Temporary values are not destructed in C, e.g. string interpolation leaks memory

[yeputons]

Consider OpAddAssignString.ci:

		s2 += p + $"{p}" + s; //FAIL: cpp - should work with C++20

gets compiled to

	CiString_Assign(&s2, CiString_Format("%s%s%s%s", s2, p, p, s));

here, the result of CiString_Format is a result of malloc. It's passed to CiString_Assign and abandoned afterwards, leaking memory. This does not happen in C++ because temporary values are automatically destructed in C++, including std::string.

I assume there may be other instances where temporary objects are created and not immediately stored into variables.

[pfusik]

All tests are run in Travis CI. C, C++ and Swift tests are run with the Address Sanitizer and OpAddAssignString passes.

There's this line in the C translation of OpAddAssignString:

free(s2);

Why do you think it leaks memory?

[yeputons]

I'm sorry, I misread the example. Here is a simpler one:

public static class Test
{
	static void Foo(string x) {
	}
	public static void Run()
	{
	    int x;
	    Foo($"{x}");
	}
}

gets compiled to

static void Test_Foo(const char *x)
{
}

void Test_Run(void)
{
	int x;
	Test_Foo(CiString_Format("%d", x));
}

There is no free in this program at all.

[pfusik]

Yes, it's a known issue. I've just added a test and will fix it.

Interpolated strings are even worse when targetting C++, because they use C++20 <format>. Even though C++20 was ratified some time ago, clang only now starts implementing <format>.

[yeputons]

I believe the problem also exists with dynamic references in general. E.g. the following code:

class Foo {
}
public static class Test
{
	public static void Consume(Foo f) {}
	public static Foo# Create() { return new Foo(); }
	public static void Run() {
		Consume(Create());
	}
}

leaks memory and does not induce the Dynamically allocated objects must be assigned to a Foo# reference error, as Consume(new Foo()) would.

[pfusik]

cito cannot determine all the dangling reference scenarios. If it were easy, C and C++ compilers would do it, given five decades of their development. Consume(Create()) would be valid code if Create() stored the dynamic reference somewhere. The Create method itself is valid if used correctly.

[jedwards1211]

FWIW, I just learned that smart pointers are possible in C with GNU compilers: https://stackoverflow.com/questions/799825/smart-pointers-safe-memory-management-for-c

[Timerix22]

This issue is still unfixed for string concatination. The following code

string s1 = "1";
string s2 = "2";
string s3 = "3";
string s4 = "4";
string s5 = "5";
string() total = s1+s2+s3+s4+s5;

is transpiling to

const char *s1 = "1";
const char *s2 = "2";
const char *s3 = "3";
const char *s4 = "4";
const char *s5 = "5";
char *futemp0 = FuString_Format("%s%s", s1, s2);
char *futemp1 = FuString_Format("%s%s", FuString_Format("%s%s", s1, s2), s3);
char *futemp2 = FuString_Format("%s%s", FuString_Format("%s%s", FuString_Format("%s%s", s1, s2), s3), s4);
char *total = FuString_Format("%s%s", FuString_Format("%s%s", FuString_Format("%s%s", FuString_Format("%s%s", s1, s2), s3), s4), s5);

There are temporal variables declared correctly (futemp0, futemp1, futemp2), but this variables are not used in the FuString_Format() calls.

[Timerix22]

P.S. In the case of interpolation there is no such issue.

string() total = $"{s1}{s2}{s3}{s4}{s5}";

All FuString_Format() calls are merged in one:

char *total = FuString_Format("%s%s%s%s%s", s1, s2, s3, s4, s5);

[pfusik]

@Timerix22 af1f308 fixes the concatenation of multiple strings.

I will work on the remaining cases.

[pfusik]

All cases reported here are now fixed.