[Code scan] Treat short POSIX restart reads and writes as failures

[njzjz]

This issue is a result of a Codex global repository scan.

The binary restart helpers call POSIX write() and read() once and only treat -1 as failure. POSIX allows a positive return smaller than the requested byte count. In that case these helpers currently report success even though only part of the restart payload was transferred.

abacus-develop/source/source_io/module_restart/restart.cpp

Lines 36 to 55 in 84ca04b

	bool Restart::write_file2(const std::string& file_name, const void* const ptr, const size_t size, const bool error_quit) const
	{
	const int file = open(file_name.c_str(), O_WRONLY|O_CREAT|O_TRUNC, S_IRUSR|S_IWUSR);
	if (-1 == file){
	if (error_quit){
	throw std::runtime_error("can't open restart save file. \nerrno=" + ModuleBase::GlobalFunc::TO_STRING(errno) + ".\n" + std::string(__FILE__) + " line " + std::to_string(__LINE__));
	} else {
	return false;
	}
	}
	auto error = write(file, ptr, size);
	if (-1 == error) {
	if (error_quit) {
	throw std::runtime_error("can't write restart save file. \nerrno=" + ModuleBase::GlobalFunc::TO_STRING(errno) + ".\n" + std::string(__FILE__) + " line " + std::to_string(__LINE__));
	} else {
	return false;
	}
	}
	close(file);
	return true;

abacus-develop/source/source_io/module_restart/restart.cpp

Lines 63 to 82 in 84ca04b

	bool Restart::read_file2(const std::string& file_name, void* const ptr, const size_t size, const bool error_quit) const
	{
	const int file = open(file_name.c_str(), O_RDONLY);
	if (-1 == file) {
	if (error_quit) {
	throw std::runtime_error("can't open restart load file. \nerrno=" + ModuleBase::GlobalFunc::TO_STRING(errno) + ".\n" + std::string(__FILE__) + " line " + std::to_string(__LINE__));
	} else {
	return false;
	}
	}
	auto error = read(file, ptr, size);
	if (-1 == error) {
	if (error_quit) {
	throw std::runtime_error("can't read restart load file. \nerrno=" + ModuleBase::GlobalFunc::TO_STRING(errno) + ".\n" + std::string(__FILE__) + " line " + std::to_string(__LINE__));
	} else {
	return false;
	}
	}
	close(file);
	return true;

Caller-facing byte count construction:

abacus-develop/source/source_io/module_restart/restart.h

Lines 32 to 38 in 84ca04b

	template<typename T>
	bool save_disk(const std::string label, const int index, const int size, T* data, const bool error_quit = true) const
	{
	return write_file2(folder + label + "_" + ModuleBase::GlobalFunc::TO_STRING(GlobalV::MY_RANK) + "_"
	+ ModuleBase::GlobalFunc::TO_STRING(index),
	data,
	size * sizeof(T),

Relevant code:

auto error = write(file, ptr, size);
if (-1 == error) { ... }
...
auto error = read(file, ptr, size);
if (-1 == error) { ... }

Impact:

A truncated restart shard can be accepted as successfully loaded, leaving part of the destination buffer stale and making restart behavior non-deterministic.

Suggested fix:

Loop until exactly size bytes are transferred, or return/throw on EOF or a short write. Use RAII for the file descriptor so open descriptors are closed on all error paths.