Test Result
Some might wondering what happened to the Free Pascal test result on the Amiga. There are some reasons I did not write about the results. 1. it took much longer than though, especially later tests need much longer to compile and run (more like 30s per test) and 2. some tests crash what leaded to idling computer half of the night until I can let it continue, overall it stopped 8 time, of which it was 3 times a „Guru“ and 4 times the enforcer (of Draco) killed the program before it could crash the computer because the program accessed some memory it’s not supposed to and once the Hard disk went full and it stopped with a Disk full requester (I did not notice that the logs and programs need such big amount of HD).
Overall 326 tests are failing either do not compile or the compiled program does not work correctly + for mentioned 7 that crash, which is much more than the m68k-linux test for this compiler version which should be around 40 or. I guess that must be some implementations are missing for the Amiga. I saw scrolling by some Link LibC or other c library errors, which will not work because we did not implement that. The rest I have to check slowly the next days.
It’s not bad after all, 333 tests give problems means around 5600 tests ran well without problems and my Draco did run 3 days without interruption without any problem, also nice to know. After checking the errors, maybe I can fix some of them and make a test list without the 7 crashing tests that I can run on one go. I should repeat that test with the latest Free Pascal and compiled for FPU, especially some tests which are not working with the qemu setup I’m using atm. (if they really show an FPU compiler error or just an qemu problem)
Some of the test we cannot really fix, because for example test for 16 byte static variable alignment, or read only data sections, neither of which is supported on Amiga, obviously. The FPU exception handler is also inoperable under AmigaOS, so we run with exceptions off, this also causes a bunch of test failures, and I’m not sure it’s worth to fix. And yes, libc linking is also inoperable. Well, there isn’t really a „standard“ libc to link against for a start… Different C compilers will use different ABIs and so on, which is non-trivial to support. So it’s complicated. Overall, thanks for this experiment! Please share if anything stands out which should work, but it doesn’t.