From: Charles Sandmann Subject: Re: object file format change? To: nigel AT algor DOT co DOT uk (Nigel Stephens) Date: Thu, 8 Dec 1994 08:52:44 -0600 (CST) Cc: mat AT ardi DOT com, djgpp AT sun DOT soe DOT clarkson DOT edu, elf AT netcom DOT com > cross-compiler using --target=i386-go32. Linux is a red herring: > there all sorts of different Unix systems (e.g. FreeBSD) with > different native object code formats, why should we choose to be > "compatible" with Linux only? A very good point. So far, the best argument I have heard is the superior functionality (C++ debugging, overlinking, etc). > And when Windoze95/Chicago eventually arrives, how much longer will > vanilla MSDOS survive: Chicago runs 32-bit NT/COFF executables too. DJGPP apps still run in Win 95 DOS windows, actually much faster than NT/COFF images do. DJGPP is still the best for text apps. What is obvious to me is that in a few years non-DPMI systems will be rare. Enhancing our djgpp libc to be long file name aware under Win95 is possible (I have some working prototypes). > Although I think that ELF is a much more elegant format (we use it > ourselves for our tools), it looks like compilers whose primary > emphasis is DOS (and therefore soon Chicago) should standardise on > NT-compatible COFF. It is a golden opportunity that Microsoft picked > an object code format which is easily supported by the GNU tools. > > Products like Toolbuster use GCC to provide a Unix compiler and > run-time environment for NT/Chicago, but you have to pay for them. It > would be great if DJGPP could form the basis of a free alternative. People are also working on GCC for NT; this may turn out to be the entire Win 95 solution or provide a lot of the code necessary to turn it on. You can also run Win 32s images under Win 3.1 with a DLL, so there may be a Win 3.1 solution eventually also. Unfortunately, performance is not what I would like.