I have some code written in C, and there is one section that refuses to cooperate when using Visual Studio 2015 Community (clang has no problems). I have a simple struct:
/** Options for enumerating over all documents. */
typedef struct {
unsigned skip; /**< The number of initial results to skip. */
C4EnumeratorFlags flags; /**< Option flags */
} C4EnumeratorOptions;
enum {
kC4Descending = 0x01, /**< If true, iteration goes by descending document IDs. */
kC4InclusiveStart = 0x02, /**< If false, iteration starts just _after_ startDocID. */
kC4InclusiveEnd = 0x04, /**< If false, iteration stops just _before_ endDocID. */
kC4IncludeDeleted = 0x08, /**< If true, include deleted documents. */
kC4IncludeNonConflicted = 0x10, /**< If false, include _only_ documents in conflict. */
kC4IncludeBodies = 0x20 /**< If false, document bodies will not be preloaded, just
metadata (docID, revID, sequence, flags.) This is faster if you
don't need to access the revision tree or revision bodies. You
can still access all the data of the document, but it will
trigger loading the document body from the database. */
};
typedef uint16_t C4EnumeratorFlags;
And I also have a constant "default" value for it:
// In header
extern const C4EnumeratorOptions kC4DefaultEnumeratorOptions;
// In implementation
const C4EnumeratorOptions kC4DefaultEnumeratorOptions = {
0, // skip
kC4InclusiveStart | kC4InclusiveEnd | kC4IncludeNonConflicted | kC4IncludeBodies
};
However, when debugging I noticed that the initialization is not doing anything when I try to use the default value:
// options winds up with a "skip" value of something like 117939945
// and a flags value of 59648
C4EnumeratorOptions options = kC4DefaultEnumeratorOptions;
The section defining is in a DLL, and the second using is in an exe. Again, this only happens on Windows. Furthermore, the value in "options" is garbage but for some reason it's not even the same garbage that is stored in kC4DefaultEnumeratorOptions
. I know MSVC is notorious for snubbing C, but this kind of initialization is so old that even MSVC should get it right, shouldn't it? So it must be something I am doing but I can't figure out what.
EDIT The symbol is being exported via a export definitions file. I checked with dumpbin, and found the symbol in the exported symbols list
41 46 00A6EA8 kC4DefaultEnumeratorOptions = kC4DefaultEnumeratorOptions
Also as one more bit of info, the calling code is C++ and the DLL code is C, which I suspect may be playing a part in this madness.