Document the meanings of bootstrap_foo vs fooprog.
On SunOS, if we find /usr/bin/bash and use it for bootstrap_sh, also
use it for shprog. Resolves failure to run libarchive's configure on
old SunOS. (It seems likely libarchive's configure is using modern
constructs that are best avoided, but it is also good to use a better
shell if available.)
Problem and solution from Vivianne Ombretta via pkgsrc-users. The
explanatory comments are gdt's fault.
NATIVE_PKG_*_CMD is what gets executed at build-time, and must be the
bootstrapped pkg_install which lives in the working tree; PKG_*_CMD
may get baked into packages for use at run-time.
This allows the user to specify an exact SDK to use, and can be used to build
packages for an older release of macOS than the host. The user should ideally
set this via environment variable at bootstrap time, and pkgsrc will then
encode that into mk.conf and use it for all builds.
Tested on macOS 12.x building against an 11.3 SDK for both arm64 and x86_64.
This was introduced many years ago in PR#26143 to try and work around a
misconfiguration in how Apple's GCC was built at the time. Newer compilers no
longer look in /usr/local/include, and this workaround can now be harmful when
trying to pick a specific SDK to build against.
I was able to reproduce an OpenBSD user's boostrap failure because they
followed the instructions in this file, which doesn't mention
'--prefer-pkgsrc' at all. And yet it was required in order for me to
succeed on OpenBSD/amd64 7.1
Fold README.IRIX5.3 into the IRIX README; we don't have multiple
READMEs per OS, and it's clutter, especially given the information is
from 2006.
Add last-updated dates (2005/2006) as a clue.
Separate active use, maintained, and published bulk builds. This
amounts to a new users/maintained-but-no-bulk section for FreeBSD.
Sort OpenBSD into "platforms with active use" based on list comments.
Drop mention of Darwin, as I can find on evidence that Darwin
continues to exist outside of macOS in any meaningful way.
Convert to markdown because the previous style was too confusing. A
lot of this is ancient; mark it as such. Fold in comments from Peter
Lai about working/non-working gcc versions.