Re: [Mingw-users] Problem when building newlib-1.16.0, could be a gnu Make problem?

[Cesar S. <ces...@gm...>]

Hi, Keith

Keith Marshall wrote:
> On Tuesday 27 May 2008 16:21, Song, Shan (GE Infra, Transportation) 
> wrote:
>> Obviously, we should run arm-elf-gcc <option stuff ... > crt0.S
>> It's not crt0.s.
>> gcc treat *.s and *.S differently.
> 
> To be strictly accurate, it's `gas' which treats `*.s' and `*.S' 
> differently, and IMO that is a bug in `gas'; it is simply unacceptable 
> for it to rely exclusively on any such distinction, if it expects to 
> target a case-insensitive file system, such as that of Win32.

I don't think gas/binutils is the culprit here. It is only when you use
the gcc front-end to the assembler that *.S is treated differently.

>From the gcc 4.3.0 manual:
http://gcc.gnu.org/onlinedocs/gcc-4.3.0/gcc/Overall-Options.html

  For any given input file, the file name suffix determines
  what kind of compilation is done:
  [...]
  file.s
      Assembler code.
  file.S
  file.sx
      Assembler code which must be preprocessed.

> Exactly the point.  Just as `gcc' provides alternative methods to `*.c' 
> vs. `*.C' to discriminate between `C' and `C++', `gas' needs at least 
> one alternative to `*.s' vs. `*.S', to support use on case-insensitive 
> file systems. [...]

As you see from above, this is already fixed in gcc 4.3.0, by using the
.sx extension.

Regards,
Cesar