fix: pass -Tp on MSVC for .cc so they are not treated as objects

[cestercian]

Summary

MSVC does not treat .cc as C++. With cpp(true), mimalloc 0.1.52 writes OUT_DIR/mimalloc-static.cc; without /TP, cl.exe assumes that file is an object (D9024) and never produces the -Fo output (*-mimalloc-static.o).

Pass -TP on MSVC-like compilers when cpp(true).

Fixes #1877

Test plan

• cargo +stable test --locked --test test — 59 passed (includes msvc_cpp_cc_source_not_treated_as_object)
• clippy on cc clean

[NobodyXu]

Thank you!

Is it possible to use -Tp instead, to recognise the immediate following .cc as c++ source files?

/TP could potentially break c code as C++ as slightly different rules, sometimes stricter

[cestercian]

@NobodyXu Switched to per-file -Tp immediately before the .cc path (no compiler-wide /TP / -TP), so C inputs and flag probes stay C. Regression test now asserts that shape. Head: de5683e.

[NobodyXu]

Thank you, just want the implementation to be changed as described and have the doc updated

You would probably want to update regression test to ensure C files aren't interpreted as C++, and that /Tp isn't used for clang-cl

[NobodyXu]

Ok there's a few issues we need to solve here

• we shouldn't need to check !is_assembler_msvc for /Tp
• -- is used for clang-cl to signal end of the options here so we cannot pass /Tp afterwards, however the good news is, we don't need it because clang-cl can recognise .cc
• we need to pass /Tp only for .cc files

so we need to do something like this:

Suggested change

	if self.cpp && compiler.is_like_msvc() && !is_assembler_msvc {
	// MSVC recognizes only `.c` / `.cpp` / `.cxx` as source. A `.cc`
	// file (mimalloc 0.1.52 writes `OUT_DIR/mimalloc-static.cc`) is
	// assumed to be an object unless `-Tp` forces C++ compilation
	// (#1877).
	cmd.arg("-Tp");
	} else if compiler.supports_path_delimiter() && !is_assembler_msvc {
	// #513: For `clang-cl`, separate flags/options from the input file.
	// When cross-compiling macOS -> Windows, this avoids interpreting
	// common `/Users/...` paths as the `/U` flag and triggering
	// `-Wslash-u-filename` warning.
	cmd.arg("--");
	}
	if
	self.cpp &&
	src.extension == Some("cc") &&
	matches!(compiler.family, ToolFamily::Msvc { clang_cl: false })
	{
	// MSVC recognizes only `.c` / `.cpp` / `.cxx` as source. A `.cc`
	// file (mimalloc 0.1.52 writes `OUT_DIR/mimalloc-static.cc`) is
	// assumed to be an object unless `-Tp` forces C++ compilation
	// (#1877).
	cmd.arg("-Tp");
	}
	if compiler.supports_path_delimiter() && !is_assembler_msvc {
	// #513: For `clang-cl`, separate flags/options from the input file.
	// When cross-compiling macOS -> Windows, this avoids interpreting
	// common `/Users/...` paths as the `/U` flag and triggering
	// `-Wslash-u-filename` warning.
	cmd.arg("--");
	}

[cestercian]

makes sense, narrowed it to -Tp only for .cc on MSVC (not clang-cl), and left the -- path alone

[NobodyXu]

Suggested change

	/// On MSVC this also passes `-Tp` immediately before each source file, so
	/// inputs such as `.cc` are compiled as C++ rather than assumed to be
	/// On MSVC this also passes `-Tp` immediately before each `.cc` source file to ensure that
	/// they are compiled as C++ rather than assumed to be

[cestercian]

updated the cpp() docs to match

[NobodyXu]

Thank you!

A new release is scheduled on next weekend, if you need it early, I can cut an early release