Re: binutils is broken for Linux/sh

[Klee Dienes <klee at apple dot com>]

There's two ways to fix this, pick your favorite:

Choice a) is to do the standard "make a separate table for the 
read/write portion and use that" I did for the other assemblers and 
disassemblers.

Choice b) is to remove the requirement that the "name" portion of the 
table have the same address if the strings are equal.  This invariant 
isn't used in tc-mcore.c; in tc-sh.c, it's used as an optimization in 
the opcode search to speed up the check for "have we switched to an 
opcode with a different name".

I propose that we go with b), and just change the one compare to a 
compare followed by a streq.  In the worst-case, this should be a 
relatively insignificant inefficiency.  But my guess is that the 
worst-case would never happen, as all the strings are constant strings 
declared in the same header, and should end up assigned to the same 
address anyway.

If we go with choice b), here's the patch:

Attachment: more-const.txt
Description: Text document

Both targets (mcore-elf, sh-elf) are tested and pass the test-suite 
clean.

On Thursday, November 21, 2002, at 02:35 AM, Alan Modra wrote:

> On Wed, Nov 20, 2002 at 03:42:31PM -0800, H. J. Lu wrote:
> > I got so many
> [snip errors]
>
> Yeah, the assembler is bombing at tc-sh.c:893 due to the opcode table
> constification.
>
> --
> Alan Modra
> IBM OzLabs - Linux Technology Centre