Use consistent name for OS on GNU/Linux.

[hobophobe]

Somehow the platform directories and the platform.rb seem to be desynched for GNU/Linux.

I assume this will fix issue #137, though the submitter of that issue didn't indicate which platform.

[terceiro]

The OS is said to be GNU because other OSes with GNU userland (but not Linux) are also supported.

Also, I don't see where that OS constant is used for anything concrete. Can you ellaborate on why this is a problem? Do you have a concrete use case (or unit test) that fails because of this?

[hobophobe]

On my system rbconfig.rb has CONFIG["host_os"] = "linux-gnu". That means platform.rb's OS = case gets set gnu. platform.rb uses that to set CONF_DIR. That means when a consumer tries to resolve a type, the FFI::TypeDefs/types.conf is sought in ffi/platform/x86_64-gnu instead of ffi/platform/x86_64-linux.

The resulting error is the same as that issue linked before, or also Debian Bug #643788.

I see your point about GNU meaning more than Linux, but I don't see type definitions for GNU here. Also, in platform.rb IS_LINUX is defined as equal to IS_GNU = is_os("gnu").

At least the OS as used in the CONF_DIR should depend on the kernel in use rather than GNU?

For example, Debian Sid: libruby1.8: downloads includes versions for Debian using a FreeBSD kernel. The rbconfig.rb distributed with those have host_os of kfreebsd-gnu, which would also look in, say, x86_64-gnu (it would otherwise have correctly matched freebsd). (Note that under the logic mentioned above, it would also mean a FreeBSD system would have true for IS_LINUX.)

Hopefully my updated commit will be clean enough to resolve the issues. Does anything rely on OS being gnu?

[terceiro]

That makes sense - now we only need a concrete failing spec. The reason why this was not caught before is that there is not a single spec for this.

(BTW I am the Debian maintainer)

[terceiro]

done, I wrote a failing spec. @hobophobe, please merge my pull request on your master branch so that this pull request gets updated with the extra commit I did.

[terceiro]

Just noticed that this changes make 5 specs fails on debian kfreebsd-amd64, so you are not quite there yet

[terceiro]

fixed that. @hobophobe, please merge my other pull request

[terceiro]

@wmeissner, could you please consider merging this branch into the official master branch?

[ghost]

I think it looks ok - I haven't run a linux system in ages, so as long as you're happy it works on all debian platforms (and I assume all other linuxen work the same), then I'm ok with merging it.

[ghost]

btw, can you check the equivalent files in JRuby to see if they need changes?

It will have much the same platform detection logic.

[terceiro]

Yes, I guess JRuby needs similar changes:

• it should use LIBC_SO defined in gnu/lib-names.h instead of hardcoding libc.so.6 (not sure how you would do that in Java, though)
• it should actually test for GNU instead of Linux when deciding the file extension for dynamic libraries