Implement fmemopen

[no92]

Supersedes #540.
Fixes #527.

[no92]

The failing assert will not be done on glibc due to (incorrectly) failing there. NetBSD's libc tests also guard against this in the same way. My implementation is adhering to POSIX, and thus correct; musl and the BSDs agree with me here. This should be ready for review now.

[ikbenlike]

LGTM

[Geertiebear]

Haven't taken a good look at this yet, will do later, but just a nitpick already when skimming it over: the comments aren't in the same style :) In general we follow that comments should // Look like this., if you could fix that in the relevant places that would be really nice.

[no92]

Fixed the comments.

[Geertiebear]

In general, I'm not a fan of this code. It seems that the _allowResizing flag was used to cram two different classes into one implementation, which makes for a very confusing read of the code. In this case it would be appropriate to have two different classes, and possibly a common superclass that abstracts some of the common behaviour away. Right now (what should be) virtual calls are emulated using the _allowResizing flag.

In summary, please add missing test cases, since there were several untested cases I've noticed. And separate the two different implementations using inheritance, this will clean up the implementation and hopefully reduce the likelihood of bugs.

[Geertiebear]

I feel like io_seek() and io_read() can be fully abstracted by just making _buffer_size() a protected virtual function in the base class. io_write is unfortunately a little bit trickier it seems since there is a special case for fmemopen().

[Dennisbonke]

LGTM