[dmd-internals] Struct initialization in nested functions

[Sean Kelly]

Here are three variations of code using the curl client.  The first works, the second fails with a segfault, and the third works:


#1

void main() {
   auto http = HTTP();
   go(http);
}

void go(ref HTTP http) {
   // compose request
   http.perform();
}

works!


#2

void main() {
   HTTP http;
   init(http);
   go(http);
}

void init(ref HTTP http) {
   http = HTTP();
}

void go(ref HTTP http) {
   // compose request
   http.perform();
}

segfault! (see below)


#3

void main() {
   HTTP http;
   init(http);
   go(http);
}

void init(ref HTTP http) {
   http.__ctor(null);
}

void go(ref HTTP http) {
   // compose request
   http.perform();
}

works!


I'm guessing the segfault in the second case is a compiler bug, but I haven't tried to narrow it down yet.  The problem initially came up because I wanted to use the HTTP struct as a class member, and the documented means for initializing it was causing a segfault.  Here's the stack trace:


Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00000001000028ec
0x0000000100077803 in D3std3net4curl4HTTP15onReceiveHeaderMFNdDFxAaxAaZvZv12__lambda2026MFxAaZv ()
(gdb) bt
#0  0x0000000100077803 in D3std3net4curl4HTTP15onReceiveHeaderMFNdDFxAaxAaZvZv12__lambda2026MFxAaZv ()
#1  0x0000000100078bfd in D3std3net4curl4Curl15onReceiveHeaderMFNdDFxAaZvZv12__lambda2029MFxAaZv ()
#2  0x0000000100078e75 in _D3std3net4curl4Curl22_receiveHeaderCallbackUxPammPvZm ()
#3  0x00007fff98ed6eb9 in Curl_client_write ()
#4  0x00007fff98ed616f in Curl_http_readwrite_headers ()
#5  0x00007fff98eebe11 in Curl_readwrite ()
#6  0x00007fff98eed709 in Curl_do_perform ()
#7  0x0000000100078a3b in D3std3net4curl4Curl7performMFbZi ()
#8  0x00000001000770f8 in D3std3net4curl4HTTP8_performMFbZi ()
#9  0x0000000100076e9a in D3std3net4curl4HTTP7performMFZv ()
#10 function go(ref HTTP)
#11 function main()


Can someone confirm that either the static opCall method that HTTP uses for initialization is bad and should be changed, or that this is a compiler issue?