On Tue, 11 Oct 2022 08:37:59 GMT, Aleksey Shipilev <sh...@openjdk.org> wrote:
>>> > > Fix looks good as far as it goes (can't believe I didn't see what was
>>> > > going on here!) - but `os::signal` is still broken as it uses
>>> > > `sa_handler` instead of `sa_sigaction`.
>>> >
>>> >
>>> > Yikes! I think this change should proceed as is, and `os::signal` should
>>> > be looked at as a new issue. That looks messy :(
>>>
>>> There is no rush, because we are waiting for another build system change to
>>> drop.
>>>
>>> Why can't we do the same thing we did for `SR_handler` in `SR_initialize`?
>>>
>>> ```
>>> @@ -864,7 +864,7 @@ void* os::signal(int signal_number, void* handler) {
>>> remove_error_signals_from_set(&(sigAct.sa_mask));
>>>
>>> sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
>>> - sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
>>> + sigAct.sa_sigaction = CAST_TO_FN_PTR(sa_sigaction_t, handler);
>>>
>>> if (sigaction(signal_number, &sigAct, &oldSigAct)) {
>>> // -1 means registration failed
>>> ```
>>>
>>> It matches what we should do for `SIG_INFO` flag, and as Kim said, it is
>>> still likely yields the same code as `sa_handler` and `sa_sigaction` are
>>> probably the same on currently supported systems.
>>
>> For POSIX we expect the handler argument for os::signal to be a sigaction
>> handler (taking 3 arguments). For Windows we expect os::signal to take a 1
>> argument handler, since Windows seemingly doesn't support the sigaction
>> stuff. How is that a portability layer?
>>
>> So all of the calls are going to need their handler functions examined. In
>> at least one use (in os_windows), the UserHandler function takes 3 arguments
>> though only uses one, and is called with only one.
>>
>> Looking at where os::signal is called, it's not clear it needs to be in os,
>> and given the inconsistent expectations it probably shouldn't be. All calls
>> are in posix-specific or windows-specific files. I'd suggest removing
>> os::signal entirely, and having windows code use ::signal and posix code use
>> ::sigaction. Maybe with some convenience wrappers around each, but those
>> wrappers are windows-specific or posix-specific and never the twain shall
>> meet.
>>
>> That all seems to me like it would be better to separate from this change,
>> keeping this one nice and simple.
>
>> For POSIX we expect the handler argument for os::signal to be a sigaction
>> handler (taking 3 arguments). For Windows we expect os::signal to take a 1
>> argument handler, since Windows seemingly doesn't support the sigaction
>> stuff. How is that a portability layer?
>
> D'oh.
>
>> That all seems to me like it would be better to separate from this change,
>> keeping this one nice and simple.
>
> Agreed.
> ...unless @shipilev really wants it? :)
No, I do NOT want it :) I'll be happy to test it, though.
-------------
PR: https://git.openjdk.org/jdk/pull/10494
