Re: gettext patches for cygwin #4: accessing fields of exported structs/arrays

[Charles Wilson]

Charles Wilson wrote:

> Plus, if you're going to modify how gcc figures out which section to put 
> variables into, you might as well go all the way: marking .text writable 
> was a shortcut.  Why not put all variables holding values that must be 
> updated by relocation machinery, and are currently placed in .text, into 
> .data instead?  Ditto vars in .rdata meeting that specification?  Then 
> both .text and .rdata can stay non-writable.

<begin state-the-obvious>

D'oh.  Major thinko.  .text === program text.  That is, code, not 
variables.  The auto-import support infrastructure needs to update the 
actual code *at loadtime*.  So every program gets its own copy of the 
DLL's code.

</end>

Yeah, that's a bummer -- there are thus *NO* memory advantages to DLLs. 
 The only advantages you get are (1) synchronization, and (2) dynamic 
updates: (1) every app will(*) use the currently installed version of 
foo.dll, not whatever version of foo.a it was compiled with.  (2) each 
up will automatically pick up the newly-installed DLL, possibly 
benefiting from bugfixes, etc.

(*) assuming you haven't done something stupid with $PATH and multiple 
DLLs...

It'd still be nice if gcc was smarter regarding the .rdata/const 
problem, tho: e.g. "if 'const var foo' contains reference to data 
imported from DLL, put in .data not .rdata"

--
Chuck