• 1.0 -> scripts/param.h, paxmirabilis: MFC the new version from trunk
• both 1.0 and trunk: implement “make targz”, “make tarbz2”
  (I like the gzip(1)d versions better though)
• 1.0 -> package/config/Makefile: quieten the “clean” target to be consistent

new target “distclean” for packages

optimise

hack to make new “make wconfig”, developer-only, to generate Config.default
feel free to improve

remove target_clean, it can't do what it appears to do since it removes the di-
rectory where all the compiled IPKs are stored, and it's not documented anyway

some wording fixes, 80-column limit, and “make package=foo” will NEVER
resolve dependencies. description of update-patches was already here @wbx

use host-system mksh instead of bash to speed up the splitting process,
if installed

new: “make slow” – compile, but output only as few as can be transmitted
via a slow line, such as a serial link

wbx@ only allowed me to default to V=99 if I commit “something like” this

future improvements: change the TRACE variables to echo special control
characters which are then picked up by “make slow” as well; convert it
from using tail(1) to a shell-based line buffer (slightly more CPU in-
tensive; mksh would be more than 3 times as performant as GNU bash here)

Ok, this is the drawback of integrating new functionalities:
in the end it is always a lot more work than one expected. :)

So as requested, here comes the big documentation. To prevent
screen flooding, I split everything up into three parts:
* general make targets
* package specific targets (though generic in some way)
* package maintainers target(s)

The only maintainer target is 'update-patches', but as the most
complicated part of using it is to know how, this should fit
into a separate target. (And will hopefully grow in the future.)

some missing additions

integrate a 'help' target

oops, "v" was not verbose any more

default to verbose, this is for developers, and the users have WIB

talked over with wbx@

if a build system has a non-GNU make, or a too old gmake, just
build our own

tested on MirBSD odem.66h.42h.de 10 Kv#10n11-20070318 GENERIC#1014 i386
without gmake installed

if this is not working for you for the one or other reason,
please drop me a private mail with a detailled error description.

this procedure was explicitly ok'd by wbx@

• revert the addition of “build gmake as part of tools“, it had too
  many problems
• for now, gmake 3.81 IS REQUIRED
• clean up

try to work around hen-egg problem, sorry dnehring@, please continue to test this, I cannot

by request of markus@ use the “old” $PATH when calling the “real” ${CC},
one without ${TOPDIR}/lbin in it, so that the commonly used, broken, idiom
of 'CC="ccache gcc"' can be supported (untested)

to use this: make prereq

add “oldconfig” target, thanks to nbd@openwrt for the hint
(untested)

check if host compiler supports -fwrapv
should fix PR#201

unbreak damage introduced by dnehring@ in r1771

Request tg: move -DNDEBUG in CPPFLAGS. r1768 was wrong.

Disable assert() by default to save space.

the already-existing global ${HOSTCFLAGS} was not used when
determining ${HOST_CFLAGS} from ${CFLAGS} and setting a default;
use it (so that HOSTCFLAGS = HOST_CFLAGS) and the defaults

* we override LC_ALL (already), no need to do it twice
* LANG is overridden by LC_ALL anyway
* on some systems, GNU dcgettext overrides LC_ALL with
  LANGUAGE, so we force-set that to 'C' as well

* '%' as line addressing is a BSD extension to the text editor,
  which GNU ed 0.3+ doesn't support, but isn't listed in the
  ChangeLog relative to GNU ed 0.2 - use ',' instead
  bug cought by n0-1@
* correct line addressing of "TOPDIR:=..." line in prereq.mk

got another great idea: why not check for supported OSes
during 'make prereq', tell the user to contact us (we may
even have a (commercial) contractor available for the task
of porting to a new build (not host) system) and what the
issues with some of the OSes are

also move other uname-related stuff from Makefile into
scan-tools.sh so that we keep them in one place

Introduced a new make target `snapshotconfig'.

This target is quite equal to the well-known `allmodconfig', 
with the extension that all possible target devices are selected
at the same time. This leads to also enabling packages which depend
on certain target devices, what is useful for snapshot builds.

Enabled the well-known "allmodconfig" target.
The idea is merely to build something that fits everywhere,
but rather for testing builds including everything. (This
way one can deselect senseless packages, like e.g. busybox
which is totally useless with everything enabled.)

Also synced Config.in.collections with the one from 1.0.

when changing something, don't forget to remove the old
code when introducing new code ;)

cought by wbx@

experimental support for having CC="ccache gcc", etc.
reported by tntnet in IRC, not tested

new target: allnoconfig - creates a config with everything disabled

remove autobuild and snapshot targets. snapshots and releases will be done with ib.sh

we use the httpd redirect here, my bad, this is by design
and actually makes sense

by request of markus@ communicated via wbx@, add new facilities:
$ make [V=99] package=mksh clean prepare compile install
$ gmake [V=99] subdir=tools/paxmirabilis clean prepare compile install

This works as follows: if ${subdir:-${package+package/${package}}} (in shell
syntax) is specified (you can only specify one subdirectory, and the package
variable has precedence over the subdir variable), all make targets specified
on the command (if any) are passed to a GNU make run in the subdirectory spe-
cified (packag/${package} or simply ${subdir}), relative to ${TOPDIR}, with
environment appropriately set and ${MAKEFLAGS} passed.

also, while here, recind the advertising clause on the top-level Makefile and
integrate this stuff with the new boilerplate (attn: licence URI adjusted);
the requirement to credit my name is contained in the GPL as well

merge wbx-target-cleanup to trunk.
- mk directory for some top level makefiles (build.mk,vars.mk)
- add a separate configfs partition to all models, 128 kb big
- add model and platform specific startup files and kernel configs
- delete target/linux/package, add all addon kernel packages to packages
- simplify target/linux/brcm-2.4/Makefile, abstraction to mk/ comes later
- add target/image directory, place where the different images are created
- default off for all extra packages, be sure that you enable all packages
  you need to get a dsl-capable router. snapshots will be created by a specific config which contains all needed stuff (iptables, pppoe, pptp, haserl, webif)

stupid typo, cought by wbx@

wbx@ requested another iteration, per-filesystem.

*sigh*

first I'm told conf -d is needed, then I'm told it overwrites .config...
remove it

typo

* .defconfig: remove superfluous commented-out targets
* build.mk: move some targets out of the if-have-.config
* build.mk, Makefile: hopefully fix wbx@' snapshot target
* Makefile: sync licence

rename target, suggested by tg

add first hackish parts of a snapshot target

even better, use more relative pathnames,
causes less breakage when TOPDIR is a symbolic link ;)

relativify TOPDIR in prereq.mk and make it the first line

add autobuild, still busybox need to be excluded in any good way. no idea for that at the moment. remove unused stuff.

hide output of find, add kernel packages to update-package

as discussed with wbx@ some 2-3 weeks ago, change
the BR2_ prefix to FWRT_

tested with a standard build (only the squashfs
image since the jffs2 didn't get built)

revert r348
described in Message-ID: <Pine.BSM.4.64L.0607221359490.32684@odem.66h.42h.de>

i think this is some leftover from yesterday's hacking session, seems to work without this line, verified with a build and reflash

I guess I should use := instead of = in prereq.mk to benefit GNU make
also add GMAKE there, as MAKE can be both

speed up build and ease transition to BSD make by determining
HOST_ARCH during make prereq and cacheing it in prereq.mk too

add TOPDIR to prereq.mk

move the export line from prereq.mk to rules.mk and remove SHELL from it
since that's special (always use ${BASH} if you must) according to TFM

sanitise the way tools which must be in lbin/ (currently,
sed and pax/cpio/tar, but there will be more...) are handled:
make prereq invokes a special recursive target install-lbin
which copies them from the staging dir into lbin if they exist
there (i.e. are already installed), or removes the install stamp
otherwise. the tools' install-lbin target does however duplicate
a little code, this could be done better but it works OK like this

also, add a missing include to brcm-utils and fix pax' clean target

be quiet

it gets even weirder with support for Darwin 7.9.0 ...

the macintosh never ceases to amaze me... csh(1) which...

switch to paxmirabilis

ok wbx@

NB: not yet totally tested, but if anything breaks I'll fix it today/tonight

keep everything within 80 characters width

forgot one dollar escape

try some autobuild stuff

use the GNU sed built by toolchain again, but prevent it
from picking up I18N support (causes it to segfault at
least under mirbsd - GNU libregex clashes)

ok wbx@

this is not linux, er kansas, any more

save the old $PATH as _PATH in 'make prereq',
so that it can be picked up later
please re-run 'make prereq' after this commit

hint: /usr/mpkg/bin/bash wasn't in the path

ok wbx@

obviously, perl isn't compatible either. well, what did
you expect of trying to throw an espie?

on systems without sed --in-place in --help, use a shell
script which emulates that behaviour. (requires /usr/bin/sed)

this is so horrifyingly cheap I cannot believe I'm committing this...

the lunox kernel makefiles are so unbelievably broken that
we cannot override $(HOSTCC) from gcc *EVEN IF SET* on the
gmake command line, so symlink $(HOSTCC) to lbin/gcc - for
now...

suggested by wbx@

require GNU tar and install it as tar, for now

ok wbx@

stupid GNU make incompatibility

noticed by js in IRC

* have 'make prereq' and 'make prereq-noerror' be .PHONY
  targets, i.e. if they're called directly, they will be
  always remade; the other targets now depend on a regu-
  lar '.prereq_done' target to not redo the prerequisite
  checks unless needed
* do not call gmake with --no-print-directory if verbose

ok wbx@

expose $BASH too

ok wbx@

tell gmake to use GNU bash

ok wbx@

after creating prereq, tell the user it's done

ok wbx@

On Debian, gmake 3.80 "seems to be" enough.
Allow the user to override the version check,
point to the FAQ.

requested by and "okay, cool" wbx@

Check some of the prerequisite tools (WIP, more to come)
* that bash exists and works
* that GNU make and a host compiler (gcc) exist
* that we can run a compiled test programme (for the host)
* the GNU make version number

ok wbx@

Require, sadly, GNU bash to be installed before proceeding.

"sehr schön" wbx@

* instead of using a $(VARS_TO_PASS) write the variables into a
  *.mk include file, so that people can cd to some package direc-
  tory and invoke gmake directly
* fake OpenBSD as uname for MirBSD, so that we get a little further
  in the build process there (stops at linux requiring bash now)

Tested with a full build (standard config) on Linux.
Invoking gmake in a subdirectory *not* tested, but
that didn't work before at all, so it's not a problem.

quiet build

set --no-print-directory for make calls, so that non-verbose is really non-verbose

prepend a blank line and a descriptive line with date
to a continuation of 'make.log' to make it easier to
follow

resolve a bad interaction between shell quoting
and shell evaluation of inside commands in the
$(HOSTCFLAGS) determination algorithm

it's a pity we have to be compatible to a lowest
common denominator of make and sh for all systems...

change 'make v' target, which was the same as 'make verbose',
to additionally append standard output and error (i.e. a full
build log) to a 'make.log' file; svn:ignore the make.log file

copy this:
 # Ensure consistent sort order, 'gcc -print-search-dirs' behavior, etc.
 LC_ALL:= C
from package/config/Makefile; I think it's a good idea to
normalise behaviour on systems which do support locales,
because e.g. in Spanish locales, sort order varies, etc.
and that introduces hard-to-find bugs (non-locale-aware
operating systems such as OpenBSD not affected).

provide sensible defaults (-O2) for systems with no default CFLAGS
tested on lunox

note: this also, as a not unwelcome side effect, changes the
default flags used for builds from "-g -O2" to "-O2" on the
standard GNU installation (and "-O2 -pipe" on OpenBSD, I
think). More flags (e.g. debugging) can and will be added
later.

Correctly determine HOSTCC and HOSTCFLAGS (from host make)
and pass them as CC/CFLAGS to the toolchain as well.

Now this at least goes as far as I suspected the first
breakage would be (on MirBSD) - autotools.

Detecting works by invoking $(MAKE) with a BSD-specific
makefile and just using $(HOSTCFLAGS) if that fails.
This works better on OpenBSD too, FWIW.

cosmetic changes to Makefile/build.mk
remove configtest.pl
cosmetic changes in freewrt/toolchain/sed/Makefile, add md5sum

change target distclean to also remove the download area
(distclean = "as clean as if we were making a distfile")

new target cleandir removes everything removed by distclean
except the configuration and downloaded distfiles
(cleandir = clean + toolchain directories)

removed target dirclean, replaced by ordinary clean target
(clean = clean packages, bindir, builddir)

ok wbx@

use pax to extract if it exists, instead of tar,
because it handles certain archives created by
GNU tar better (on BSD), no change on Linux

ok wbx@

add 'make verbose' target

ok wbx@

Initial cut at supporting FreeWRT builds under other operating
systems (MirOS BSD, OpenBSD, probably the other BSDs as well;
Mac OSX and Cygwin to come)

The basic idea is:
* the old GNU make "Makefile" is renamed to "build.mk"
* a wrapper Makefile which uses neither GNU nor BSD make
  features calls GNU make with -f build.mk
* the wrapper Makefile also checks for prerequisite
  tools (as of now: existence of gmake and md5sum)
  and their versions (not yet, but that's why it's
  being used on GNU/Linux as well)
* scripts/md5sum.bsd uses BSD cksum(1) to generate
  GNU md5sum compatible output (only when the arguments
  are files; no options are supported)
* GNU make is symlinked into ./lbin/; md5sum.bsd is
  copied there if needed; ./lbin/ is prepended to
  $PATH for gmake -f build.mk invocations
* make distclean removes prerequisites (and lbin)