Re: gettext patches for cygwin #1: rdata

[Charles Wilson]

Bruno Haible wrote:

> True. And I try to choose the workarounds that require the least knowledge
> of the code or are the most automatic.

So, for instance, making both --enable-auto-import 
--enable-pseudo-runtime-reloc the default setting for cygwin ld, so that 
DLLs can be compiled EXACTLY like static libs with NO changes, and NO 
#ifdef this_flag and #define that_macro?

(ignoring, for the moment, the brokenness w.r.t .rdata/const)

> It is this kind of low-level reasoning - and the amount of time it takes
> to understand it - that takes away brain cycles from the programmer. This
> knowledge is essential for the person who will implement the workaround in
> gcc. It's good that you wrote it down once, for reference. But for an
> average application programmer, this excursion to assembly-language and
> compiler internals drags his attention away from the actual application's
> purpose. That's what I mean by "valid ANSI C code": everything below this
> level is the business of the system implementor, not of the application
> programmer.

Platform note: don't put the address of variables loaded from an 
external library into const storage.

The app programmer doesn't need to know why.  Just what (not) to do. 
It's OBVIOUSLY not ideal.  But it's not THAT bad.  Try coding for 
embedded systems sometime ("whaddaya mean I can't use printf?")

> Are you sure this would work without the DLL_VARIABLE declaration? 

You are right, as I discovered and detailed in the later email with the 
code smippets.

--
Chuck