The porting infrastructure includes several scripts under
infrastructure/bin that facilitate the creation of new ports:
- check-lib-depends
- invoked through
make lib-depends-check, to verify shared
libraries dependencies.
- update-patches
- invoked through
make update-patches, which should
always be used to regenerate patches.
- update-plist
- invoked through
make update-plist.
This takes care of most of the finest points of crafting accurate
packing-lists.
OpenBSD packing-lists are significantly different from those of other
BSD projects, in part because the package tools have been completely
rewritten.
Check the infrastructure/bin directory for more useful scripts.
Most of them have man pages.
Generic Infrastructure Issues
OpenBSD's make(1) supports
${VAR:U} and ${VAR:L} to transform a variable's value
into uppercase or lowercase. Accordingly, make tests should be coded in
a case-independent way.
For example:
.if ${NEED_XXX:L} == "yes"
do stuff if yes
.else
do other stuff
.endif
In theory, all boolean variables recognized by bsd.port.mk should
always be defined, so that code like defined(USE_FOO) should not
be necessary, and ${USE_FOO:L} != "no" ought to work.
The main bsd.port.mk file has been heavily streamlined and fixed.
In particular, it is parallel-make ready.
The scripts/{pre,do,post}-* feature has been lost in the process.
To replace that feature,
invoke the script manually from the Makefile.
Using make Properly
Note that if you invoke make as make VAR=value,
the assignment will override whatever value VAR may get from the
Makefile.
This means that many Makefile patches are not necessary.
It is much better to set MAKE_FLAGS correctly, which decreases the
maintenance burden.
Fetching Sources
There are two kinds of source archives: DISTFILES and
PATCHFILES.
OpenBSD processes them in a uniform way and retrieves everything from
MASTER_SITES by default.
There are no PATCH_SITES
nor PATCH_SITES_SUBDIR.
If all files to fetch don't come from the same set of sites, OpenBSD allows
the extension filename:0 to filename:9, in which case
it will use MASTER_SITES0 to MASTER_SITES9 to
retrieve the file.
Some architectures may need specific distfiles.
In the past, this has caused trouble where mirroring distfiles was concerned.
OpenBSD supports a third set of files: SUPDISTFILES.
Those will be considered only for creating checksums and mirroring purposes.
Note that SUPDISTFILES may have an overlap with
DISTFILES or
PATCHFILES.
For instance:
DISTFILES=foo-1.0.tgz
.if ${ARCH} == "i386"
DISTFILES+=foo-i386.tgz
.elif ${ARCHI} == "amd64"
DISTFILES+=foo-amd64.tgz
.endif
SUPDISTFILES=foo-i386.tgz foo-amd64.tgz
The WRKDIR Infrastructure
We don't want ports that use NO_WRKDIR.
All OpenBSD ports must have a work directory.
Naming details of those work directories should not be a porter's concerns.
If you need to find out about such a name, ask the Makefile:
$ cd that_ports_dir && make show=WRKDIR
This will yield that port's idea of WRKDIR.
The main reason behind this prohibition is that OpenBSD's
bsd.port.mk
acts like a real Makefile, with dependencies.
The fetch stage depends upon the distfiles and patchfiles, and all
other stages depend on real files living in the working directory (cookies),
so they can't exist without a working directory.
If the DISTFILES extraction is special, set
EXTRACT_ONLY=
and do the extraction in post-extract.
WRKDIR- The port working directory, where it puts its own cookies.
WRKDIST- Subdirectory of
WRKDIR where the port actually unpacks.
It is also the base directory for patch.
Other BSDs currently don't have the WRKDIST/WRKSRC distinction and
have only WRKSRC.
WRKSRC- Subdirectory of
WRKDIST where the actual source lives.
WRKBUILD- Subdirectory of
WRKDIR where the port configure and build will
occur.
Other BSDs don't have the WRKBUILD/WRKSRC distinction.
Programs based on autoconf (mostly) can usually set SEPARATE_BUILD
to let the port build happen in a WRKBUILD distinct from
WRKSRC.
WRKCONF- Subdirectory of
WRKDIR where configure scripts should be run.
Defaults to WRKBUILD, which is correct 99% of the time.
WRKINST- Directory where the port will be installed before being packaged (see
"faking ports" below).
Note that NO_WRKSUBDIR has been removed: its functionality can
be achieved by setting WRKDIST=$(WRKDIR) instead.
Fake Port Installation
Introduction
After a build is complete, other BSDs proceed to install a port, then build
a package from the installed port.
OpenBSD uses faked installation instead.
- An OpenBSD port is configured and built normally (e.g., to install under
PREFIX, usually /usr/local).
- But it's told to install elsewhere, namely under
WRKINST, which
is usually a subdirectory of WRKDIR.
- Then the false installation is packaged, using the
-B option of
pkg_create.
- Finally, the resulting package can be installed, using pkg_add.
Advantages
- For a package builder, it means that most ports don't have to actually be
installed, which removes a large number of potential compromises and general
nastiness from badly-installed ports.
It also allows building several conflicting packages on the same machine.
Finally, it allows building a new set of untested packages without hosing
a correct installation.
- For a port writer, it greatly simplifies the task of finding problems in
packing lists, since the fake area of installation is empty before the port
gets installed.
Also, if a port installs too many files, it is no longer necessary to tweak
the port installation: it is enough not to record the extraneous files in
the packing list.
- For the end user, it improves the quality of packages: since the final
port is installed using pkg_add, the end user gets exactly the same
software that was prepared on the porter's machine.
How to do it
The targets invoked for make fake are the usual install
targets, except for a few differences:
FAKE_FLAGS is used instead of MAKE_FLAGS.
By default, FAKE_FLAGS sets DESTDIR=${WRKINST}.
FAKE_TARGET is used instead of INSTALL_TARGET.
- The
{pre,do,post}-install fragments are invoked with
TRUEPREFIX set to $(PREFIX),
PREFIX set to $(WRKINST)$(PREFIX), and
DESTDIR set to $(WRKINST).
Ports using imake should work as is, since the imake fragments are configured
to use DESTDIR.
Similarly, recent GNU configure ports should need no change.
Another good technique is a "late binding" trick: configure the ports to use
a prefix of $(DESTDIR)/usr/local, so that the resulting
Makefile has the following set:
prefix=$(DESTDIR)/usr/local
When the port gets built, since DESTDIR is set to nothing,
/usr/local is used.
The fake install will put everything into ${WRKINST}/usr/local
(e.g., for GNU configure, use CONFIGURE_STYLE= gnu dest).
Pitfalls
- Some ports are inconsistent in their
DESTDIR processing: most
of the port is happy with DESTDIR set, except for one or two
offenders.
Patch the problem away.
- Be careful to distinguish between the actual location where the port is
installed, and the location recorded in the configuration files of the package.
This is very easy to overlook, but easy to fix using
TRUEPREFIX.
- Absolute symlinks always need to be tweaked.
Luckily,
bsd.port.mk will notice problems in that area.
- A few ports don't want to leave
$(DESTDIR) alone at the
configure stage.
A post-configure fragment that tweaks all Makefiles to add the
DESTDIR is needed.
- Very seldom, a port will resist all reasonable attempts to use FAKE.
A brute force approach should work: use
pre-fake to link or copy
everything the port wants to find in the WRKINST area, then perform
the install under chroot.
The package tools know about quite a few file types and can do a lot of things
automatically: in most cases @exec commands or INSTALL
scripts are unneeded.
Note that all unneeded scripts should be banned, as they have scalability
issues.
It is much easier to debug one single package infrastructure than to modify
hundreds of scripts to handle new problems.
For instance:
@exec ldconfig is not needed, as shared libraries are
annotated with @lib libfoo.so.1.0 and ldconfig
runs only when needed, and handles chroot gracefully.
@exec install-info is not needed, as info documentation
files are annotated with @info file.info.
This also takes care of multiple info files, and removes the need for
makeinfo --no-split.
- fonts get integrated automatically thanks to
@font and
@fontdir.
- New users and groups get handled with
@newuser and
@newgroup instead of installation scripts.
They also get created early enough so that further package extraction can
use them.
- Most third party database handling is handled through
@tag
which triggers running tools like update-desktop-database once at the
end of installation.
- Configuration files are handled through
@sample instead
of installation scripts.
Refer to pkg_create(1) for
more details.
In most cases, make update-plist will write a very good
approximation of a complete packing-list, and will carry hand-made tweaks
from one version to the next.
Flavors
Options have been rationalized as flavors, so that package building can be
consistent.
A port with options should set FLAVORS to the list of all options
that make sense for that port (e.g., FLAVORS=foo bar zoinx),
then use
FLAVOR to test what options have actually been selected (e.g.,
FLAVOR=zoinx foo).
bsd.port.mk provides some support:
- The
PKGNAME is tweaked to include dash separated options (e.g.,
package-foo-zoinx).
- The
WRKDIR is tweaked so that distinct flavors can be built
concurrently without colliding.
- Constructs of the form
%%flavor%% will trigger the inclusion of
PFRAG.flavor.
bsd.port.subdir.mk understands the extension
SUBDIR=directory,opt1,opt2 to say
"build port in directory with FLAVOR=opt1 opt2."
Checking that a given flavor has been selected is as simple as:
.if ${FLAVOR:Mzoinx}
There is an extra extension, known as MULTI_PACKAGES.
Generally speaking, MULTI_PACKAGES and FLAVORS
are orthogonal mechanisms.
Together, they account for OpenBSD ports tree being somewhat smaller than the
other BSDs, as they allow one single port directory to build lots of distinct
packages.
bsd.port.mk(5) has
a full
section devoted to FLAVORS AND MULTI_PACKAGES.
¡®Yes, sir. I felt sure you understood that. She said she had told you.¡¯ "Why, eh,--I--I don't know that my movements need have anything to do with his. Yours, of course,--" "Ah, but if it saved your life!" "No, I'm not," grumbled the Doctor, "I've had enough of this wild-goose chase. And besides, it's nearly dinner time." "I am coming to that," Lawrence said, lighting a fresh cigarette. "As soon as Bruce was in trouble and the plot began to reel off I saw that it was mine. Of course there were large varyings in the details, but the scheme was mine. It was even laid on the same spot as my skeleton story. When I grasped that, I knew quite well that somebody must have stolen my plot." Judy In a coach-house, through which we passed on our way to see the prince's favourite horses with the state carriages¡ªquite commonplace and comfortable, and made at Palitana¡ªwas a chigram,[Pg 68] off which its silk cover was lifted; it was painted bright red and spangled with twinkling copper nails. This carriage, which is hermetically closed when the Ranee goes out in it, was lined with cloth-of-gold patterned with Gohel Sheri's initials within a horseshoe: a little hand-glass on one of the cushions, two boxes of chased silver, the curtains and hangings redolent of otto of roses. "Are you certain of it? You have seen so very little of him, and you may be mistaken." "And your wife?" "I drawed on my man's bundle o' wood," said Gid, "and then dropped a little, so's to git him where he was biggest and make sure o' him." HoME²¨¶àÒ°½áÒÂ×óÏßÊÓƵ
ENTER NUMBET 0016eureka.org.cn
lwchain.com.cn
inqu.com.cn
www.hlchain.com.cn
mfchain.com.cn
www.idpro.com.cn
www.hyboao.com.cn
tjftz.com.cn
www.oizsml.com.cn
wjflhs.com.cn