Re: Someone using COMPAT_SVR4(_32) ?

[Maxime Villard <max%m00nbsd.net@localhost>]

Le 13/09/2017 à 17:25, Jerome Ibanes a écrit :
> Then it's likely that one of the changes between the 5 and 6 branches broke
> compat_svr4, other than going through each patch in arch/sparc64 and running
> a test on 6.0.x; which other option do you suggest to narrow this down?

That's perhaps gross, but I would diff the sys/arch/sparc64, sys/compat/svr4 and
sys/compat/svr4_32 directories between 5 and 6, and try to figure out which
part got modified in a way that breaks svr4_32. The diff shouldn't be too big.

Or you could try to manually follow the execution of the syscall, from the very
entry point to the generic mmap code, and see if changes have been made in this
area. But it is possible that nothing got changed in here, and that the bug is
the result of some distant, hard-to-flush-out mistake in an unrelated sub-
system.

Apparently no one realized a regression was introduced in compat_svr4 before
you did - more than seven years of zero testing.

Maxime