Re: Please don't drop support for AmigaOS
lists.gnu.org
lists.gnu.org
--
First, if you look through the Make code, you'll see it's riddled with #ifdef lines for each supported OS. In modern code you'd separate these out into per-OS sub-libraries with a common API, but in GNU code it all gets mashed together.
Here's a good example I found from a quick search -- it's trying to find a default value for the user's home directory, which could easily be a per-OS function, except that that's not how it worked in the '80s so instead they embedded environment variable lookup into a SEVEN HUNDRED LINE glob() FUNCTION IN lib/glob.c.
#ifndef VMS
if ((flags & (GLOB_TILDE|GLOB_TILDE_CHECK)) && dirname[0] == '~')
{
if (dirname[1] == '\0' || dirname[1] == '/')
{
/* Look up home directory. */
#ifdef VMS
/* This isn't obvious, RTLs of DECC and VAXC know about "HOME" */
const char *home_dir = getenv ("SYS$LOGIN");
#else
const char *home_dir = getenv ("HOME");
#endif
# ifdef _AMIGA
if (home_dir == NULL || home_dir[0] == '\0')
home_dir = "SYS:";
# else
# ifdef WINDOWS32
if (home_dir == NULL || home_dir[0] == '\0')
home_dir = "c:/users/default"; /* poor default */
# else
# ifdef VMS
How do you test that? In modern code you'd have an OS-independent unit test for glob expansion, and a separate test for "get homedir on AmigaOS", and the AmigaOS code would never change because why would it?Plus, the code itself is horrible, like if you read this and think "wait, they're nesting `#ifdef VMS` inside `#ifndef VMS`, what is their intention" then buckle the fuck up because the entire project looks like that.
--
Second, many GNU projects use a library named gnulib, which is a sort of anti-portability library developed to make every system look like GNU libc. To achieve this goal the gnulib code digs deep into opaque implementation details of the underlying system libraries, stuff like the layout of `struct FILE` or whether particular symbols are constants or macros. As you might expect this shit breaks constantly, even the most dirt-common gcc+glibc+Linux system might fail to build a GNU project from a year ago because the glibc folks changed something that gnulib implicitly depended on.
Things get even worse from there, like if you try to build with musl or Clang then you're in for potentially days of reading mailing lists and digging up old CVS checkouts to figure out how to get the damn thing to compile. And we haven't even talked about OS portability yet! It took me weeks to get a working GNU Bison build on Windows because I had to hunt down dozens of places where they assumed that sizeof(int) == sizeof(long) == sizeof(void*). That's on the world's most popular workstation OS, good fucking luck getting any sort of confidence that the code works as expected on an Amiga.
--
In summary, the GNU folks have intentionally structured their code such that it's difficult to maintain across platforms, and this email list is merely one of the observable outcomes of their decision.
I should be able to run the test "expand this glob as if on AmigaOS, with given homedir path" on a Linux desktop.
Anyone compiling on an Amiga is more likely to be using SAS/C
SAS/C was already dead in the 90s.
GCC and VBCC are the options today.
Anyone using either of those is cross-compiling (hence my comment specifying "on an Amiga")
It's unlikely to find many people running GCC on AmigaOS itself so it makes sense to deprecate the support for it
Yes they are. People actually use their Amiga computers and love doing so. Outside of modern web browsing, they are very capable machines.
Cross-compilation does definitely have some users, but is complex and prone to issues.
And we're far from a situation where these people don't care; LLVM m68k support was merged recently. It was an effort funded mostly by Amiga and Atari ST users.
What complexity and issues? Everyone I know is developing their Amiga software, libraries, devices etc using things like bebbo gcc or vbcc on their PCs
> And we're far from a situation where these people don't care; LLVM m68k support was merged recently. It was an effort funded mostly by Amiga and Atari ST users.
Where can I find an LLVM build for AmigaOS m68k though? Not on Aminet I suppose anyone using that must be running it on a pc
I still use my Amiga constantly but I'm not going to pretend there's more than a handful of people actually running bleeding edge versions of GCC under AmigaOS 3.1
Just because they no longer support GCC builds for Amiga doesn't mean it won't be able to build for the target just that you won't be running newer versions of GCC under AmigaOS itself