doc,langref: mention diffs of Zig and C packed structs

[pgy]

I've been bitten by this nuance in the wild while using std.mem.sliceAsBytes on an array of packed structs and felt like it deserved a paragraph in the docs. Feel free to reject this PR without explanation if you disagree -- I'm not very familiar with the Zig project.

Closest issue I could find was #12960

[rohlem]

I agree this is a gotcha; two little nitpicks:

• Instead of "the storage size of the struct in memory" I'd write "the storage size of the struct in a non-packed memory location".
  Specifically as a field of another packed struct, it does exactly bound the size as one would expect.
  (If this sounds too alien, I guess we could also add ", unless as a field of another packed struct.
• The proposed wording if the backing integer is not a power of two is an unnecessary oversimplification in my eyes.
  From what I understand, the compiler is always able to pad and over-align any type it desires, unless specifically requested/guaranteed otherwise via the language.
  It might be better to mention directly (f.e. as inline comment in the example) that it's up to the compiler (+ compilation options, target etc.) what @sizeOf and @alignOf for any given type yield,
  so the correct solution is to always document any assumptions your code makes using comptime assertions (or tests).

[pgy]

I think the confusion arises because "packed" means something very different in the C world. In C, a packed struct means that the size of the struct "is the sum of the lengths of its fields", while in Zig (as far as I understand it) a packed struct is really syntax sugar for bit-packing into a backing integer or another packed struct.

For example CPacked in C has sizeof 13 and alignof 1, while ZigPacked in Zig has @sizeOf 16 and @alignOf 8 in an "unpacked context" and can be packed into 104 bits in a "packed context" (in which case there is no point in talking about byte size or byte alignment). These are very different things, so I think the best would be to mention this difference in the docs.

struct __attribute__((packed)) CPacked {
    double x;
    float y;
    char q;
};

const ZigPacked = packed struct {
    x: f64,
    y: f32,
    q: u8,
};

[pgy]

I changed the PR to document what I described in #21413 (comment)

[alexrp]

Related: #12852

[pgy]

I am closing this PR as there are already many other issues about clarifying "packed" semantics and docs and I do not want to add to the noise. Thanks.

#19660
#19754
#19755