// file : doc/intro.cli
// copyright : Copyright (c) 2014-2017 Code Synthesis Ltd
// license : MIT; see accompanying LICENSE file
"\name=build2-toolchain-intro"
"\subject=toolchain"
"\title=Toolchain Introduction"
// TODO
//
// @@ refs to further docs
//
// STYLE
//
// @@ section boundary page breaks (
)
// @@ when printed, code background is gone, but spaces still there
//
// PDF
//
// @@ tree output is garbled
// @@ Install list margins missing
// @@ Could we use a nicer font, seeing that we embed them?
//
// NOTES
//
// - Maximum line is 70 characters.
//
"
\h#tldr|TL;DR|
\
$ bpkg create -d hello cc
created new configuration in hello/
$ cd hello/
$ bpkg add https://build2.org/pkg/1/hello/stable
added repository build2.org/hello/stable
$ bpkg fetch
fetching build2.org/hello/stable
2 package(s) in 1 repository(s)
$ bpkg build hello
build libhello/1.0.0 (required by hello)
build hello/1.0.0
continue? [Y/n] y
libhello-1.0.0.tar.gz 100% of 2428 B 983 kBps 00m01s
fetched libhello/1.0.0
unpacked libhello/1.0.0
hello-1.0.0.tar.gz 100% of 1057 B 6882 kBps 00m01s
fetched hello/1.0.0
unpacked hello/1.0.0
configured libhello/1.0.0
configured hello/1.0.0
c++ hello-1.0.0/cxx{hello}
c++ libhello-1.0.0/hello/cxx{hello}
ld libhello-1.0.0/hello/libs{hello}
ld hello-1.0.0/exe{hello}
updated hello/1.0.0
\
"
"
\h#warning|Warning|
The \c{build2} toolchain \c{0.X.Y} series are alpha releases. Interfaces
\i{will} change in backwards-incompatible ways, guaranteed. Currently, it is
more of a technology preview rather than anything final. But if you want to
start playing with it, welcome and join the \l{https://lists.build2.org
mailing list}!
Our approach to developing \c{build2} is to first get the hard parts right
before focusing on completeness. So while we might still only be building
serially, we do handle auto-generated source code (and, in particular,
headers) properly. In other words, we go depth rather than breadth-first. As a
result, there are plenty of limitations and missing pieces, especially in the
build system. The most notable ones are:
\ul|
\li|Very limited documentation.|
\li|No support for parallel builds.|
\li|No support for custom build system rules/modules.|
|
"
"
\h#intro|Introduction|
The \c{build2} toolchain is a set of tools designed for building and packaging
C and C++ code (though, if it can handle C++, it can handle anything,
right?). The toolchain currently includes the \i{build system} (\c{build2}),
the \i{package manager} (\c{bpkg}), and the \i{repository web interface}
(\c{brep}). More tools, such as the \i{build robot} (\c{bbot}), are in the
works. Then there is \l{https://cppget.org/ cppget.org} (running \c{brep})
which we hope will become \i{the C++ package repository}.
The goal of this document is to give you a basic idea of what the \c{build2}
toolchain can do so that you can decide if you are interested and want to learn
more. Further documentation is referenced at the end of this introduction.
The \c{build2} toolchain is self-hosted and self-packaged (and, yes, it is on
\l{https://cppget.org/ cppget.org}). It could have served as its own example,
however, before the toolchain can build itself, we have to bootstrap it (that
chicken and egg problem again). And this step wouldn't serve our goal of
quickly learning what \c{build2} is about. So, instead, we will start with a
customary \i{\"Hello, World!\"} example which you won't yet be able to try
yourself (but don't worry, complete terminal output will be shown). If at the
end you find \c{build2} appealing, you can jump right to
\l{build2-toolchain-install.xhtml The \c{build2} Toolchain Installation and
Upgrade} (and, yes, there you get to run that coveted \c{bpkg build bpkg}).
Once the \c{build2} installation is complete, you can come back to the
\i{\"Hello, World!\"} example and try all of the steps for yourself.
This introduction explores the \i{consumer} side of \i{\"Hello, World!\"}.
That is, we assume that someone was kind enough to create and package the
\c{libhello} library as well as the \c{hello} program and we will learn how to
obtain and build them as well as keep up with their updates. At the end we
will also see how to write our own, \c{hello2}, program that depends on
\c{libhello}. And so, without further ado, let's begin.
The first step in using \c{bpkg} is to create a \i{configuration}. A
configuration is a directory where packages that require similar compile
settings will be built. You can create as many configurations as you want: for
different C++ compilers, targets (\c{build2} is big on cross-compiling),
debug/release, 32/64-bit, or even for different days of the week, if you are
so inclined. Say we are in the mood for a GCC 5 release build today:
\
$ mkdir hello-gcc5-release
$ cd hello-gcc5-release
$ bpkg create cxx config.cxx=g++-5 config.cxx.coptions=-O3
created new configuration in /tmp/hello-gcc5-release/
\
Or perhaps you are on Windows and prefer Visual Studio (running from the
Visual Studio Tools Command Prompt):
\
> mkdir hello-vc14-release
> cd hello-vc14-release
> bpkg create cxx config.cxx=cl config.cxx.coptions=/O2
created new configuration in C:\projects\hello-vc14-release\
\
One of the primary goals of the \c{build2} toolchain is to provide a uniform
build interface across all the platforms and compilers. While the following
examples use the \c{hello-gcc5-release} configuration and assume a UNIX-like
operation system, everything will work if you use \c{hello-vc14-release} (or
\c{hello-mingw-release}) on Windows. Just use appropriate paths, compilers,
and options.
Let's discuss that last command line: \l{bpkg-cfg-create(1) \c{bpkg create}}
is the command for creating a new configuration. As a side note, if you ever
want to get help for any \c{bpkg} command, run \c{bpkg help \i{}}. To
see the list of commands, run just \l{bpkg-help(1) \c{bpkg help}} (or see
\l{bpkg(1)}). While we are at it, if you ever want to see what \c{bpkg} is
running underneath, there is the \c{-v} (essential commands) and \c{-V} (all
commands) options. And if you really want to get under the hood, use
\l{bpkg-common-options(1) \c{--verbose }}.
After the command we have \c{cxx} which is the name of the \c{build2} build
system module. As you might have guessed, \c{cxx} provides support for the C++
compilation. By specifying this module when creating the configuration we
configure it (yes, with those \c{config.cxx.*} variables that follow) for the
entire configuration. That is, every package that we will build in this
configuration and that uses the \c{cxx} module will by default inherit these
settings.
The rest of the command line are the configuration variables for the \c{cxx}
module with \c{coptions} standing for \i{compile options} (there are also
\c{poptions} for \i{preprocess options}, \c{loptions} for \i{link options}, and
\c{libs} for extra libraries to link).
There is also the \c{c} module for the C compilation. So if we were planning
to build both C and C++ projects, then we could have run:
\
$ bpkg create c cxx ...
\
The problem, of course, is that you may not know what mix of languages those
projects (or their dependencies) might use. For example, the use of C might be
an implementation detail of a C++ library. To solve this, \c{build2} provides
another module called \c{cc} which stands for \i{C-common}. So, in this
context, instead of using the \c{c} and \c{cxx} modules directly, it's a good
idea to get into the habit of using \c{cc}:
\
$ bpkg create cc config.cxx=g++-5 config.cc.coptions=-O3
\
Notice two things about this command line: we don't need to specify the C
compiler with \c{config.c} \- \c{build2} is smart enough to figure it out
from \c{config.cxx} (or vice versa). We also used \c{config.cc.coptions}
instead of \c{config.cxx.coptions} so that the options apply to all the
C-common languages (we can still use \c{config.{c,cxx\}.*} for the
language-specific options).
Ok, configuration in hand, where can we get some packages? \c{bpkg} packages
come from \i{repositories}. A repository can be a local filesystem directory
or a remote URL. Our example packages come from their own remote \i{\"Hello,
World!\"} repository: \c{\l{https://build2.org/pkg/1/hello/stable/}} (go ahead,
browse it, I will wait).
Instead of scouring repository manifests by hand (I know you couldn't resist),
we can ask \c{bpkg} to interrogate a repository location for us:
\
$ bpkg rep-info https://build2.org/pkg/1/hello/stable
warning: authenticity of the certificate for repository build2.org/hello/stable cannot be established
certificate is for build2.org, \"Code Synthesis\"
certificate SHA256 fingerprint:
FF:DF:7D:38:67:4E:C3:82:[...]:30:56:B9:77:B9:F2:01:94
trust this certificate? [y/n]
\
The \c{bpkg} repositories are normally signed to prevent tampering with
packages. If the repository certificate is seen (in this configuration) for
the first time, \c{bpkg} will ask you to authenticate it. A good way to
authenticate a certificate is to compare the displayed fingerprint to the one
you have received earlier, for example, in an email announcement. The
repository's about page also lists the fingerprint (see the
\l{https://build2.org/pkg/hello/?about about page} for our repository). For
more details on repository signing see the \l{bpkg-repository-signing(1)} help
topic.
If we answer \i{yes}, we will see the basic repository information (its
\i{canonical name}, location, certificate subject and fingerprint) followed
by the list of available packages:
\
build2.org/hello/stable https://build2.org/pkg/1/hello/stable
CN=build2.org/O=Code Synthesis/admin@build2.org
FF:DF:7D:38:67:4E:C3:82:[...]:30:56:B9:77:B9:F2:01:94
hello/1.0.0
libhello/1.0.0
\
We can also use the repository's web interface (implemented by \c{brep}). Our
repository has one, check it out: \c{\l{https://build2.org/pkg/hello/}}.
Ok, back to the command line. If we want to use a repository as a source of
packages in our configuration, we have to first add it:
\
$ bpkg add https://build2.org/pkg/1/hello/stable
added repository build2.org/hello/stable
\
If we want to add several repositories, we just execute the \l{bpkg-rep-add(1)
\c{bpkg add}} command for each of them. Once this is done, we fetch the list of
available packages for all the added repositories:
\
$ bpkg fetch
fetching build2.org/hello/stable
2 package(s) in 1 repository(s)
\
Note that you would normally re-run the \l{bpkg-rep-fetch(1) \c{bpkg fetch}}
command after you've added another repository or to refresh the list of
available packages.
Now that \c{bpkg} knows where to get the packages, we can finally get down to
business:
\
$ bpkg build hello
build libhello/1.0.0 (required by hello)
build hello/1.0.0
continue? [Y/n]
\
Let's see what's going on here. We ran \l{bpkg-pkg-build(1) \c{bpkg build}} to
build the \c{hello} program which happens to depend on the \c{libhello}
library. So \c{bpkg} presents us with a \i{plan of action}, that is, the steps
it will have to perform in order to build us \c{hello} and then asks us to
confirm if that's what we want to do (you can add \c{--yes|-y} to skip the
confirmation). In the real-world usage the plan will be more complex, with
upgrades/downgrades, reconfigurations, etc.
Let's answer \i{yes} and see what happens:
\
libhello-1.0.0.tar.gz 100% of 2428 B 1364 kBps 00m01s
fetched libhello/1.0.0
unpacked libhello/1.0.0
hello-1.0.0.tar.gz 100% of 1057 B 20 MBps 00m01s
fetched hello/1.0.0
unpacked hello/1.0.0
configured libhello/1.0.0
configured hello/1.0.0
c++ hello-1.0.0/cxx{hello}
c++ libhello-1.0.0/hello/cxx{hello}
ld libhello-1.0.0/hello/libs{hello}
ld hello-1.0.0/exe{hello}
updated hello/1.0.0
\
While the output is mostly self-explanatory, in short, \c{bpkg} downloaded,
unpacked, and configured both packages and then proceeded to building the
\c{hello} executable which happens to require building the \c{libhello}
library. Note that the download progress may look differently on your machine
depending on which \i{fetch tool} (\c{wget}, \c{curl}, or \c{fetch}) is
used. If you ever considered giving that \c{-v} option a try, now would be a
good time. But let's first drop (\l{bpkg-pkg-drop(1) \c{bpkg drop}}) the
\c{hello} package so that we get the same build from scratch:
\
$ bpkg drop hello
following prerequisite packages were automatically built and will no longer be necessary:
libhello
drop prerequisite packages? [Y/n] y
drop hello
drop libhello
continue? [Y/n] y
disfigured hello
disfigured libhello
purged hello
purged libhello
\
Ok, ready for some \c{-v} details? Feel free to skip the following listing
if you are not interested.
\
$ bpkg build -v -y hello
fetching libhello-1.0.0.tar.gz from build2.org/hello/stable
curl ... https://build2.org/pkg/1/hello/stable/libhello-1.0.0.tar.gz
% Total % Received Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 2428 100 2428 1121 0 0:00:01 0:00:01 --:--:-- 1122
fetched libhello/1.0.0
tar -xf libhello-1.0.0.tar.gz
unpacked libhello/1.0.0
fetching hello-1.0.0.tar.gz from build2.org/hello/stable
curl ... https://build2.org/pkg/1/hello/stable/hello-1.0.0.tar.gz
% Total % Received Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 1057 100 1057 773 0 0:00:01 0:00:01 --:--:-- 772
fetched hello/1.0.0
tar -xf hello-1.0.0.tar.gz
unpacked hello/1.0.0
b -v configure(./libhello-1.0.0/)
cat >libhello-1.0.0/build/config.build
configured libhello/1.0.0
b -v configure(./hello-1.0.0/)
cat >hello-1.0.0/build/config.build
configured hello/1.0.0
hold package hello
b -v update(./hello-1.0.0/)
g++-5 -I libhello-1.0.0 -O3 -std=c++11 -o hello-1.0.0/hello.o -c hello-1.0.0/hello.cxx
g++-5 -I libhello-1.0.0 -O3 -std=c++11 -fPIC -o libhello-1.0.0/hello/hello.so.o -c libhello-1.0.0/hello/hello.cxx
g++-5 -O3 -std=c++11 -shared -o libhello-1.0.0/hello/libhello-1.0.so libhello-1.0.0/hello/hello.so.o
g++-5 -O3 -std=c++11 -o hello-1.0.0/hello hello-1.0.0/hello.o libhello-1.0.0/hello/libhello-1.0.so
updated hello/1.0.0
\
Another handy command is \l{bpkg-pkg-status(1) \c{bpkg status}}. It can be
used to examine the state of a package in the configuration. Here are a few
examples (if you absolutely must know what \c{hold_package} and \c{sys:?}
mean, check \l{bpkg-pkg-status(1)}):
\
$ bpkg status libhello
configured 1.0.0; available sys:?
$ bpkg status hello
configured 1.0.0 hold_package; available sys:?
$ bpkg drop -y hello
disfigured hello
disfigured libhello
purged hello
purged libhello
$ bpkg status hello
available 1.0.0 sys:?
$ bpkg status libfoobar
unknown
\
Let's say we got wind of a new development: the \c{libhello} author released a
new version of the library. It is such an advance in the art of \i{\"Hello,
World!\"}, it's only currently available from \c{testing}. Of course, we must
check it out.
Now, what exactly is \c{testing}? You must have noticed that the repository
location that we've been using so far ended with \c{/stable}. Quite often it is
useful to split our repository into sub-repositories or \i{sections}. For
example, to reflect the maturity of packages (say, \c{stable} and \c{testing},
as in our case) or to divide them into sub-categories (\c{misc} and \c{math})
or even some combination (\c{math/testing}). Note, however, that to \c{bpkg}
these sub-repositories or \i{sections} are just normal repositories and there
is nothing special about them.
We are impatient to try the new version so we will skip interrogating the
repository with \c{rep-info} and just add it to our configuration. After all,
we can always check with \c{status} if any upgrades are available for packages
we are interested in. Here we assume the configuration has \c{hello} built (run
\c{bpkg build -y hello} to get to that state).
\
$ bpkg add https://build2.org/pkg/1/hello/testing
added repository build2.org/hello/testing
$ bpkg fetch
fetching build2.org/hello/stable
fetching build2.org/hello/testing
5 package(s) in 2 repository(s)
\
Notice that this time we don't see any authentication-related messages or
prompts since \c{bpkg} remembered (in this configuration) that we trust the
certificate (\c{testing} naturally uses the same one as \c{stable}).
Let's see what's new:
\
$ bpkg status libhello
configured 1.0.0; available 1.1.0 sys:?
\
Ok, \c{libhello/1.1.0} is now available. How do we upgrade? We can try to
build \c{hello} again:
\
$ bpkg build -y hello
info: dir{hello-1.0.0/} is up to date
updated hello/1.0.0
\
Why did nothing happenned? Because \c{bpkg} will only upgrade (or downgrade)
to a new version if we explicitly ask it to. As things stand, all dependencies
for \c{hello} are satisfied and \c{bpkg} is happy to twiddle its thumbs. Let's
tell \c{bpkg} to build us \c{libhello} instead:
\
$ bpkg build libhello
build libformat/1.0.0 (required by libhello)
build libprint/1.0.0 (required by libhello)
upgrade libhello/1.1.0
reconfigure hello (dependent of libhello)
continue? [Y/n]
\
Ok, now we are getting somewhere. It looks like the new version of \c{libhello}
went really enterprise-grade (or is it called web-scale these days?). There are
now two new dependencies (\c{libformat} and \c{libprint}) that we will have to
build in order to upgrade. Maybe we should answer \i{no} here?
Notice also that \c{reconfigure hello} line. If you think about this, it makes
sense: we are getting a new version of \c{libhello} and \c{hello} depends on it
so it might need a chance to make some adjustments to its configuration.
Let's answer \i{yes} if only to see what happens:
\
update dependent packages? [Y/n]
\
Another question. This one has to do with that \c{reconfigure hello} line we
just talked about. If you were wondering why we were only offered to
reconfigure and not actually update the dependent package, you should know
that \c{bpkg} is a very lazy package manager, it only does what it must do,
not what might be nice to do. It must reconfigure but it doesn't really have
to update. And this could be a good thing if, for example, you have a hundred
dependents in your configuration but right now you only want to build just
those specific packages. However, quite often, you do want to keep all the
packages in your configuration up to date and \c{bpkg} graciously offers to
take care of this task. Ok, let's answer \i{yes} again:
\
...
update dependent packages? [Y/n] y
disfigured hello/1.0.0
disfigured libhello/1.0.0
libformat-1.0.0.tar.gz 100% of 1064 B 11 MBps 00m01s
fetched libformat/1.0.0
unpacked libformat/1.0.0
libprint-1.0.0.tar.gz 100% of 1040 B 9 MBps 00m01s
fetched libprint/1.0.0
unpacked libprint/1.0.0
libhello-1.1.0.tar.gz 100% of 1564 B 4672 kBps 00m01s
fetched libhello/1.1.0
unpacked libhello/1.1.0
configured libformat/1.0.0
configured libprint/1.0.0
configured libhello/1.1.0
configured hello/1.0.0
c++ libhello-1.1.0/hello/cxx{hello}
c++ libformat-1.0.0/format/cxx{format}
ld libformat-1.0.0/format/liba{format}
c++ libprint-1.0.0/print/cxx{print}
ld libprint-1.0.0/print/liba{print}
ld libhello-1.1.0/hello/liba{hello}
c++ libhello-1.1.0/hello/cxx{hello}
c++ libformat-1.0.0/format/cxx{format}
ld libformat-1.0.0/format/libs{format}
c++ libprint-1.0.0/print/cxx{print}
ld libprint-1.0.0/print/libs{print}
ld libhello-1.1.0/hello/libs{hello}
c++ libhello-1.1.0/tests/test/cxx{driver}
ld libhello-1.1.0/tests/test/exe{driver}
c++ hello-1.0.0/cxx{hello}
ld hello-1.0.0/exe{hello}
updated libhello/1.1.0
updated hello/1.0.0
\
A lot of output but nothing really new. If you were to answer \i{no} to the
\"update dependent packages?\" question above, it is easy to make sure a
package is up-to-date at a later time with the \l{bpkg-pkg-update(1) \c{bpkg
update}} command (there is also \l{bpkg-pkg-clean(1) \c{bpkg clean}}), for
example:
\
$ bpkg clean hello
rm hello-1.0.0/exe{hello}
rm hello-1.0.0/obje{hello}
cleaned hello/1.0.0
$ bpkg update hello
c++ hello-1.0.0/cxx{hello.cxx}
ld hello-1.0.0/exe{hello}
updated hello/1.0.0
\
Let's say we really don't like the direction \c{libhello} is going and would
rather stick to version \c{1.0.0}. Just like upgrades, downgrades are explicit
plus, in this case, we need to specify the version (you can also specify
the desired version for upgrades).
\
$ bpkg build libhello/1.0.0
downgrade libhello/1.0.0
reconfigure hello (dependent of libhello)
continue? [Y/n] y
update dependent packages? [Y/n] y
disfigured hello/1.0.0
disfigured libhello/1.1.0
libhello-1.0.0.tar.gz 100% of 2428 B 983 kBps 00m01s
fetched libhello/1.0.0
unpacked libhello/1.0.0
configured libhello/1.0.0
configured hello/1.0.0
following prerequisite packages were automatically built and will no longer be necessary:
libprint
libformat
drop prerequisite packages? [Y/n] y
disfigured libprint
disfigured libformat
purged libprint
purged libformat
c++ libhello-1.0.0/hello/cxx{hello}
ld libhello-1.0.0/hello/liba{hello}
c++ libhello-1.0.0/hello/cxx{hello}
ld libhello-1.0.0/hello/libs{hello}
c++ libhello-1.0.0/tests/test/cxx{driver}
ld libhello-1.0.0/tests/test/exe{driver}
c++ hello-1.0.0/cxx{hello}
ld hello-1.0.0/exe{hello}
updated libhello/1.0.0
updated hello/1.0.0
\
Notice how \c{bpkg} helpfully offered to get rid of \c{libprint} and
\c{libformat} which we won't be needing anymore. Note also that while we
can use \c{--yes|y} as an answer to all the numerous prompts, there are
also more granular options. For example, this is how we can instruct
\c{bpkg} to drop prerequisites (\c{--drop-prerequisite|-D}) but leave
dependents just reconfigured (\c{--leave-dependent|-L}):
\
$ bpkg build -D -L libhello/1.0.0
\
Ok, so all this might look nice and all, but we haven't actually seen anything
of what we've presumably built; it can all be a charade, for all we know. Can
we see some libraries and run the \c{hello} program?
There are several ways we can do this. If the package provides tests (as all
good packages should), we can run them with the \l{bpkg-pkg-test(1) \c{bpkg
test}} command:
\
$ bpkg test libhello hello
test libhello-1.0.0/tests/test/exe{driver}
test hello-1.0.0/exe{hello}
tested libhello/1.0.0
tested hello/1.0.0
\
But that doesn't quite count for seeing libraries and running programs. Well,
if you insist, let's see what's inside \c{hello-gcc5-release/}. The \c{bpkg}
configuration (this \c{hello-gcc5-release/} directory) is, in the \c{build2}
build system terms, an \i{amalgamation} \- a project that contains
\i{subprojects}. Not surprisingly, the subprojects in this amalgamation are the
packages that we've built:
\
$ ls -1F
build/
hello-1.0.0/
libhello-1.0.0/
buildfile
hello-1.0.0.tar.gz
libhello-1.0.0.tar.gz
\
And if we look inside \c{hello-1.0.0/} we will see what looks like the
\c{hello} program:
\
$ ls -1F hello-1.0.0/
build/
buildfile
hello*
hello.d
hello.cxx
hello.o
hello.o.d
manifest
test.out
version
$ hello-1.0.0/hello
usage: hello ...
$ hello-1.0.0/hello World
Hello, World!
\
The important point here is this: the \c{bpkg} configuration is not some black
box that you should never look inside of. On the contrary, it is a normal and
predictable concept of the build system and as long as you understand what you
are doing, feel free to muck around.
Another way to get hold of a package's goodies is to install it with
\l{bpkg-pkg-install(1) \c{bpkg install}}. Let's try that:
\
$ bpkg install \
config.install.root=/opt/hello \
config.install.sudo=sudo \
hello
install /opt/hello/
install /opt/hello/include/
install /opt/hello/include/hello/
install libhello-1.0.0/hello/hxx{hello}
install libhello-1.0.0/hello/hxx{export}
install /opt/hello/lib/
install libhello-1.0.0/hello/libs{hello}
install /opt/hello/bin/
install hello-1.0.0/exe{hello}
install /opt/hello/share/
install /opt/hello/share/doc/
install /opt/hello/share/doc/hello/
install hello-1.0.0/doc{version}
installed hello/1.0.0
\
The \c{config.install.sudo} value is the optional \i{sudo}-like program
that should be used to run the \c{install} program. For those feeling queasy
running \c{sudo make install}, here is your answer. If you are wondering
whether you could have specified those \c{config.install.*} values during the
configuration creation, the answer is yes, indeed!
Let's see what we've got:
\
$ tree -F /opt/hello/
/opt/hello/
├── bin/
│ └── hello*
├── include/
│ └── hello/
│ ├── export
│ └── hello
├── lib/
│ ├── libhello-1.0.so*
│ └── libhello.so -> libhello-1.0.so*
└── share/
└── doc/
└── hello/
└── version
\
We can also try to run the installed program:
\
$ /opt/hello/bin/hello World
/opt/hello/bin/hello: error while loading shared libraries: libhello-1.0.so: cannot open shared object file: No such file or directory
\
Not what we hoped to see. Note to the Windows users: this will actually work
since \c{hello-1.0.dll} will be installed into \c{bin\\}, next to the
executable; for once things are working better on Windows.
The problem is with our installation location: the runtime linker won't look
for \c{libhello-1.0.so} in \c{/opt/hello/lib} unless we somehow tell it to
(for example, using \c{LD_LIBRARY_PATH} or equivalent). There are several
ways we can resolve this. We could give up on shared libraries and link our
prerequisite libraries statically (\c{config.bin.exe.lib=static}). Or we could
use the \i{rpath} mechanism:
\
$ bpkg install \
config.install.root=/opt/hello \
config.install.sudo=sudo \
config.bin.rpath=/opt/hello/lib \
hello
ld hello-1.0.0/exe{hello}
install /opt/hello/
install /opt/hello/include/
install /opt/hello/include/hello/
install libhello-1.0.0/hello/hxx{hello}
install libhello-1.0.0/hello/hxx{export}
install /opt/hello/lib/
install libhello-1.0.0/hello/libs{hello}
install /opt/hello/bin/
install hello-1.0.0/exe{hello}
install /opt/hello/share/
install /opt/hello/share/doc/
install /opt/hello/share/doc/hello/
install hello-1.0.0/doc{version}
installed hello/1.0.0
$ /opt/hello/bin/hello World
Hello, World!
\
Notice that \c{ld} line above \- this is where our executable is re-linked
with the \c{-rpath} option.
We can also uninstall what we have installed with \l{bpkg-pkg-uninstall(1)
\c{bpkg uninstall}}:
\
$ bpkg uninstall \
config.install.root=/opt/hello \
config.install.sudo=sudo \
hello
uninstall hello-1.0.0/doc{version}
uninstall /opt/hello/share/doc/hello/
uninstall /opt/hello/share/doc/
uninstall /opt/hello/share/
uninstall hello-1.0.0/exe{hello}
uninstall /opt/hello/bin/
uninstall libhello-1.0.0/hello/libs{hello}
uninstall /opt/hello/lib/
uninstall libhello-1.0.0/hello/hxx{export}
uninstall libhello-1.0.0/hello/hxx{hello}
uninstall /opt/hello/include/hello/
uninstall /opt/hello/include/
uninstall /opt/hello/
uninstalled hello/1.0.0
$ ls /opt/hello
ls: cannot access /opt/hello: No such file or directory
\
What if we wanted to use \c{libhello} in our own project? While installing it
is always an option, this may not be convenient when we develop our code. We
may have multiple builds per project, for example, with GCC and Clang to catch
all the warnings. We may also want to make sure our application works well
with several versions of \c{libhello} (and maybe even with that heinous
\c{1.1.X}). While we can install different configurations into different
directories, it's hard to deny things are getting a bit hairy: multiple
configurations, multiple installations... I guess we will have to get our
hands into that cookie jar, I mean, configuration, again.
In fact, let's just start writing our own version of the \c{hello} program
and see how it goes:
\
$ mkdir hello2
$ cd hello2
$ cat >hello.cpp
#include
int main ()
{
hello::say (\"World\");
}
\
What build system shall we use? I can't believe you are even asking this
question...
\
$ mkdir build
$ cat >build/bootstrap.build
project = hello2 # project name
using config # config module (those config.*)
$ cat >build/root.build
cxx.std = 11 # C++ standard
using cxx # C++ module
cxx{*}: extension = cpp # C++ source file extension
$ cat >buildfile
import libs = libhello%lib{hello}
exe{hello}: cxx{hello} $libs
\
While some of this might not be crystal clear (like why do we have
\c{bootstrap.build} \i{and} \c{root.build}), I am sure you at least have a
fuzzy idea of what's going on. And that's enough for what we are after here.
Completely explaining what's going on here and, more importantly, \i{why} it's
going this way is for another time and place (the \c{build2} build system
manual).
To recap, these are the contents of our project so far:
\
$ tree -F
.
├── build/
│ ├── bootstrap.build
│ └── root.build
├── buildfile
└── hello.cpp
\
Let's try to build it and see what happens \- maybe it will magically work
(\l{b(1)} is the \c{build2} build system driver).
\
$ b config.cxx=g++-5
error: unable to import target libhello%lib{hello}
info: consider explicitly specifying its project out_root via the config.import.libhello command line variable
info: while applying rule cxx.compile to update obje{hello}
info: while applying rule cxx.link to update exe{hello}
info: while applying rule alias to update dir{./}
\
No magic, unfortunately (or fortunately). But we got a hint: looks like we
need to tell \c{build2} where \c{libhello} is using
\c{config.import.libhello}. Without fretting too much about what exactly
\c{out_root} means, let's point \c{build2} to our \c{bpkg} configuration and
see what happens. After all, that's where, more or less, our \i{out}-put for
\c{libhello} is.
\
$ b config.cxx=g++-5 \
config.import.libhello=/tmp/hello-gcc5-release
c++ cxx{hello}
ld exe{hello}
\
Almost magic. Let's see what we've got:
\
$ tree -F
.
├── build/
│ ├── bootstrap.build
│ └── root.build
├── buildfile
├── hello*
├── hello.d
├── hello.cpp
├── hello.o
└── hello.o.d
$ ./hello
Hello, World!
\
Let's change something in our source code and try to update:
\
$ touch hello.cpp
$ b
error: unable to import target libhello%lib{hello}
info: consider explicitly specifying its project out_root via the config.import.libhello command line variable
info: while applying rule cxx.compile to update obje{hello}
info: while applying rule cxx.link to update exe{hello}
info: while applying rule alias to update dir{./}
\
Looks like we have to keep repeating those \c{config.*} values and who wants
that? To get rid of this annoyance we have to make our configuration
\i{permanent}. Also, seeing that we plan to have several of them (GCC/Clang,
different version of \c{libhello}), it makes sense to create them \i{out of
source tree}. Let's get to it:
\
$ cd ..
$ mkdir hello2-gcc5-release
$ ls -1F
hello2/
hello2-gcc5-release/
$ b config.cxx=g++-5 \
config.cc.coptions=-O3 \
config.import.libhello=/tmp/hello-gcc5-release \
'configure(hello2/@hello2-gcc5-release/)'
mkdir -p hello2-gcc5-release/build/
save hello2-gcc5-release/build/config.build
\
Translated, \c{configure(hello2/@hello2-gcc5-release/)} means \i{\"configure
the \c{hello2/} source directory in the \c{hello2-gcc5-release/} output
directory\"}. In \c{build2} this \i{source directory} is called \c{src_root}
and \i{output directory} \- \c{out_root}. Hm, we've already heard \c{out_root}
mentioned somewhere before...
Once the configuration is saved, we can develop our project without any
annoyance:
\
$ b hello2-gcc5-release/
c++ hello2/cxx{hello}
ld hello2-gcc5-release/exe{hello}
$ cd hello2-gcc5-release/
$ b
info: dir{./} is up to date
$ b clean
rm exe{hello}
rm obje{hello}
$ b -v
g++-5 -I/tmp/hello-gcc5-release/libhello-1.0.0 -O3 -std=c++11 -o hello.o -c ../hello2/hello.cpp
g++-5 -O3 -std=c++11 -o hello hello.o /tmp/hello-gcc5-release/libhello-1.0.0/hello/libhello-1.0.so
\
Some of you might have noticed that \c{hello2-gcc5-release/} and
\c{/tmp/hello-gcc5-release/} look awfully similar and are now wondering if we
could instead build \c{hello2} \i{inside} \c{/tmp/hello-gcc5-release/}? I am
glad you've asked. In fact, we can just do:
\
$ cd ..
$ ls -1F
hello2/
hello2-gcc5-release/
$ b 'configure(hello2/@/tmp/hello-gcc5-release/hello2/)'
mkdir -p /tmp/hello-gcc5-release/hello2/build/
save /tmp/hello-gcc5-release/hello2/build/config.build
$ b /tmp/hello-gcc5-release/hello2/
c++ hello2/cxx{hello}@/tmp/hello-gcc5-release/hello2/
ld /tmp/hello-gcc5-release/hello2/exe{hello}
\
Now that might seem like magic, but it's actually pretty logical. Why don't we
need to specify any of the \c{config.c*} values this time? Because they are
inherited from those specified for \c{/tmp/hello-gcc5-release} when we created
the configuration with \c{bpkg create}. What about \c{config.import.libhello},
don't we need at least that? Not really \- \c{libhello} will be found
automatically since it is part of the same amalgamation.
Of course, \c{bpkg} has no idea \c{hello2} is now part of its configuration:
\
$ bpkg status -d /tmp/hello-gcc5-release/ hello2
unknown
\
This is what I meant when I said you can muck around in \c{bpkg}'s back yard as
long as you understand the implications.
But is there a way to make \c{bpkg} aware of our little project? You seem to
really have all the right questions today. Actually, there is a very good
reason why we would want that: if we upgrade \c{libhello} we would want
\c{bpkg} to automatically reconfigure our project. As it is now, we will have
to remember and do it ourselves.
The only way to make \c{bpkg} aware of \c{hello2} is to turn it from merely a
\c{build2} \i{project} into a \c{build2} \i{package}. While the topic of
packaging is also for another time and place (the \c{build2} package manager
manual), we can get away with something as simple as this:
\
$ cat >hello2/manifest
: 1
name: hello2
version: 1.0.0
summary: Improved \"Hello World\" program
license: proprietary
url: http://example.org/hello2
email: hello2@example.org
depends: libhello >= 1.0.0
\
For our purposes, the only really important value in this manifest is
\c{depends} since it tells \c{bpkg} which package(s) we need. Let's give it a
try. But first we will clean up our previous attempt at building \c{hello2}
inside \c{/tmp/hello-gcc5-release/}:
\
$ b '{clean disfigure}(/tmp/hello-gcc5-release/hello2/)'
rm /tmp/hello-gcc5-release/hello2/exe{hello}
rm /tmp/hello-gcc5-release/hello2/obje{hello}
rm /tmp/hello-gcc5-release/hello2/build/config.build
rmdir /tmp/hello-gcc5-release/hello2/
\
Next, we use the \l{bpkg-pkg-build(1) \c{bpkg build}} command but instead of
giving it a package name like we did before, we will point it to our \c{hello2}
package directory (\c{bpkg} can fetch packages or it can build local package
archives or package directories):
\
$ bpkg build -d /tmp/hello-gcc5-release/ ./hello2/
build hello2/1.0.0
continue? [Y/n] y
unpacked hello2/1.0.0
configured hello2/1.0.0
c++ hello2/cxx{hello}@/tmp/hello-gcc5-release/hello2-1.0.0/
ld /tmp/hello-gcc5-release/hello2-1.0.0/exe{hello}
updated hello2/1.0.0
\
Let's upgrade \c{libhello} and see what happens:
\
$ bpkg build -d /tmp/hello-gcc5-release/ -L libhello
build libformat/1.0.0 (required by libhello)
build libprint/1.0.0 (required by libhello)
upgrade libhello/1.1.0
reconfigure hello2 (dependent of libhello)
continue? [Y/n] y
disfigured hello2/1.0.0
disfigured libhello/1.0.0
[ ... fetching & unpacking ... ]
configured libformat/1.0.0
configured libprint/1.0.0
configured libhello/1.1.0
configured hello2/1.0.0
[ ... updating libprint, libformat, and libhello ... ]
updated libhello/1.1.0
\
As promised, \c{hello2} got reconfigured (it didn't get updated because of the
\c{-L} option). We can now update it and give it a try:
\
$ bpkg update -d /tmp/hello-gcc5-release/ hello2
c++ hello2/cxx{hello}@/tmp/hello-gcc5-release/hello2-1.0.0/
ld /tmp/hello-gcc5-release/hello2-1.0.0/exe{hello}
updated hello2/1.0.0
$ /tmp/hello-gcc5-release/hello2-1.0.0/hello
Hello, World!
\
To finish off, let's see how hard it will be to get a Clang build going:
\
$ cd /tmp
$ mkdir hello-clang36-release
$ cd hello-clang36-release
$ bpkg create cc config.cxx=clang++-3.6 config.cc.coptions=-O3
created new configuration in /tmp/hello-clang36-release/
$ bpkg add https://build2.org/pkg/1/hello/testing
added repository build2.org/hello/testing
$ bpkg fetch
fetching build2.org/hello/testing
[... certificate authentication ...]
fetching build2.org/hello/stable (complements build2.org/hello/testing)
5 package(s) in 2 repository(s)
$ bpkg build libhello/1.0.0 path/to/hello2/
build libhello/1.0.0
build hello2/1.0.0
continue? [Y/n] y
libhello-1.0.0.tar.gz 100% of 2428 B 983 kBps 00m01s
fetched libhello/1.0.0
unpacked libhello/1.0.0
unpacked hello2/1.0.0
configured libhello/1.0.0
configured hello2/1.0.0
c++ libhello-1.0.0/hello/cxx{hello}
ld libhello-1.0.0/hello/liba{hello}
c++ libhello-1.0.0/hello/cxx{hello}
ld libhello-1.0.0/hello/libs{hello}
c++ libhello-1.0.0/tests/test/cxx{driver}
ld libhello-1.0.0/tests/test/exe{driver}
c++ /path/to/hello2/cxx{hello}@hello2-1.0.0/
ld hello2-1.0.0/exe{hello}
updated libhello/1.0.0
updated hello2/1.0.0
\
Are you still there? Ok, one last example. Let's see how hard it is to
cross-compile.
\
$ mkdir hello-mingw64
$ cd hello-mingw64
$ bpkg create cc config.cxx=x86_64-w64-mingw32-g++
created new configuration in /tmp/hello-mingw64/
$ bpkg add https://build2.org/pkg/1/hello/stable
added repository build2.org/hello/stable
$ bpkg fetch
fetching build2.org/hello/stable
[... certificate authentication ...]
2 package(s) in 1 repository(s)
$ bpkg build -y hello
libhello-1.0.0.tar.gz 100% of 2428 B 983 kBps 00m01s
fetched libhello/1.0.0
unpacked libhello/1.0.0
hello-1.0.0.tar.gz 100% of 1057 B 6882 kBps 00m01s
fetched hello/1.0.0
unpacked hello/1.0.0
configured libhello/1.0.0
configured hello/1.0.0
c++ hello-1.0.0/cxx{hello}
c++ libhello-1.0.0/hello/cxx{hello}
ld libhello-1.0.0/hello/libs{hello}
ld hello-1.0.0/exe{hello}
updated hello/1.0.0
$ wine hello-1.0.0/hello.exe Windows
Hello, Windows!
\
In fact, on a properly setup GNU/Linux machine (that automatically uses
\c{wine} as an \c{.exe} interpreter) we can even run tests:
\
$ bpkg test libhello hello
c++ libhello-1.0.0/tests/test/cxx{driver}
ld libhello-1.0.0/tests/test/exe{driver}
test libhello-1.0.0/tests/test/exe{driver}
test hello-1.0.0/exe{hello}
tested libhello/1.0.0
tested hello/1.0.0
\
"