2DeskLib won't build with DDE5 - compiler bug?

[gah! idiot typo in the subject line…]

Hi,

As the subject says. Builds just fine with DDE24, however when using DDE25, the “Event” code aborts with the message:

Serious error   expected '{'

The line in question is:

event_pollmask event_mask = 0;

(event_pollmask was earlier defined as a union – what is it expecting, “= { 0 }” ?)

I have created a listing file showing what the compiler “sees” here: http://www.heyrick.co.uk/random/event.txt

If I remove the error message line in the newer listing, the rest of the content is identical to the listing created by the DDE24 compiler.

I can get it to build by removing the “= 0” part of the line so it isn’t a critical issue, however… Why is this happening? Or, more specifically, why is the newer compiler having differently to the previous with the same source?

For reference, compiler options are: cc -c -ffahi -list -depend !Depend -throwback -D_DeskLib_Event -I,C: ^.c.Event -o o.Event

You’re right:

typedef union {
  unsigned int value;
  struct {
    unsigned int bitmask:32;
  } data;
} MyUnion;

MyUnion example = 0;

$ gcc test.c
test.c:8:1: error: invalid initialiser

$ clang test.c
test.c:8:9: error: initializing 'MyUnion' with an expression of incompatible type 'int'
MyUnion example = 0;
        ^         ~
1 error generated.

(as ever, clang’s error messages are more helpful). If you put it in { } then it compiles. If you remove the = 0 part then you aren’t initialising the global variable – with unknown consequences for whatever it gets used for.

To be explicit you can also do:
MyUnion example = { .value=0 };
(this is a C99-ism)
clang doesn’t allow:
MyUnion example = { .value=0, .data=42 };
but gcc 4.6.3 and 4.8.4 do.

If you remove the = 0 part then you aren’t initialising the global variable – with unknown consequences for whatever it gets used for.

Not true – the spec says that objects with static storage duration (i.e. globals, statics) will be zero-initialised by default.

Yes, you’re right – since it’s global it’ll get put in the zero-initialised segment. However I think it’s generally good practice to do explicit initialisation and not rely on the behaviour of the runtime system – means you can read the code with having to understand the scope.