From mboxrd@z Thu Jan  1 00:00:00 1970
From: Nick Clifton <nickc@cygnus.com>
To: snowball3@bigfoot.com
Cc: binutils@sourceware.cygnus.com
Subject: Re: Minor fix to COFF .section change
Date: Mon, 03 Jul 2000 10:42:00 -0000
Message-id: <200007031742.KAA27100@elmo.cygnus.com>
X-SW-Source: 2000-07/msg00026.html

Hi Mark,

: There seems to be another problem in my opinion as a consequence of
: the current default.
: 
: .section .foo,"w"
: .section .foo2,"d"
: 
: produces:
: 
:   3 .foo          00000000  00000000  00000000  00000000  2**2
:                   ALLOC, LOAD, CODE
:   4 .foo2         00000000  00000000  00000000  00000000  2**2
:                   ALLOC, LOAD, DATA
: 
: The !BFD version handles 'w' and 'd' the same way, I think the BFD version 
: should too. Thoughts?

Hmm, well I agree that .foo ought to be a DATA section not a CODE
section, so that in this case they ought to end up with the same
attributes.

Looking at the source in obj-coff.c it seems that "d" type sections
explicitly have SEC_DATA and SEC_LOAD set, whereas "w" sections do
not.  I am not sure why this should be, although I suspect it is
because the m88k does not default to setting these attributes for its
sections.

I assume that your proposed patch fixes this problem as well ?

: > I'm concerned that there may be software out there that may 
: > create a section, expecting it to be executable (the old way) but 
: > finding it's not (the new way).
: 
: I do believe the proposed new default matches the documentation, but
; I see your point. I looked through the GCC config files that define 
: ASM_OUTPUT_SECTION, but it wasn't much help because they all use
: pretty much the same in always adding the attribute so there's never
: a doubt about what's intended. DJGPP should do the same in gcc 3.0.  
: 
: Anyway, as I see it there are several ways to go:
: 
: 1. Go ahead and change the default attribute to match the docs.
: 
: 2. Go with #1, but if it turns out to be a problem for a target, let
:    it default to code.
: 2a. Let a target choose between defaulting to code or to data.
: 
: 3. Recognize .eh_frame (as the .section doc allows) and set SEC_DATA
: for it. 
: 
: My preference is for door #2. Door #3 goes in the direction of 
: automatically setting the right attributes for every special section
: there is.

I like 2/2a too.  Can you try generating a new version of the patch
that creates a new target macro (eg TC_COFF_DEFAULT_SECTION_ATTRIBUTES) 
which if not defined defaults to SEC_LOAD | SEC_DATA, and which then
uses this macro in obj_coff_section().  If you can also generate a
paragraph to go into doc/internals.texi describing this new macro,
that would be great.

Cheers
	Nick