Bug 6308 – Destruction of temporaries on exception causes unhandled access violation

My reply about this issue on dmd-beta is pending moderation, but a bug about this should probably be filed anyway. import std.stdio; void main() { foreach (i; 0..100) foreach (line; File("test.d").byLine) {} } On Windows, when the above program is compiled with DMD from current git and ran, it crashes with an unhandled exception (Windows displays two standard "has stopped working" dialogs in succession).

More by Vladimir: I did a bit of research on this a few days ago. Using a certain debugger and the map file, I found that the crash occurred in the autogenerated scope(exit) for getDependencies - the block of code that calls the File destructor. I didn't look too closely, but it looked like a bad "this" pointer is passed to the File dtor. I tried a full DMD/Phobos regression test, but didn't really get anywhere.

Minimized testcase: struct C { this(int) { throw new Exception("Oops!"); } ~this() { } int bar() { return 1; } } void foo(int) {} void main() { foo(C(1).bar()); } The key finding is that the exception is thrown in the constructor.

*** Issue 6329 has been marked as a duplicate of this issue. ***

Issue 6329 is an example where the the exception isn't thrown in the constructor, but in the body of a delegate (dirEntries uses opApply). I'll try to minimize it.

(In reply to comment #4) > I'll try to minimize it. Reduced it to a small variation of the above. struct C { void oops() { throw new Exception("Oops!"); } ~this() { } } void main() { C().oops(); } I think DustMite may have taken some shortcuts while reducing Andrej's test case, though - but this test case definitely does show that it's not about throwing in a constructor.

By the way, I should probably clearly mention that this is Windows-only.

I had a quick look at the disassembly generated for the code in comment 5, and it turns out that the frame handler pointer is zero, instead of identical code "C c; c.oops();" that generates something like mov edx,dword ptr fs:[0] push offset __Dmain+77h // this is 0 for "C().oops();" push edx mov dword ptr fs:[0],esp

dmd 2.053 does not crash, but also does not create any exception handling code at all, while dmd 2.054+ seems to generate setup code for the exception frame, but no exception code.

*** Issue 6363 has been marked as a duplicate of this issue. ***

This code: import std.file; void main() { string[] dirs; foreach (string dir; dirEntries(r"C:\", SpanMode.shallow)) { if (dir.isDir) dirs ~= dir; } } Compiled via DMD 2.055 commit cfab198ee186f6e69c364aaf4206434220d83204 (Aug 15), enters an infinite loop at runtime where exceptions are kept throwing: === Bypassed === std.file.FileException@std\file.d(1156): C:\pagefile.sys: The process cannot access the file because it is being used by another process. ---------------- 42A540 42A3B7 402A7D 40206D 402950 40298F 40258B 45B855 45B7AC ---------------- object.Error: Access Violation ---------------- 42A540 42A3B7 402950 40298F 40258B 45B855 45B7AC ---------------- Bypasses std.file.FileException@std\file.d(1156) object.Error: Access Violation ---------------- 42A540 42A3B7 402950 40298F 40258B 45B855 45B7AC ---------------- object.Error: Access Violation ---------------- 42A540 42A3B7 402950 40298F 40258B 45B855 45B7AC It seems after it tries to open the protected pagefile.sys it keeps rethrowing exceptions and not exiting the foreach loop. It doesn't stop at all, just keeps rethrowing.

(In reply to comment #11) > It seems after it tries to open the protected pagefile.sys it keeps rethrowing > exceptions and not exiting the foreach loop. It doesn't stop at all, just keeps > rethrowing. I can't reproduce this, it throws a single exception for me. Was this fixed in the meantime? The fact that you can't list C:\ because Phobos can't read a file's attributes is worth a separate bug report, though.

(In reply to comment #12) > (In reply to comment #11) > > It seems after it tries to open the protected pagefile.sys it keeps rethrowing > > exceptions and not exiting the foreach loop. It doesn't stop at all, just keeps > > rethrowing. > > I can't reproduce this, it throws a single exception for me. Was this fixed in > the meantime? The final 2.055 release still has this issue, just checked.

(In reply to comment #13) > (In reply to comment #12) > > (In reply to comment #11) > > > It seems after it tries to open the protected pagefile.sys it keeps rethrowing > > > exceptions and not exiting the foreach loop. It doesn't stop at all, just keeps > > > rethrowing. > > > > I can't reproduce this, it throws a single exception for me. Was this fixed in > > the meantime? > > The final 2.055 release still has this issue, just checked. Note that my bug6329 which was labeled as duplicate of this has a similar issue. First I get a range violation, then `object.Error: Access Violation` is printed many times before the app exits.