Bug 2479 – Cannot use variadic arguments inside a closure

char [32] tmp; sformat(tmp,"hi"); writefln(tmp); tmp is null after calling sformat. In some other cases an EXCEPTION_ACCESS_VIOLATION is thrown. Oddly enough format works fine The problem seems to be in passing the variable argument list to sformat, the argptr passed to sformat is null. Workaround is to go back to 2.021

Set importance to regression.

A reduced test case (larger, but complete and doesn't import large swaths of code). This is a massive reduction of all the irrelevant parts of doFormat torn away, and sformat reduced a good bit as well. module bug2479; import std.stdarg; void doFormat(void delegate(dchar) myputc, TypeInfo[] arguments, va_list argptr) { string fmt = va_arg!(string)(argptr); assert(!(fmt is null)); // fires when it shouldn't. } char[] sformat(char[] s, ...) { void putc(dchar c) { s[0] = cast(char)c; } doFormat(&putc, _arguments, _argptr); return s[0 .. 1]; } void main() { char[32] tmp; sformat(tmp, "%s", "hi"); } Changing putc to not touch 's' causes the va_arg call to succeed an fmt is no longer null.

Adding a printf at the top of doFormat... The broken case: argptr = 0xf7ceeffc [email protected](11): Assertion failure With putc emptied out: argptr = 0xffb004d4 (no assert)

Even further reduced, removing all imports. My patch to bug 814 does not fix this. ------ alias void* va_list; T va_arg(T)(inout va_list _argptr) { T arg = *cast(T*)_argptr; _argptr = _argptr + ((T.sizeof + int.sizeof - 1) & ~(int.sizeof - 1)); return arg; } void doFormat(void delegate(dchar) myputc, TypeInfo[] args, va_list argptr){ string fmt = va_arg!(string)(argptr); assert(!(fmt is null)); // fires when it shouldn't. } char[] sformat(char[] s, ...){ void putc(dchar c) { s[0] = cast(char)c; } doFormat(&putc, _arguments, _argptr); return s[0 .. 1]; } void main(){ char[32] tmp; sformat(tmp, "%s", "hi"); }

Reduced test case. This does not seem to be a Phobos bug. If you remove any reference to s from the delegate (so that it stops being a closure), the code works correctly. ---- void doFormat(void delegate() myputc, void * argptr){ assert(!(*cast(string*)argptr is null)); } void sformat(string s, ...){ void putc() { assert(s[0]==s[0]); } doFormat(&putc, _argptr); } void main(){ sformat("xxx", "hi"); }

I have patched Phobos so that original symptom is fixed. The compiler bug is that in FuncDeclaration::buildClosure() in toir.c, variadic arguments aren't supported, but _argptr is set in FuncDeclaration::semantic3() as if they were. Partial patch, to turn this from a wrong-code into a rejects-valid bug, until the bug is actually fixed: func.c, line 1282. --- if (argptr) { // Initialize _argptr to point past non-variadic arg #if IN_GCC // Handled in FuncDeclaration::toObjFile v_argptr = argptr; v_argptr->init = new VoidInitializer(loc); #else + // BUG: Fix this in FuncDeclaration::buildClosure() in toir.c. + if (needsClosure()) + error("Closures are not yet supported with variadic arguments (Bugzilla 2479)"); Type *t = argptr->type; VarDeclaration *p; unsigned offset;

SVN commit (workarround in Phobos, not a real fix): http://www.dsource.org/projects/phobos/changeset/1318

I'll add a message for now, dmd svn 239.

Direct link to the SVN commit: http://www.dsource.org/projects/dmd/changeset/239

Was: Badcode regression: closure corrupts _argptr value In DMD2.036, this is now just a rejects-valid rather than wrong-code.

Removing the 'regression' keyword, since variadic arguments in closures have never worked.

Git master does not exhibit this behavior. The code compiles and runs successfully. Closing as WORKSFORME.