Fix compatibility with zeromq 4.1

[yuyichao]

Fix #83.

The padding size matches what's in the zmq header now. (On 0.4-dev this can be a NTuple{8, Int64}...)

[tkelman]

So did this change between different ZMQ versions? Any compatibility issues with distros that are not as bleeding-edge? I guess Travis will build 3.2.4 from source?

[stevengj]

See also zeromq/libzmq@d9fb1d3 ... do we need to have an explicit Int64 field here to ensure alignment?

[stevengj]

@tkelman, yes, apparently they expanded the size of the struct in a recent version. As long as we use a padding size that is ≥ any size used in any version of ZMQ, we should be fine, because we don't actually access any of that data ourselves.

[yuyichao]

do we need to have an explicit Int64 field here to ensure alignment?

I hoped that a big bitstype would do but I don't mind adding a Int64 at the beginning just to be safe.

Any compatibility issues with distros that are not as bleeding-edge?

Hopefully not. And this is what I want to check with the CI builds.

[yuyichao]

Is zeromq/libzmq@d9fb1d3 released yet? I don't see that on 4.1.2. Should we increase the size of the structure by 1 pointer then?

P.S. I think the GC allocation are aligned now, which should be all what matters? @carnaval

[stevengj]

@yuyichao, that patch does not increase the size of zmq_msg_t — it changes the struct to a union in order to ensure alignment.

If our GC allocation is aligned, that is all that matters.

[yuyichao]

it changes the struct to a union in order to insure alignment.

Ah.... I think this is not the first time I missed sth like this .............. = = ....

[yuyichao]

To the best of my knowledge, a huge bitstype should garentee maximum alignment

https://github.com/JuliaLang/julia/blob/f4283920037399a0ecefa4507adbc6eb9a25708e/src/alloc.c#L654

[stevengj]

Sound good. Let's merge and tag this once Travis is green.

[yuyichao]

I haven't checked the upstream tracker but at least I can confirm that the tests all pass and I have just plotted a curve in IJulia with PyPlot.

(Still find it wierd that I had a working IJulia on earlier master. Did we tighten the allocation recently?....)

[yuyichao]

Should we tag a new release?

[stevengj]

Yes, I'll tag it.

[yuyichao]

THX.

[stevengj]

Tagged as v0.1.19.