Slashdot Mirror
Sybase to Open Souce Watcom C/C++ & Fortran Compiler
Kendall Bennett writes "Rumours have been running around for some time, but now it is official. Sybase has announced that it will be releasing the full source code to the Watcom C/C++ and Fortran compilers under an Open Source license. For more information see the new Open Watcom website." The press release states as well that the license will comply with an OSI approved license.
In theory, we should be able to take some of the smarts from the Watcom compiler and put them in GCC. I bet this won't be very possible in practice simply because of architectural differences in the compilers.
"It may be remarked in passing that success is an ugly thing. Men are deceived by its false resemblences to merit."
I really hope the code is released as GPL (or a GPL-compatible license). That way, the egcs (now gcc) team could incorporate some code (mostly optimizations I guess) in gcc. Also, I *think* the fortran frontend (g77) is missing a lot of features (like pointers). Maybe they'd be able to use Watcom fortran for these features... or even take the code to build a new frontend.
What's clear to me is that few people would be willing to continue developping watcom. I see more future in incorporating watcom stuff into gcc.
Opus: the Swiss army knife of audio codec
Direct from the OpenWatcom FAQ:
Q: What compilers will I need to compile the source code?
A: Initially the Open Watcom 1.0 compiler will require an official copy of the Watcom C/C++ 11.0b compiler, with the 11.0c binary patch release applied in order to successfully compile it. This is due to the need to rely upon proprietry SDK's for platforms such as Windows, and OS/2 that cannot be distributed along with the Open Source 1.0 compiler release. It is planned to eliminate this dependency in the Open Source 2.0 release, by allowing the compiler to work with freely downloadable SDK's for the supported platforms.
So in other words, to compile the 1.0 Open-Source version, you need the commercial 11.0c version. So everyone run out and buy the 11.0c version for ~$200 so you can compile your free Open-Source version. Or you could wait god knows how long for version 2.0 to come available without the 11.0c dependency.
However, in graduate school, I was developing a biology simulation in Matlab. The development went well, but the execution was painfully slow. Since Fortran 90/95 was syntactically similar to Matlab I was able to port his huge simulation in a few days and keep working in Fortran without much of a learning curve. Say what you want, for high performance numerical computing with lots of vector operations, Fortran is one of the best languages out there.
Bringing this back to the open source world, numerically intense vector computations can be developed in Octave, and easily ported to Fortran 90/95. It's a one-two punch that I would definitely find useful. Unfortunately, the only free Fortran compilers I've found are for Fortran 77.
Just my two bits.
. . . but Fortran *is* still changing, sometimes rapidly. No, it doesn't have everything that C does--nor should it. For many types of heavy computation, it is still the language of choice (Including my own research).
The strengths aren't just from what is *in* fortran, but from what is left *out*. Because certain classes of pointers/objects/whatever aren't there, stronger assumptions can be made while optimizing. You can quickly write fast code faster in Fortran than C.
Don't get me wrong--it's not that C *can't* produce numerical code as fast as the corresponding Fortran; it can. However, this happens after hand-tuning and optimizing the C, and it generally reaches roughly parity with the initial Fortran program.
hawk, still using Fortran
This is a very good thing. I see a lot of people either putting down the Watcom compiler or making fun of Watcom for no longer being a major player in compilers. I'm surprised to see such sentiments from what is supposedly the open source community.
Watcom's open sourcing of their C++ compiler is a good thing all around. It has generally been accepted that open source is of the most value for system software, because it serves as a foundation for the work of many other people. It is difficult to bet on a lesser known vendor, when they could go out of business without warning. This has happened to be me several times. Open sourcing compilers, or even just releasing them free of charge, as Borland has done, is The Right Thing. It removes unneeded pressure from choosing development tools.
It is also excellent that gcc is getting more competition. gcc is a good compiler, and it has many boosters in the UNIX world, but it is crusty in manys ways, especially in the Windows environment. gcc has also been peculiarly slow in evolving for the x86 architecture, presumably because of initial personal grievances from FSF members. Watcom throws in a little competition. Maybe it is a better compiler. Maybe it generates better code. Maybe it has better error reporting. I don't know if it has any of these things, but it certainly has a good reputation.
I would love to see other vendors follow suit. If Borland opened it's Object Pascal compiler (independent of Delphi), I'd switch to it for many development projects.
> It is unfortunate that it requires some proprietary components in order to function.
The only proprietry components are the SDK's, not the tools that are used. Included in the Open Source release will be full source code to *all* utilities such as the compiler, linker, librarian, assembler, resource compiler and much more.
We are planning the Open Source 1.0 release to require the 11.0c compiler in order to build, but be able to rely on freely available SDK's to actually produce programs. Ie: you can download the Open Source 1.0 official binaries and the Microsoft Platform SDK (from the M$ web site), and start happily writing Win32 programs.
One useful addition to the project would be additional support for non-OMF and non-PE executeable formats. Ie: support for ELF would be nice, so that it can directly generate Linux executeables.
- Kendall
Watcom was an excellent C compiler; I hope some of their floating-point optimizations will eventually be folded into egcs, because I remember the default DOS-compiled BYTEMarks had a bit of an edge there.
:)
(Also, someone told me that when IBM recompiled the Windows source with it, it was about 30% faster at the time. I think that was for OS/2, but it was funny at the time.
---
pb Reply or e-mail; don't vaguely moderate.
pb Reply or e-mail; don't vaguely moderate.
I hope we see lots of great new additions to Fortran as a result of this :)
DrLunch.com The site that tells you what's for lunch!