Hi All, First of all let me explain our situation. On the m68k port of gcc, libgcc_s.a is used to provide floating point math functions. We also have software math functions in our crt (stdc.library), however they both use their own idea of what state the softfloat support is in (e.g. currently set exceptions). We would therefore like to include an object file when those functions are used in a binary, to setup the glue between gcc's softfloat support (by installing a suitable trap handler), and the crt's math functions. We could just add the file to the STARTFILE_SPEC, of course, however it would always be used in binaries then even if they do not use floating point math - which is undesirable. While looking into this, ive noticed the STARTFILE_SPEC can call a function and have it return an addition to the SPEC. This seems like it is the obvious route to implementing what we need, but a few things still aren't clear, and I was wondering if anyone has enough knowledge on the matter to give me some feedback or put me on the correct path. My Questions are -: Is there any existing implementation that uses the ability of the STARTFILE_SPEC to call a function? How/where should this function be provided? We would like to use something like AROS_STDC_crtobjects as the function name, and have it check the symbols in the compiled file to see if any of the float functions from our crt are used - and if so add the necessary object file to the STARTFILE_SPEC for inclusion in the final binary - is this possible? We propose creating our object files (e.g. crt.fenv.o) aswell as a file containing the symbols to match against (e.g. crt.fenv.lst), and using this list in the called function to query the symbols. If there is any match - include the object file. Any help or information would be greatly appreciated, Yours Nick Andrews.
