the default implicit copy constructor of FormatSectionMatcher is copying over every field, where fields such as FormatSection::conv_val_ptr and FormatSection::conv_val_raw have not been initialized!

I wonder if we should either:

• have the unit tests that default construct FormatSection's initialize conv_val_ptr and conv_val_raw
• explicitly delete the implicit default constructor for FormatSection and update the tests
• other?

Observed while trying to test #124036.

[llvmbot]

@llvm/issue-subscribers-libc

Author: Nick Desaulniers (nickdesaulniers)

I'm seeing a large spew of: ``` In constructor ‘__llvm_libc_20_0_0_git::testing::FormatSectionMatcher::FormatSectionMatcher(__llvm_libc_20_0_0_git::printf_core::FormatSection)’, inlined from ‘virtual void LlvmLibcPrintfParserTest_DoublePercentIsAllowedInvalidIndex::Run()’ at /android0/llvm-project/libc/test/src/stdio/printf_core/parser_test.cpp:626:3: /android0/llvm-project/libc/test/UnitTest/PrintfMatcher.h:25:9: error: ‘expected0.__llvm_libc_20_0_0_git::printf_core::FormatSection::conv_val_ptr’ is used uninitialized [-Werror=uninitialized] 25 | : expected(expectedValue) {} | ^~~~~~~~~~~~~~~~~~~~~~~ /android0/llvm-project/libc/test/src/stdio/printf_core/parser_test.cpp: In member function ‘virtual void LlvmLibcPrintfParserTest_DoublePercentIsAllowedInvalidIndex::Run()’: /android0/llvm-project/libc/test/src/stdio/printf_core/parser_test.cpp:621:46: note: ‘expected0.__llvm_libc_20_0_0_git::printf_core::FormatSection::conv_val_ptr’ was declared here 621 | LIBC_NAMESPACE::printf_core::FormatSection expected0; | ^~~~~~~~~

when building the unit tests with gcc-14.

Perhaps `EXPECT_PFORMAT_EQ` is a macro that constructs a `FormatSectionMatcher`, which copy constructs the `FormatSection` `expected` (highlighted in the diagnostic), but the default implicit copy constructor of `FormatSectionMatcher` is copying over every field, where fields such as `FormatSection::conv_val_ptr` and `FormatSection::conv_val_raw` have not been initialized!

I wonder if we should either:
- have the unit tests that default construct `FormatSection`'s initialize `conv_val_ptr` and `conv_val_raw`
- explicitly delete the implicit default constructor for `FormatSection` and update the tests
- other?

Observed while trying to test #<!-- -->124036.
</details>

[nickdesaulniers]

I wonder if we should either:

• default initialize FormatSection::conv_val_ptr, FormatSection::conv_val_raw (oh, and FormatSection::conv_name; that's why there's a spew of diagnostics; all three are not initialized by default).