Initialise dispatch_queue_attr_concurrent on Linux

[seabaylea]

Three of the tests are failing because the DISPATCH_QUEUE_CONCURRENT structure hasn't been initialised (the value for DISPATCH_QUEUE_CONCURRENT->do_vtable is NULL.

This is occurring because DISPATCH_QUEUE_CONCURRENT is aliased to _dispatch_queue_attrs in xcodeconf/libdispatch.aliases, but aliasing (the use of the -alias_list linker option) is only supported on Darwin.

Up until libdispatch-4xx, DISPATCH_QUEUE_CONCURRENT was initialised explicitly in init.c. Taking that approach for non-Apple platforms certainly solves the issue, but there may be a better approach to doing the equivalent of aliasing on Linux.

[MadCoder]

ELF supports aliases from gcc/clang.

you should use:

#ifndef __APPLE__
struct dispatch_queue_attr_s _dispatch_queue_attr_concurrent
        __attribute__((__alias__("_dispatch_queue_attrs")));
#endif

that should work just fine.

[MadCoder]

you may need extern on top, like this:

#ifndef __APPLE__
extern struct dispatch_queue_attr_s _dispatch_queue_attr_concurrent
        __attribute__((__alias__("_dispatch_queue_attrs")));
#endif

I'll let you play with it until you find the right combination that works, glibc does this everywhere in its source

[seabaylea]

Thanks - I'll do some experimenting and then update the PR.

[MadCoder]

I'll ask you to submit a new one so that we get a single patch once you get the right one though, but let leaev that one open until you submit the new one so that nobody tries to fix that concurrently ;)

[seabaylea]

I can just squash the commits under this PR if that helps?

[MadCoder]

I don't think github lets you do that :) but if it allows for it, sure go ahead!

[seabaylea]

Looks like your suggestion was all that was needed. I've built/tested and updated the commit in the PR (by squashing the commits and doing a force push to the branch).