no method value(::CLPackedAttr)

[vchuravy]

ERROR: LoadError: MethodError: no method matching value(::CLPackedAttr)
Closest candidates are:
  value(!Matched::CLEnumConstantDecl) at /home/vchuravy/.julia/packages/Clang/NONKM/src/cursor.jl:246
Stacktrace:
 [1] wrap!(::DefaultContext, ::CLEnumDecl) at /home/vchuravy/.julia/packages/Clang/NONKM/src/wrap_c.jl:97
 [2] run(::WrapContext) at /home/vchuravy/.julia/packages/Clang/NONKM/src/compat.jl:122
 [3] top-level scope at none:0

It seems to be just:

typedef enum {
    /* Operation completed successfully */
    UCS_OK                         =   0,

    /* Operation is queued and still in progress */
    UCS_INPROGRESS                 =   1,

    /* Failure codes */
    UCS_ERR_NO_MESSAGE             =  -1,
    UCS_ERR_NO_RESOURCE            =  -2,
    UCS_ERR_IO_ERROR               =  -3,
    UCS_ERR_NO_MEMORY              =  -4,
    UCS_ERR_INVALID_PARAM          =  -5,
    UCS_ERR_UNREACHABLE            =  -6,
    UCS_ERR_INVALID_ADDR           =  -7,
    UCS_ERR_NOT_IMPLEMENTED        =  -8,
    UCS_ERR_MESSAGE_TRUNCATED      =  -9,
    UCS_ERR_NO_PROGRESS            = -10,
    UCS_ERR_BUFFER_TOO_SMALL       = -11,
    UCS_ERR_NO_ELEM                = -12,
    UCS_ERR_SOME_CONNECTS_FAILED   = -13,
    UCS_ERR_NO_DEVICE              = -14,
    UCS_ERR_BUSY                   = -15,
    UCS_ERR_CANCELED               = -16,
    UCS_ERR_SHMEM_SEGMENT          = -17,
    UCS_ERR_ALREADY_EXISTS         = -18,
    UCS_ERR_OUT_OF_RANGE           = -19,
    UCS_ERR_TIMED_OUT              = -20,
    UCS_ERR_EXCEEDS_LIMIT          = -21,
    UCS_ERR_UNSUPPORTED            = -22,
    UCS_ERR_REJECTED               = -23,

    UCS_ERR_FIRST_LINK_FAILURE     = -40,
    UCS_ERR_LAST_LINK_FAILURE      = -59,
    UCS_ERR_FIRST_ENDPOINT_FAILURE = -60,
    UCS_ERR_LAST_ENDPOINT_FAILURE  = -79,
    UCS_ERR_ENDPOINT_TIMEOUT       = -80,

    UCS_ERR_LAST                   = -100
} __attribute__((packed)) ucs_status_t ;

[Gnimuc]

Since __attribute__(xxx) is not supported by Julia, one might need to wrap this kinda structures manually with additional dummy padding fields to ensure the correct alignement. Indeed, wrap! should be able to catch __attribute__ keywords and print warning messages instead of throwing errors.

This is a "packed" enum, no padding is needed, so maybe we could make wrap! work on this special case.
EDIT: I'm not sure whether Julia compiler will add padding between fields in the same way as C or not. But this enum only has integer fields which seems to be compatible with Julia.

[Gnimuc]

It looks like libclang provides a way to get the fieldoffset info:

Clang.jl/gen/libclang_api.jl

Lines 904 to 910 in 289bdd5

	function clang_Type_getOffsetOf(T, S)
	ccall((:clang_Type_getOffsetOf, libclang), Clonglong, (CXType, Cstring), T, S)
	end
	function clang_Cursor_getOffsetOfField(C)
	ccall((:clang_Cursor_getOffsetOfField, libclang), Clonglong, (CXCursor,), C)
	end

maybe we could write a macro/function for constructing structs with dummy padding fields automatically inserted?

[vchuravy]

I think a packed enum means that it will be only using Int8 in this case? But I am not a 100% on all the C-semantics.

In any case Clang.jl should at least ignore it and print a warning.

[ihnorton]

I'm not sure whether Julia compiler will add padding between fields in the same way as C or not. But this enum only has integer fields which seems to be compatible with Julia.

Yes, it has to respect the platform alignment rules (e.g. base reflects a bunch of structs from libraries like libuv that would be very broken otherwise).

maybe we could write a macro/function for constructing structs with dummy padding fields automatically inserted?

The only issue there is that the output will probably be wrong on systems where ptrdiff_t is 32-bit (ARM and x86 Windows; not sure if Julia still even supports the latter). I don't know a good solution to that... Ideally this would be pushed down in to the compiler, but it is enough of an edge-case that everybody has obviously gotten by without it for this long, using https://github.com/Keno/StructIO.jl and StrPack, etc.

I think a packed enum means that it will be only using Int8 in this case? But I am not a 100% on all the C-semantics.

IIUC yes, from a quick read of GCC's page about it -- should pack to the smallest type which can represent the range. That said, this is about the point where I might start thinking about C helper shims to get the as-compiled size, especially if you are going to be using UCX on any "interesting" hardware.