There should be an option with packages to re-use an existing user or
group, even if a specific uid/gid is provided.  This commit introduces a
user variable to configure this: PKGINSTALL_IGNORE_UIDGID.

When PKGINSTALL_IGNORE_UIDGID is true then pkg_install will not try to
create the user/group as specified.  If the user/group does not exist
but the specified uid/gid is, then the uid/gid will be stripped so that
the system can allocate one at random.

This allows specifying additional libraries in the output phase, and is
a special case compared to _WRAP_EXTRA_ARGS as we need to ensure that
these libaries are not applied when using Sun ld in -r mode.

  * Disable zlib so that we don't pull in platform libz.
  * Set OVERRIDE_DIRDEPTH.depcomp=1 to disable depcomp-override

This allows arguments to be added when a wrapper is generating an
object, and is useful for inserting a specific library into each,
for example -lumem on SunOS.

 - Remove TOOLS_BASEDIR references.
 - Set search path with COMPILER_LIB_DIRS.
 - Don't hardcode GCC library paths.
 - Can't use -nostdlib (due to the above change).

mk/compiler/gcc.mk: remove libtool-base handling.

We need this to ensure that sys_lib_search_path_spec is correct and
prefers the runtime library.  Without this we see failures where
packages using libtool try to use both libstdc++.so's.

Ensure we can use an external libtool for binutils.

Avoids accidentally picking up a broken/non-working one from the
environment, or linking against one we don't have installed everywhere.

We already set them explicitly in pkgbuild and besides avoiding
duplicates this also allows us to change them without having to rebuild
bootstrap.

Resolves issues when bootstrapping without a suitable pkg_info
available, as the bootstrap pkg_info is unable to operate on binary
packages.  Fixes TritonDataCenter/pkgsrc-legacy#564

It is only used for c_rehash and a couple of other example tools, and
users can install perl manually if they require them.  It bloats too
much by default for something that is used so little, if at all.

 - Add ALTERNATIVES files.
 - Include necessary object files in the -libs package.
 - Prune unnecessary libraries from the -libs package.
 - Set local prefix to LOCALBASE (except tools build), and link against
   it by default.
 - Disable -fomit-frame-pointer, we like stack traces.
 - Disable --as-needed to avoid libgcc_s dependency.
 - Disable __stack_chk_fail_local (fixes 32-bit)
 - Support -fclone-functions, -fstrict-calling-conventions, -msave-args
 - Fix check-shlibs in -libs
 - Fix TritonDataCenter/pkgsrc-legacy#264 (avoid tls emulation)

 - Add ALTERNATIVES files.
 - Include the .o files in the -libs packages.
 - Prune unnecessary libraries from the -libs package.
 - Set local prefix to LOCALBASE (except tools build), and link against
   it by default.
 - Disable -fomit-frame-pointer, we like stack traces.
 - Disable __stack_chk_fail_local (fixes 32-bit)
 - DATASET-953 disable full gcc47 dependency
 - Implement -fstrict-calling-conventions, from richlowe.
 - Cleanup DEPENDS when using -libs
 - Fix check-shlibs in -libs
 - Fix TritonDataCenter/pkgsrc-legacy#264 (avoid tls emulation)
 - Fix for building with newer GCC.

These update a lot but not in a way that will affect the scan results,
and if that is the case then we remove old bulklogs anyway.

This adds a chroot handling script and a few additional variables to
pbulk.conf:

  - $chroot_create and $chroot_delete point to scripts which create and
    delete a sandbox respectively, each taking a path argument of the
    chroot location.

  - $chroot_dir is the base directory where chroots are created.

  - $build_chroots and $scan_chroots specify the number of build and
    scan chroots to create respectively.

If the number of chroots is 1, they are named:

  ${chroot_dir}-build
  ${chroot_dir}-scan

otherwise they are named i=1..N:

  ${chroot_dir}-build-${i}
  ${chroot_dir}-scan-${i}

The chroot_dir variable isn't brilliantly named, but at this point is
baked into too many of our pbulk branches to change.

Currently users of this need to provide their own create/delete scripts,
though hopefully at some point I'll provide sysutils/mksandbox wrappers
as examples for getting started.

This allows a single pbulk install to handle multiple builds easily
without having to switch out the main pbulk.conf each time.

These are passed in as SUBDIR entries by the top-level Makefile.

This allows local customisations to the client-clean script, for example
our OSX /tmp ramdisk cleanup:

  https://github.com/joyent/pkgbuild/blob/master/scripts/client-clean-osx

which is necessary to keep the ramdisk usage pruned.

This is useful when e.g. using Oracle Java which cannot be
redistributed.  It should still be possible to install packages which
depend upon it after the user has installed the restricted package
manually first.

There's no guarantee that the pbulk master is running the same OS
version that the pbulk clients are.

Skipped on Darwin as it has a centralised OpenDirectory database,
however finding a way to do this on Darwin would be greatly beneficial.

Assumes pbulk.conf is populated with the appropriate variables and that
nodejs and the 'manta' npm package is installed under $MANTA_PATH.

This saves a decent amount of time at the end of a pbulk run (around 20
minutes) by avoiding a single-threaded operation to generate them.

XXX: not currently handled by custom $sync_package scripts

This is a directory where the bulklog files are written to temporarily
before either being moved to $bulklog at the end of a failed build or
removed.

This avoids failures when writing to an NFS-backed $bulklog (seen most
commonly with 'gmake: write error'), and also improves performance.

This is set to a script that is executed by the upload script prior to
starting the actual upload, and is useful for performing any sanity
checks on what is about to be uploaded.

Add variables used by Joyent to ensure a certain number of packages
built ($pkg_rsync_min) and that a list of core packages all built
successfully ($pkg_rsync_required_pkgs) to the example config.

Avoids issues seen on SunOS and Darwin where LIST_NEXT would return
garbage.  Use event_del() while here.

This updates the mysql.buildlink3.mk file to support multiple versions,
and adds some support for building multiple p5-DBD-mysql and py-mysqldb
packages.

Based on a patch from richlowe, expanded to handle all current possible
-std= variants except e.g. "-std=iso9899:1999" as it's not clear how to
handle the embedded colon in the startfile syntax.