Issue 14147 - Compiler crash on identical functions in a single module #7577
Conversation
| @@ -120,6 +120,7 @@ static foreach(member;__traits(allMembers,S)){ | |||
| } | |||
|
|
|||
| // print prime numbers using overload sets as state variables. | |||
| /+ | |||
| static assert(is(typeof(bad57))); | |||
| static assert(!is(typeof(bad53))); | |||
There was a problem hiding this comment.
I'm not sure how I can "fix" this test. It generated bad57 a couple of times and AFAICT there's no break for static foreach yet.
CC @tgehr
wilzbach
force-pushed
the
pr-4396
branch
3 times, most recently
from
bf511f9
to
1045399
Compare
wilzbach
force-pushed
the
pr-4396
branch
2 times, most recently
from
93b6b04
to
44969df
Compare
If two extern (C) functions are just declared with different signatures, they don't conflict. extern(C): alias sigfn_t = void function(int); alias sigfn_t2 = void function(int) nothrow @nogc; sigfn_t bsd_signal(int sig, sigfn_t func); sigfn_t2 bsd_signal(int sig, sigfn_t2 func) nothrow @nogc; // no error This behavior must be reconsidered in the future.
|
So we are all good here? Anything blocking this? |
I'm just waiting for the tests to pass before giving this one last review |
FYI the Jenkins failure just means that the GCE has terminated the runner - it has nothing to do with this PR: dlang/ci uses cheap preemptible GCE instances: https://cloud.google.com/compute/docs/instances/preemptible
|
|
This broke a subtler usage of |
|
Can we revert this, due to issue 18385? |
but yeah, it looks like this needs a deprecation cycle, so we should probably revert it for now and re-add it with deprecations. |
I don't see why it has to be deprecated. The actual code that hit the regression have different mangled names: class Foo
{
extern (C) static void callback(int) {}
extern (C) static void callback(char*) {}
}In the above example the mangled names of |
|
Well, let's go with #7969 for now. Once, that's in we can can still decide what is going to be deprecated with 2.080 and what not (the test cases of #7969 are definitely sth. that should be deprecated).
AFAICT the check already mangles the function signature and checks for that, but as said: "revert first, talk (in-depth) later". Sounds fair? |
Temporarily disable #7577, s.t. a proper deprecation cycle can be started
|
Isn't the regression caused by your special handling of extern(C) symbols? Therefore that special handling should just be removed outright. GDC has thrown errors about duplicate functions for a long time now, and it's surprising to see a couple of changes to the testsuite here that don't really cause symbol conflicts ( |
Yeah, I think so. IIRC, Walter has also, in the |
Revived #4396