Bitfields cause major issues on Windows about cbforest HOT 8 CLOSED

This is weird. It’s true that a bool can hold values other than 0 or 1, but in general it shouldn’t, as it only gets initialized by the constants true or false, or by a logical operation like &&.

The values you’re seeing make me suspect something’s wrong. Where did that c4options struct come from? I suspect it may have been allocated/declared without getting initialized, so the fields contain garbage…

from cbforest.

The portion of the test I noticed this in was here:

C4AllDocsOptions options = kC4DefaultAllDocsOptions;
options.includeBodies = false;
e = c4db_enumerateAllDocs(db, kC4SliceNull, kC4SliceNull, &options, &error);

from cbforest.

In that case kC4DefaultAllDocsOptions must not be initialized properly … the bools in there should all be 1 or 0. Is MSVC leaving the fields unmentioned in the initializer as garbage?

from cbforest.

I have it setup with the C position-based struct initializer (see https://github.com/couchbaselabs/cbforest/blob/feature/c_api/C/c4Database.cc#L611-L618). I don't quite know what is going on with that. I could try using (bool)1 I guess. I'm almost sure it is some weird compiler optimization because the C# bindings don't have this problem (You can't bind to a variable in C# so the defaults, etc, are all redefined in C#)

from cbforest.

Could you set a breakpoint and examine the contents of the struct to see if they're garbage? They shouldn't be, from what I know of the C standard, but it's possible this is a quirk of MSVC…

from cbforest.

I already did that, it's how I found the values I noted in the original ticket. The contents of the struct are weird. The variable examiner will show things like "true (233)" I wonder which standard is being applied here, because bool is a C99 thing and we all know how MSVC fares with C99. However, it is also a C++ thing. Everything I look at says "this shouldn't happen" which frustrates me.

from cbforest.

This shouldn't happen. :/ It's in C++ code so there's no problem with bool, and you're explicitly initializing every field of the struct to a valid value. If the fields don't have those values at runtime, it seems like the compiler screwed up.

from cbforest.

I've found another clue in this matter. It seems like ANY global const struct getting exported for use in CppTests is getting completely blown away for some reason with Visual Studio. Everything inside of it is garbage. This garbage just happened to be closer to the truth than the garbage I am dealing with now.

from cbforest.