Fix segfaults on error on osx. #141
Conversation
|
Updated with a fix for all thread related issues on Mac. Do we really not have a single threading unittest in phobos that also calls the GC? I mean this is broken: class MyThread : public Thread {
int run() {
new byte[1024*1024];
return 0;
}
}
int main(char[][] args) {
auto t = new MyThread();
t.start();
t.wait();
return 0;
}Cheers Jakob. |
|
Bugzilla entry http://d.puremagic.com/issues/show_bug.cgi?id=6290 |
| @@ -165,7 +165,8 @@ class Thread | |||
| return fp(arg); | |||
| else if (dg) | |||
| return dg(); | |||
| assert(0); | |||
| error("neither delegate nor function pointer supplied to run"); | |||
| assert(0); // Not reached but compilred doesn't notice | |||
There was a problem hiding this comment.
?????? If you want a better message, do
assert(0, "neither delegate nor function pointer supplied to run");
The error shouldn't be necessary. If it is, there's a bug in the compiler that needs to be fixed, and that should be fixed rather than doing this. If assert(0) is broken on Mac OS X there are a lot more places than just here which aren't going to work correctly. So, there isn't much point in doing this.
|
ping? |
Fix a simple typo that lead to segfaults in the GC when spawning another thread.
Simple test case:
import std.thread;
class MyThread : public Thread
{
int run()
{
// Tickle the GC
new byte[1024*1024];
return 0;
}
}
int main(char[][] args)
{
auto t = new MyThread();
t.start();
t.wait();
return 0;
}
Signed-off-by: Jakob Bornecrantz <wallbraker@gmail.com>
Calling assert causes an segfaults on osx. Just do this on
all platforms since this is a more helpfull error message.
Simple test case:
import std.thread;
int main(char[][] args)
{
// User forgets to supply a valid delegate
auto t = new Thread(null);
t.start();
t.wait();
return 0;
}
Signed-off-by: Jakob Bornecrantz <wallbraker@gmail.com>
|
Very minor update to commits, mostly just commit messages. @andralex pong! I'm hoping somebody who actually knows the GC to take a look at the change to it, it works for me. @jmdavis For some reason asserting inside the run() function doesn't work, but this example works: import std.thread;
class Foo
{
int func()
{
assert(false);
}
}
int main(char[][] args)
{
auto f = new Foo();
auto t = new Thread(&f.func);
t.start();
t.wait();
return 0;
}The last patch isn't strictly necessary as it only makes the case in the commit message print a error message and then quite instead of segfaulting. While I prefer the error message I can live without it just as long as I get the GC fix so I can actually use Threads on OSX. Cheers Jakob |
|
Ping, can somebody with any GC/runtime knowledge take a look at this? |
|
Sean? |
|
This is Phobos for D 1.0. I could look at this, but it isn't code I maintain. |
|
Please do, I really hope this can go into 1.070 I hate to wait another for another release before my users can use threads. Cheers Jakob. |
|
For what it's worth, the problem with D 1.0 is almost definitely the same as we saw in D 2.0 where we had to switch from using a hardcoded base stack address to obtaining it via an OS call. |
|
@complexmath Okay, well the patch I supplied fixed the problem I was seeing. I think this is unrelated to the issue you mentioned IIRC I asked on that thread and Walter said that no fix was needed for D1. Cheers Jakob. |
|
Interesting. Well, if this change fixes D1 for you then that's good enough for me. I think it's too late to get this into the upcoming release, but it can be merged immediately after. |
Fix segfaults on error on osx.
Calling assert causes an segfaults on osx. Just do this on
all platforms since this is a more helpfull error message.