This is the mail archive of the mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: gnome 2.8.0 and external dependencies

Hello Charles,

Am Sonntag, 3. Oktober 2004 um 06:49 schriebst du:

> Gerrit P. Haase wrote:

> Well, you have a struct that is const, so it "never changes" and new gcc
> puts the whole struct into .rdata

 >> static const
 >> struct poptOption cl_libIDL_callback_options[] = {
 >> [snip]

> But the important part is that INSIDE the struct, one of the 
> "initialized fields" contains the address of a variable.  In this case,
> although you didn't show it, it's the address of the variable into which
> popt is supposed to store the argument for one of the command line 
> options.  E.g. (going from memory)
>     struct {
>       option = "--bob"
>       short_option = "-b"
>       type = POPT_INTEGER
>       value = &my_variable   <<<<---- right here
>       helpstr = "blah blah blah"
>    };
> So, what's supposed to happen (from popt's perspective) is that the user
> types "--bob 12" and popt will parse the "12" and store it into my_variable.

> However, my_variable is a DATA export from your DLL (it has to be an
> export, so that popt can "see" it).

> But the address of DATA exports from DLLs is unknown at link time.  The
> Windows Runtime Loader has special code (coupled with 
> declspec(dllexport)/declspec(dllimport)) to fixup this address at 
> runtime.  To avoid declspec hell, on cygwin and mingw, the platform,
> compiler and linker cooperate to do something similar, fixing up the
> address at runtime to be a function pointer to a trampoline, which 
> eventually "makes it all work".

> But it is necessary, in both "normal" windows and cygwin's 
> pseudo-runtime-reloc, to UPDATE the address stored in the struct for
> &my_variable.

> But the struct is in .rdata and its contents cannot be changed at runtime.

> Better?

Yes, cool.  Now I think I can distinguish between 'wrong' and valid
const struct definitions.  May I consider using 'const' for structs
which are not really constant (i.e. containig a variable) as harmful?


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]