DCD crashes on startup in std.experimental.logger

[Hackerpilot]

Using LDC built from commit 6e0bada

Stack trace:

#0  0x0000000000598058 in rt.monitor_.ensureMonitor(Object) ()
#1  0x0000000000598126 in _d_monitorenter ()
#2  0x00000000004665e3 in std.experimental.logger.core.Logger.memLogFunctions!(64).logImpl!(160, "src/server/server.d", "server.server.main", "int server.server.main(string[] args)", "server.server", immutable(char)[], immutable(char)[]).logImpl(lazy immutable(char)[], lazy immutable(char)[]) (this=0x7ffff7fc38e8, _param_0={string delegate() (void)} 0x7fffffffd400, _param_1={string delegate() (void)} 0x7fffffffd3f0) at /usr/local/include/d/std/experimental/logger/core.d:948
#3  0x0000000000459178 in std.experimental.logger.core.defaultLogFunction!(64).defaultLogFunction!(160, "src/server/server.d", "server.server.main", "int server.server.main(string[] args)", "server.server", immutable(char)[], immutable(char)[]).defaultLogFunction(lazy immutable(char)[], lazy immutable(char)[]) (_param_0={string delegate() (void)} 0x7fffffffd468, _param_1={string delegate() (void)} 0x7fffffffd458) at /usr/local/include/d/std/experimental/logger/core.d:487
#4  0x000000000045723d in D main (args=...) at server.d:160

[Hackerpilot]

This may related to #1377.

[JohanEngelen]

@Hackerpilot Is it fixed by this?

--- a/gen/typinf.cpp
+++ b/gen/typinf.cpp
@@ -726,8 +726,11 @@ void TypeInfoDeclaration_codegen(TypeInfoDeclaration *decl, IRState *p) {
     } else {
       irg->type = LLStructType::create(gIR->context(), decl->toPrettyChars());
     }
-    LLGlobalVariable *g = new LLGlobalVariable(gIR->module, irg->type, true,
-                                               lwc.first, nullptr, mangled);
+    // Create the symbol. We need to keep it mutable as the type is not declared
+    // as immutable on the D side, plus synchronized() can be used on the
+    // implicit monitor.
+    auto g = new LLGlobalVariable(gIR->module, irg->type, false, lwc.first,
+                                  nullptr, mangled);
     setLinkage(lwc, g);
     irg->value = g;
   }

Thanks for testing!

[Hackerpilot]

It is. With that patch DCD passes its test suite.

[JohanEngelen]

@Hackerpilot Thanks for the fix confirmation.
Closing, as this is a duplicate of #1377.

[Hackerpilot]

I made a fresh clone of LDC master (fa8fe51), and tried this again and it's failing in the same place. I checked that CPP file and see that the fix is there, but the crash is back. Any ideas?

[JohanEngelen]

@Hackerpilot What happens when you apply the patch of this PR? #1384

[Hackerpilot]

That patch is against the ltsmaster branch, so I'll have to translate it to D first.

[JohanEngelen]

@Hackerpilot Sorry didn't notice, working on new PR here: #1385

[Hackerpilot]

For some reason that didn't break the build.

The line of ASM causing the segfault is

mov    rax,QWORD PTR [rax]

gdb tells me that the rax register's value at this point is 0.

Stack trace:

#0  0x00000000005a2a28 in rt.monitor_.ensureMonitor(Object) ()
#1  0x00000000005a2af6 in _d_monitorenter ()
#2  0x000000000046a793 in std.experimental.logger.core.Logger.memLogFunctions!(64).logImpl!(160, "src/server/server.d", "server.server.main", "int server.server.main(string[] args)", "server.server", immutable(char)[], immutable(char)[]).logImpl(lazy immutable(char)[], lazy immutable(char)[]) (this=0x7ffff7fc38e0, _param_0={string delegate() (void)} 0x7fffffffd410, _param_1={string delegate() (void)} 0x7fffffffd400) at /usr/local/include/d/std/experimental/logger/core.d:948
#3  0x000000000045d258 in std.experimental.logger.core.defaultLogFunction!(64).defaultLogFunction!(160, "src/server/server.d", "server.server.main", "int server.server.main(string[] args)", "server.server", immutable(char)[], immutable(char)[]).defaultLogFunction(lazy immutable(char)[], lazy immutable(char)[]) (_param_0={string delegate() (void)} 0x7fffffffd478, _param_1={string delegate() (void)} 0x7fffffffd468) at /usr/local/include/d/std/experimental/logger/core.d:487
#4  0x000000000045b33d in D main (args=...) at server.d:160

[JohanEngelen]

Could you dustmite it?
(the crash is exactly what I would expect from an optimization bug with synchronize on an immutable object)

[Hackerpilot]

I'm working on figuring out how to write the test script to catch that specific segfault.

Edit: https://github.com/CyberShadow/DustMite/wiki/Detecting-a-specific-segfault

[Hackerpilot]

I'm no longer able to reproduce this issue with LDC at version 98c899b

[dnadlinger]

Yep, we should be emitting TypeInfo as mutable for now.

[yotamr]

Seems to happen again on LDC 1.3.0:

#0 0x00007fbcfb0caa99 in rt.monitor_.ensureMonitor(Object) () from ./libagent.so #1 0x00007fbcfb0cab66 in _d_monitorenter () from ./libagent.so #2 0x00007fbcfae048ce in std.experimental.logger.core.Logger.log!(immutable(char)[]).log(const(std.experimental.logger.core.LogLevel), lazy immutable(char)[], int, immutable(char)[], immutable(char)[], immutable(char)[], immutable(char)[]) (this=0x7fbcd4004258, ll=const(LogLevel).all, args=..., line=179, file=..., funcName=..., prettyFuncName=..., moduleName=...) at /opt/ldc2/bin/../import/std/experimental/logger/core.d:1259 #3 0x00007fbcfae031d9 in std.experimental.logger.core.log!(immutable(char)[]).log(lazy immutable(char)[], int, immutable(char)[], immutable(char)[], immutable(char)[], immutable(char)[]) (arg=..., line=179, file=..., funcName=..., prettyFuncName=..., moduleName=...) at /opt/ldc2/bin/../import/std/experimental/logger/core.d:353 #4 0x00007fbcfae31613 in service.restapi.getRules() () at restapi.d:179

[JohanEngelen]

@yotamr How about LDC 1.4.0 ?

[yotamr]

Just tested against LDC 1.4.0. Still crashes.

…

On 16 September 2017 at 04:06, Johan Engelen ***@***.***> wrote: @yotamr <https://github.com/yotamr> How about LDC 1.4.0 ? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1358 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAmBdZAexN6cJf7dffVAx0-L4nH6lFW4ks5si6u6gaJpZM4HzfRo> .

[yotamr]

My code is a shared library loaded by an already-existing process. There are three threads running D code: one belonging to the process that loaded the .so, and two other threads handling HTTP communication. I tried reproducing it in two ways, both did not reproduce the crash:

1. single-threaded app that writes as many logs as possible in a loop
2. The same as Removed Phobos 1 from LDC tree. #1, only with three threads writing in a loop instead of one. All threads share the same logger object.

[dnadlinger]

Could you please open a new issue for the crash you are seeing, including more detailed system information (OS/compiler/linker versions, processor, etc.) and a way to reproduce it?

What you are describing is a different problem than reported here, which was caused by TypeInfos being emitted as constant data.

[yotamr]

Sure, thanks.

…

On 16 September 2017 at 18:07, David Nadlinger ***@***.***> wrote: Could you please open a new issue for the crash you are seeing, including more detailed system information (OS/compiler/linker versions, processor, etc.) and a way to reproduce it? What you are describing is a different problem than reported here, which was caused by TypeInfos being emitted as constant data. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1358 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAmBdfUZ2uPQ05YdMSY21T3Z4XvreU9sks5sjHDogaJpZM4HzfRo> .

[dnadlinger]

Thanks!