> In a nutshell, I found that `cpp t.s > t_.s; as t_.s -o t.o` will
> generate t.o with no symbol:
> 0| $ cpp t.s > t_.s; as t_.s -o t.o
> t_.s: Assembler messages:
> t_.s: Warning: end of file in comment; newline inserted
> 0| $ nm t.o
> 0| $
Can you show the first line of your t_.s?
> then I begin to research, and perhaps I found where the problem comes
> from:
> (/sdcard/input-file.c is binutils-2.37/gas/input-file.c)
>
> --- /sdcard/input-file.c
> +++ input-file.c
> @@ -191,7 +191,7 @@
> else if (c == '\n')
> ungetc ('\n', f_in);
> else
> - ungetc ('#', f_in);
> + ungetc(c, f_in), ungetc ('#', f_in);
> }
> else
> ungetc (c, f_in);
>
> This part seems like where simple preprocess the input text. When there
> is '#' at the beginning of the code with next character isn't 'N' 'A' or
> '\n', I think this character and '#' should be sent to ungetc() (so that
> ftell(f_in) return 0), but there is only ungetc('#', f_in). I modified
> input-file.c, and I successfully solved the problem.
The problem with this patch is that ungetc is only guaranteed to work
for *one* character. Also, since f_in might be a pipe you can't
rewind.
--
Alan Modra
Australia Development Lab, IBM
