[Xcb] Size of errors

[Uli Schlachter]

Hi Alexander,

On 04.05.2014 15:25, Alexander Mezin wrote:
> Hi there.
> I have few questions regarding xcb error handling, and can't find
> answers in documentation.
> 
> Considering the following code:
> 
> xcb_generic_error_t *e = xcb_request_check(connection, cookie);
> if (e)
>     xcb_generic_error_t v = *e;
> 
> Is it safe?

Yes.

> Do all xcb_*_error_t structs have size <= sizeof(xcb_generic_error_t) ?

Not only "<=". They all have the exact same size.

At least they should, the first other error that I found,
xcb_xfixes_bad_region_error_t, has a different size...

Packets are read from the X server in src/xcb_in.c, function read_packet(). This
function first gets 32 bytes from the FD which contains all the needed
information to figure out more.

All replies (response_type == XCB_REPLY == 1) have their length field in byte
position 4 to 7.

All GenericEvents (see the "XGE extension", response_type == XCB_XGE_EVENT ==
35) have a length field at the same position.

All other packets are just 32 bytes in length.

All the structs that XCB defines are "just" the on-wire representation of
events. However, there is one exception: Replies additionally get a
full_sequence field inserted which doesn't exist "on the wire".

> Does libxcb always allocate a buffer with size >= sizeof(xcb_generic_error_t) ?

Same function (simplified):

    bufsize = length + eventlength + nfd * sizeof(int)  +
        (genrep.response_type == XCB_REPLY ? 0 : sizeof(uint32_t));
    buf = malloc((size_t) bufsize);

Where length is initialized to 32 at the beginning of the function and only ever
increased later.

> May I rely on this behavior in future?

Sure

> Also, do the same rules apply to checked calls with return value
> (xcb_something_reply)?

Yup, see above. Same code handles these.

Cheers,
Uli

P.S.: How's progress?
-- 
<alanc> I think someone had a Xprint version of glxgears at one point,
    but benchmarking how many GL pages you can print per second
    was deemed too silly to merge