Re: [PATCH] Linker fix for data-only sections

[Paul Brook <paul at codesourcery dot com>]

> But for arm-elf binaries, or arm-wince, or arm-pe, the EABI is not being
> used and in these cases it would make intuitive sense that sections
> without mapping symbols but with a contains-executable-code attribute of
> some kind should be treated as if they contained code.

I don't but that argument. AFAIK arm-pe doesn't have mapping symbols at all - 
it certainly shouldn't be effected by anything in elf32-arm.c.

Even for arm-elf I find your argument somewhat sketchy. If you don't have 
correct mapping symbols on you input objects then I find it hard to believe 
you care about mapping symbols in the result.

IMO assuming data is the only sane option. "Contains code" isn't a useful 
statement on ARM. You need to know the exact combination of ARM, Thumb and 
inline Data, which is why mapping symbols exist in the first place.

If you really want to try and do something sensible with legacy objects 
lacking mapping symbols you'd probably want to base it on function symbol 
types. Even then I wouldn't expect to get useful results in anything other 
than trivial cases.

Paul