nixpkgs/doc/cross-compilation.xml
John Ericson 8ba4db0f11
manual: Fix typos
Thanks @matthewbauer!

Co-Authored-By: Ericson2314 <git@JohnEricson.me>
2019-03-25 17:50:13 -04:00

679 lines
29 KiB
XML

<chapter xmlns="http://docbook.org/ns/docbook"
xmlns:xlink="http://www.w3.org/1999/xlink"
xml:id="chap-cross">
<title>Cross-compilation</title>
<section xml:id="sec-cross-intro">
<title>Introduction</title>
<para>
"Cross-compilation" means compiling a program on one machine for another
type of machine. For example, a typical use of cross-compilation is to
compile programs for embedded devices. These devices often don't have the
computing power and memory to compile their own programs. One might think
that cross-compilation is a fairly niche concern. However, there are
significant advantages to rigorously distinguishing between build-time and
run-time environments! Significant, because the benefits apply even when one
is developing and deploying on the same machine. Nixpkgs is increasingly
adopting the opinion that packages should be written with cross-compilation
in mind, and nixpkgs should evaluate in a similar way (by minimizing
cross-compilation-specific special cases) whether or not one is
cross-compiling.
</para>
<para>
This chapter will be organized in three parts. First, it will describe the
basics of how to package software in a way that supports cross-compilation.
Second, it will describe how to use Nixpkgs when cross-compiling. Third, it
will describe the internal infrastructure supporting cross-compilation.
</para>
</section>
<!--============================================================-->
<section xml:id="sec-cross-packaging">
<title>Packaging in a cross-friendly manner</title>
<section xml:id="ssec-cross-platform-parameters">
<title>Platform parameters</title>
<para>
Nixpkgs follows the
<link
xlink:href="https://gcc.gnu.org/onlinedocs/gccint/Configure-Terms.html">conventions
of GNU autoconf</link>. We distinguish between 3 types of platforms when
building a derivation: <wordasword>build</wordasword>,
<wordasword>host</wordasword>, and <wordasword>target</wordasword>. In
summary, <wordasword>build</wordasword> is the platform on which a package
is being built, <wordasword>host</wordasword> is the platform on which it
will run. The third attribute, <wordasword>target</wordasword>, is relevant
only for certain specific compilers and build tools.
</para>
<para>
In Nixpkgs, these three platforms are defined as attribute sets under the
names <literal>buildPlatform</literal>, <literal>hostPlatform</literal>,
and <literal>targetPlatform</literal>. They are always defined as
attributes in the standard environment. That means one can access them
like:
<programlisting>{ stdenv, fooDep, barDep, .. }: ...stdenv.buildPlatform...</programlisting>
.
</para>
<variablelist>
<varlistentry>
<term>
<varname>buildPlatform</varname>
</term>
<listitem>
<para>
The "build platform" is the platform on which a package is built. Once
someone has a built package, or pre-built binary package, the build
platform should not matter and can be ignored.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<varname>hostPlatform</varname>
</term>
<listitem>
<para>
The "host platform" is the platform on which a package will be run. This
is the simplest platform to understand, but also the one with the worst
name.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<varname>targetPlatform</varname>
</term>
<listitem>
<para>
The "target platform" attribute is, unlike the other two attributes, not
actually fundamental to the process of building software. Instead, it is
only relevant for compatibility with building certain specific compilers
and build tools. It can be safely ignored for all other packages.
</para>
<para>
The build process of certain compilers is written in such a way that the
compiler resulting from a single build can itself only produce binaries
for a single platform. The task of specifying this single "target
platform" is thus pushed to build time of the compiler. The root cause
of this is that the compiler (which will be run on the host) and the
standard library/runtime (which will be run on the target) are built by
a single build process.
</para>
<para>
There is no fundamental need to think about a single target ahead of
time like this. If the tool supports modular or pluggable backends, both
the need to specify the target at build time and the constraint of
having only a single target disappear. An example of such a tool is
LLVM.
</para>
<para>
Although the existence of a "target platfom" is arguably a historical
mistake, it is a common one: examples of tools that suffer from it are
GCC, Binutils, GHC and Autoconf. Nixpkgs tries to avoid sharing in the
mistake where possible. Still, because the concept of a target platform
is so ingrained, it is best to support it as is.
</para>
</listitem>
</varlistentry>
</variablelist>
<para>
The exact schema these fields follow is a bit ill-defined due to a long and
convoluted evolution, but this is slowly being cleaned up. You can see
examples of ones used in practice in
<literal>lib.systems.examples</literal>; note how they are not all very
consistent. For now, here are few fields can count on them containing:
</para>
<variablelist>
<varlistentry>
<term>
<varname>system</varname>
</term>
<listitem>
<para>
This is a two-component shorthand for the platform. Examples of this
would be "x86_64-darwin" and "i686-linux"; see
<literal>lib.systems.doubles</literal> for more. The first component
corresponds to the CPU architecture of the platform and the second to
the operating system of the platform (<literal>[cpu]-[os]</literal>).
This format has built-in support in Nix, such as the
<varname>builtins.currentSystem</varname> impure string.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<varname>config</varname>
</term>
<listitem>
<para>
This is a 3- or 4- component shorthand for the platform. Examples of
this would be <literal>x86_64-unknown-linux-gnu</literal> and
<literal>aarch64-apple-darwin14</literal>. This is a standard format
called the "LLVM target triple", as they are pioneered by LLVM. In the
4-part form, this corresponds to
<literal>[cpu]-[vendor]-[os]-[abi]</literal>. This format is strictly
more informative than the "Nix host double", as the previous format
could analogously be termed. This needs a better name than
<varname>config</varname>!
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<varname>parsed</varname>
</term>
<listitem>
<para>
This is a Nix representation of a parsed LLVM target triple with
white-listed components. This can be specified directly, or actually
parsed from the <varname>config</varname>. See
<literal>lib.systems.parse</literal> for the exact representation.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<varname>libc</varname>
</term>
<listitem>
<para>
This is a string identifying the standard C library used. Valid
identifiers include "glibc" for GNU libc, "libSystem" for Darwin's
Libsystem, and "uclibc" for µClibc. It should probably be refactored to
use the module system, like <varname>parse</varname>.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<varname>is*</varname>
</term>
<listitem>
<para>
These predicates are defined in <literal>lib.systems.inspect</literal>,
and slapped onto every platform. They are superior to the ones in
<varname>stdenv</varname> as they force the user to be explicit about
which platform they are inspecting. Please use these instead of those.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>
<varname>platform</varname>
</term>
<listitem>
<para>
This is, quite frankly, a dumping ground of ad-hoc settings (it's an
attribute set). See <literal>lib.systems.platforms</literal> for
examples—there's hopefully one in there that will work verbatim for
each platform that is working. Please help us triage these flags and
give them better homes!
</para>
</listitem>
</varlistentry>
</variablelist>
</section>
<section xml:id="ssec-cross-dependency-categorization">
<title>Theory of dependency categorization</title>
<note>
<para>
This is a rather philosophical description that isn't very
Nixpkgs-specific. For an overview of all the relevant attributes given to
<varname>mkDerivation</varname>, see
<xref
linkend="ssec-stdenv-dependencies"/>. For a description of how
everything is implemented, see
<xref linkend="ssec-cross-dependency-implementation" />.
</para>
</note>
<para>
In this section we explore the relationship between both runtime and
build-time dependencies and the 3 Autoconf platforms.
</para>
<para>
A run time dependency between two packages requires that their host
platforms match. This is directly implied by the meaning of "host platform"
and "runtime dependency": The package dependency exists while both packages
are running on a single host platform.
</para>
<para>
A build time dependency, however, has a shift in platforms between the
depending package and the depended-on package. "build time dependency"
means that to build the depending package we need to be able to run the
depended-on's package. The depending package's build platform is therefore
equal to the depended-on package's host platform.
</para>
<para>
If both the dependency and depending packages aren't compilers or other
machine-code-producing tools, we're done. And indeed
<varname>buildInputs</varname> and <varname>nativeBuildInputs</varname>
have covered these simpler build-time and run-time (respectively) changes
for many years. But if the dependency does produce machine code, we might
need to worry about its target platform too. In principle, that target
platform might be any of the depending package's build, host, or target
platforms, but we prohibit dependencies from a "later" platform to an
earlier platform to limit confusion because we've never seen a legitimate
use for them.
</para>
<para>
Finally, if the depending package is a compiler or other
machine-code-producing tool, it might need dependencies that run at "emit
time". This is for compilers that (regrettably) insist on being built
together with their source langauges' standard libraries. Assuming build !=
host != target, a run-time dependency of the standard library cannot be run
at the compiler's build time or run time, but only at the run time of code
emitted by the compiler.
</para>
<para>
Putting this all together, that means we have dependencies in the form
"host → target", in at most the following six combinations:
<table>
<caption>Possible dependency types</caption>
<thead>
<tr>
<th>Dependency's host platform</th>
<th>Dependency's target platform</th>
</tr>
</thead>
<tbody>
<tr>
<td>build</td>
<td>build</td>
</tr>
<tr>
<td>build</td>
<td>host</td>
</tr>
<tr>
<td>build</td>
<td>target</td>
</tr>
<tr>
<td>host</td>
<td>host</td>
</tr>
<tr>
<td>host</td>
<td>target</td>
</tr>
<tr>
<td>target</td>
<td>target</td>
</tr>
</tbody>
</table>
</para>
<para>
Some examples will make this table clearer. Suppose there's some package
that is being built with a <literal>(build, host, target)</literal>
platform triple of <literal>(foo, bar, baz)</literal>. If it has a
build-time library dependency, that would be a "host → build" dependency
with a triple of <literal>(foo, foo, *)</literal> (the target platform is
irrelevant). If it needs a compiler to be built, that would be a "build →
host" dependency with a triple of <literal>(foo, foo, *)</literal> (the
target platform is irrelevant). That compiler, would be built with another
compiler, also "build → host" dependency, with a triple of <literal>(foo,
foo, foo)</literal>.
</para>
</section>
<section xml:id="ssec-cross-cookbook">
<title>Cross packaging cookbook</title>
<para>
Some frequently encountered problems when packaging for cross-compilation
should be answered here. Ideally, the information above is exhaustive, so
this section cannot provide any new information, but it is ludicrous and
cruel to expect everyone to spend effort working through the interaction of
many features just to figure out the same answer to the same common
problem. Feel free to add to this list!
</para>
<qandaset>
<qandaentry xml:id="cross-qa-build-c-program-in-build-environment">
<question>
<para>
What if my package's build system needs to build a C program to be run
under the build environment?
</para>
</question>
<answer>
<para>
<programlisting>depsBuildBuild = [ buildPackages.stdenv.cc ];</programlisting>
Add it to your <function>mkDerivation</function> invocation.
</para>
</answer>
</qandaentry>
<qandaentry xml:id="cross-qa-fails-to-find-ar">
<question>
<para>
My package fails to find <command>ar</command>.
</para>
</question>
<answer>
<para>
Many packages assume that an unprefixed <command>ar</command> is
available, but Nix doesn't provide one. It only provides a prefixed one,
just as it only does for all the other binutils programs. It may be
necessary to patch the package to fix the build system to use a prefixed
`ar`.
</para>
</answer>
</qandaentry>
<qandaentry xml:id="cross-testsuite-runs-host-code">
<question>
<para>
My package's testsuite needs to run host platform code.
</para>
</question>
<answer>
<para>
<programlisting>doCheck = stdenv.hostPlatform != stdenv.buildPlatfrom;</programlisting>
Add it to your <function>mkDerivation</function> invocation.
</para>
</answer>
</qandaentry>
</qandaset>
</section>
</section>
<!--============================================================-->
<section xml:id="sec-cross-usage">
<title>Cross-building packages</title>
<para>
Nixpkgs can be instantiated with <varname>localSystem</varname> alone, in
which case there is no cross-compiling and everything is built by and for
that system, or also with <varname>crossSystem</varname>, in which case
packages run on the latter, but all building happens on the former. Both
parameters take the same schema as the 3 (build, host, and target) platforms
defined in the previous section. As mentioned above,
<literal>lib.systems.examples</literal> has some platforms which are used as
arguments for these parameters in practice. You can use them
programmatically, or on the command line:
<programlisting>
nix-build &lt;nixpkgs&gt; --arg crossSystem '(import &lt;nixpkgs/lib&gt;).systems.examples.fooBarBaz' -A whatever</programlisting>
</para>
<note>
<para>
Eventually we would like to make these platform examples an unnecessary
convenience so that
<programlisting>
nix-build &lt;nixpkgs&gt; --arg crossSystem '{ config = "&lt;arch&gt;-&lt;os&gt;-&lt;vendor&gt;-&lt;abi&gt;"; }' -A whatever</programlisting>
works in the vast majority of cases. The problem today is dependencies on
other sorts of configuration which aren't given proper defaults. We rely on
the examples to crudely to set those configuration parameters in some
vaguely sane manner on the users behalf. Issue
<link xlink:href="https://github.com/NixOS/nixpkgs/issues/34274">#34274</link>
tracks this inconvenience along with its root cause in crufty configuration
options.
</para>
</note>
<para>
While one is free to pass both parameters in full, there's a lot of logic to
fill in missing fields. As discussed in the previous section, only one of
<varname>system</varname>, <varname>config</varname>, and
<varname>parsed</varname> is needed to infer the other two. Additionally,
<varname>libc</varname> will be inferred from <varname>parse</varname>.
Finally, <literal>localSystem.system</literal> is also
<emphasis>impurely</emphasis> inferred based on the platform evaluation
occurs. This means it is often not necessary to pass
<varname>localSystem</varname> at all, as in the command-line example in the
previous paragraph.
</para>
<note>
<para>
Many sources (manual, wiki, etc) probably mention passing
<varname>system</varname>, <varname>platform</varname>, along with the
optional <varname>crossSystem</varname> to nixpkgs: <literal>import
&lt;nixpkgs&gt; { system = ..; platform = ..; crossSystem = ..;
}</literal>. Passing those two instead of <varname>localSystem</varname> is
still supported for compatibility, but is discouraged. Indeed, much of the
inference we do for these parameters is motivated by compatibility as much
as convenience.
</para>
</note>
<para>
One would think that <varname>localSystem</varname> and
<varname>crossSystem</varname> overlap horribly with the three
<varname>*Platforms</varname> (<varname>buildPlatform</varname>,
<varname>hostPlatform,</varname> and <varname>targetPlatform</varname>; see
<varname>stage.nix</varname> or the manual). Actually, those identifiers are
purposefully not used here to draw a subtle but important distinction: While
the granularity of having 3 platforms is necessary to properly *build*
packages, it is overkill for specifying the user's *intent* when making a
build plan or package set. A simple "build vs deploy" dichotomy is adequate:
the sliding window principle described in the previous section shows how to
interpolate between the these two "end points" to get the 3 platform triple
for each bootstrapping stage. That means for any package a given package
set, even those not bound on the top level but only reachable via
dependencies or <varname>buildPackages</varname>, the three platforms will
be defined as one of <varname>localSystem</varname> or
<varname>crossSystem</varname>, with the former replacing the latter as one
traverses build-time dependencies. A last simple difference is that
<varname>crossSystem</varname> should be null when one doesn't want to
cross-compile, while the <varname>*Platform</varname>s are always non-null.
<varname>localSystem</varname> is always non-null.
</para>
</section>
<!--============================================================-->
<section xml:id="sec-cross-infra">
<title>Cross-compilation infrastructure</title>
<section xml:id="ssec-cross-dependency-implementation">
<title>Implementation of dependencies</title>
<para>
The categorizes of dependencies developed in
<xref
linkend="ssec-cross-dependency-categorization"/> are specified as
lists of derivations given to <varname>mkDerivation</varname>, as
documented in <xref linkend="ssec-stdenv-dependencies"/>. In short,
each list of dependencies for "host → target" of "foo → bar" is called
<varname>depsFooBar</varname>, with exceptions for backwards
compatibility that <varname>depsBuildHost</varname> is instead called
<varname>nativeBuildInputs</varname> and <varname>depsHostTarget</varname>
is instead called <varname>buildInputs</varname>. Nixpkgs is now structured
so that each <varname>depsFooBar</varname> is automatically taken from
<varname>pkgsFooBar</varname>. (These <varname>pkgsFooBar</varname>s are
quite new, so there is no special case for
<varname>nativeBuildInputs</varname> and <varname>buildInputs</varname>.)
For example, <varname>pkgsBuildHost.gcc</varname> should be used at
build-time, while <varname>pkgsHostTarget.gcc</varname> should be used at
run-time.
</para>
<para>
Now, for most of Nixpkgs's history, there were no
<varname>pkgsFooBar</varname> attributes, and most packages have not been
refactored to use it explicitly. Prior to those, there were just
<varname>buildPackages</varname>, <varname>pkgs</varname>, and
<varname>targetPackages</varname>. Those are now redefined as aliases to
<varname>pkgsBuildHost</varname>, <varname>pkgsHostTarget</varname>, and
<varname>pkgsTargetTarget</varname>. It is acceptable, even
recommended, to use them for libraries to show that the host platform is
irrelevant.
</para>
<para>
But before that, there was just <varname>pkgs</varname>, even though both
<varname>buildInputs</varname> and <varname>nativeBuildInputs</varname>
existed. [Cross barely worked, and those were implemented with some hacks
on <varname>mkDerivation</varname> to override dependencies.] What this
means is the vast majority of packages do not use any explicit package set
to populate their dependencies, just using whatever
<varname>callPackage</varname> gives them even if they do correctly sort
their dependencies into the multiple lists described above. And indeed,
asking that users both sort their dependencies, <emphasis>and</emphasis>
take them from the right attribute set, is both too onerous and redundant,
so the recommended approach (for now) is to continue just categorizing by
list and not using an explicit package set.
</para>
<para>
To make this work, we "splice" together the six
<varname>pkgsFooBar</varname> package sets and have
<varname>callPackage</varname> actually take its arguments from that. This
is currently implemented in <filename>pkgs/top-level/splice.nix</filename>.
<varname>mkDerivation</varname> then, for each dependency attribute, pulls
the right derivation out from the splice. This splicing can be skipped when
not cross-compiling as the package sets are the same, but still is a bit
slow for cross-compiling. We'd like to do something better, but haven't
come up with anything yet.
</para>
</section>
<section xml:id="ssec-bootstrapping">
<title>Bootstrapping</title>
<para>
Each of the package sets described above come from a single bootstrapping
stage. While <filename>pkgs/top-level/default.nix</filename>, coordinates
the composition of stages at a high level,
<filename>pkgs/top-level/stage.nix</filename> "ties the knot" (creates the
fixed point) of each stage. The package sets are defined per-stage however,
so they can be thought of as edges between stages (the nodes) in a graph.
Compositions like <literal>pkgsBuildTarget.targetPackages</literal> can be
thought of as paths to this graph.
</para>
<para>
While there are many package sets, and thus many edges, the stages can also
be arranged in a linear chain. In other words, many of the edges are
redundant as far as connectivity is concerned. This hinges on the type of
bootstrapping we do. Currently for cross it is:
<orderedlist>
<listitem>
<para>
<literal>(native, native, native)</literal>
</para>
</listitem>
<listitem>
<para>
<literal>(native, native, foreign)</literal>
</para>
</listitem>
<listitem>
<para>
<literal>(native, foreign, foreign)</literal>
</para>
</listitem>
</orderedlist>
In each stage, <varname>pkgsBuildHost</varname> refers the the previous
stage, <varname>pkgsBuildBuild</varname> refers to the one before that, and
<varname>pkgsHostTarget</varname> refers to the current one, and
<varname>pkgsTargetTarget</varname> refers to the next one. When there is
no previous or next stage, they instead refer to the current stage. Note
how all the invariants regarding the mapping between dependency and depending
packages' build host and target platforms are preserved.
<varname>pkgsBuildTarget</varname> and <varname>pkgsHostHost</varname> are
more complex in that the stage fitting the requirements isn't always a
fixed chain of "prevs" and "nexts" away (modulo the "saturating"
self-references at the ends). We just special case each instead. All the primary
edges are implemented is in <filename>pkgs/stdenv/booter.nix</filename>,
and secondarily aliases in <filename>pkgs/top-level/stage.nix</filename>.
</para>
<note>
<para>
Note the native stages are bootstrapped in legacy ways that predate the
current cross implementation. This is why the the bootstrapping stages
leading up to the final stages are ignored inthe previous paragraph.
</para>
</note>
<para>
If one looks at the 3 platform triples, one can see that they overlap such
that one could put them together into a chain like:
<programlisting>
(native, native, native, foreign, foreign)
</programlisting>
If one imagines the saturating self references at the end being replaced
with infinite stages, and then overlays those platform triples, one ends up
with the infinite tuple:
<programlisting>
(native..., native, native, native, foreign, foreign, foreign...)
</programlisting>
On can then imagine any sequence of platforms such that there are bootstrap
stages with their 3 platforms determined by "sliding a window" that is the
3 tuple through the sequence. This was the original model for
bootstrapping. Without a target platform (assume a better world where all
compilers are multi-target and all standard libraries are built in their
own derivation), this is sufficient. Conversely if one wishes to cross
compile "faster", with a "Canadian Cross" bootstraping stage where
<literal>build != host != target</literal>, more bootstrapping stages are
needed since no sliding window providess the pesky
<varname>pkgsBuildTarget</varname> package set since it skips the Canadian
cross stage's "host".
</para>
<note>
<para>
It is much better to refer to <varname>buildPackages</varname> than
<varname>targetPackages</varname>, or more broadly package sets that do
not mention "target". There are three reasons for this.
</para>
<para>
First, it is because bootstrapping stages do not have a unique
<varname>targetPackages</varname>. For example a <literal>(x86-linux,
x86-linux, arm-linux)</literal> and <literal>(x86-linux, x86-linux,
x86-windows)</literal> package set both have a <literal>(x86-linux,
x86-linux, x86-linux)</literal> package set. Because there is no canonical
<varname>targetPackages</varname> for such a native (<literal>build ==
host == target</literal>) package set, we set their
<varname>targetPackages</varname>
</para>
<para>
Second, it is because this is a frequent source of hard-to-follow
"infinite recursions" / cycles. When only package sets that don't mention
target are used, the package set forms a directed acyclic graph. This
means that all cycles that exist are confined to one stage. This means
they are a lot smaller, and easier to follow in the code or a backtrace. It
also means they are present in native and cross builds alike, and so more
likely to be caught by CI and other users.
</para>
<para>
Thirdly, it is because everything target-mentioning only exists to
accommodate compilers with lousy build systems that insist on the compiler
itself and standard library being built together. Of course that is bad
because bigger derivations means longer rebuilds. It is also problematic because
it tends to make the standard libraries less like other libraries than
they could be, complicating code and build systems alike. Because of the
other problems, and because of these innate disadvantages, compilers ought
to be packaged another way where possible.
</para>
</note>
<note>
<para>
If one explores Nixpkgs, they will see derivations with names like
<literal>gccCross</literal>. Such <literal>*Cross</literal> derivations is
a holdover from before we properly distinguished between the host and
target platforms—the derivation with "Cross" in the name covered the
<literal>build = host != target</literal> case, while the other covered
the <literal>host = target</literal>, with build platform the same or not
based on whether one was using its <literal>.nativeDrv</literal> or
<literal>.crossDrv</literal>. This ugliness will disappear soon.
</para>
</note>
</section>
</section>
</chapter>