This chapter shows you how to write Nix expressions, which are the things that tell Nix how to build packages. It starts with a simple example (a Nix expression for GNU Hello), and then moves on to a more in-depth look at the Nix expression language.

This section shows how to add and test the GNU Hello package to the Nix Packages collection. Hello is a program that prints out the text Hello, world!. To add a package to the Nix Packages collection, you generally need to do three things: Write a Nix expression for the package. This is a file that describes all the inputs involved in building the package, such as dependencies, sources, and so on. Write a builder. This is a shell scriptIn fact, it can be written in any language, but typically it's a bash shell script. that actually builds the package from the inputs. Add the package to the file pkgs/top-level/all-packages.nix. The Nix expression written in the first step is a function; it requires other packages in order to build it. In this step you put it all together, i.e., you call the function with the right arguments to build the actual package.

{stdenv, fetchurl, perl}: stdenv.mkDerivation { name = "hello-2.1.1"; builder = ./builder.sh; src = fetchurl { url = ftp://ftp.nluug.nl/pub/gnu/hello/hello-2.1.1.tar.gz; md5 = "70c9ccf9fac07f762c24f2df2290784d"; }; inherit perl; } shows a Nix expression for GNU Hello. It's actually already in the Nix Packages collection in pkgs/applications/misc/hello/ex-1/default.nix. It is customary to place each package in a separate directory and call the single Nix expression in that directory default.nix. The file has the following elements (referenced from the figure by number): This states that the expression is a function that expects to be called with three arguments: stdenv, fetchurl, and perl. They are needed to build Hello, but we don't know how to build them here; that's why they are function arguments. stdenv is a package that is used by almost all Nix Packages packages; it provides a standard environment consisting of the things you would expect in a basic Unix environment: a C/C++ compiler (GCC, to be precise), the Bash shell, fundamental Unix tools such as cp, grep, tar, etc. fetchurl is a function that downloads files. perl is the Perl interpreter. Nix functions generally have the form {x, y, ..., z}: e where x, y, etc. are the names of the expected arguments, and where e is the body of the function. So here, the entire remainder of the file is the body of the function; when given the required arguments, the body should describe how to build an instance of the Hello package. So we have to build a package. Building something from other stuff is called a derivation in Nix (as opposed to sources, which are built by humans instead of computers). We perform a derivation by calling stdenv.mkDerivation. mkDerivation is a function provided by stdenv that builds a package from a set of attributes. An attribute set is just a list of key/value pairs where each value is an arbitrary Nix expression. They take the general form {name1 = expr1; ... nameN = exprN;}. The attribute name specifies the symbolic name and version of the package. Nix doesn't really care about these things, but they are used by for instance nix-env -q to show a human-readable name for packages. This attribute is required by mkDerivation. The attribute builder specifies the builder. This attribute can sometimes be omitted, in which case mkDerivation will fill in a default builder (which does a configure; make; make install, in essence). Hello is sufficiently simple that the default builder would suffice, but in this case, we will show an actual builder for educational purposes. The value ./builder.sh refers to the shell script shown in , discussed below. The builder has to know what the sources of the package are. Here, the attribute src is bound to the result of a call to the fetchurl function. Given a URL and an MD5 hash of the expected contents of the file at that URL, this function builds a derivation that downloads the file and checks its hash. So the sources are a dependency that like all other dependencies is built before Hello itself is built. Instead of src any other name could have been used, and in fact there can be any number of sources (bound to different attributes). However, src is customary, and it's also expected by the default builder (which we don't use in this example). Since the derivation requires Perl, we have to pass the value of the perl function argument to the builder. All attributes in the set are actually passed as environment variables to the builder, so declaring an attribute perl = perl; will do the trick: it binds an attribute perl to the function argument which also happens to be called perl. However, it looks a bit silly, so there is a shorter syntax. The inherit keyword causes the specified attributes to be bound to whatever variables with the same name happen to be in scope.

source $stdenv/setup PATH=$perl/bin:$PATH tar xvfz $src cd hello-* ./configure --prefix=$out make make install shows the builder referenced from Hello's Nix expression (stored in pkgs/applications/misc/hello/ex-1/builder.sh). The builder can actually be made a lot shorter by using the generic builder functions provided by stdenv, but here we write out the build steps to elucidate what a builder does. It performs the following steps: When Nix runs a builder, it initially completely clears the environment (except for the attributes declared in the derivation). For instance, the PATH variable is emptyActually, it's initialised to /path-not-set to prevent Bash from setting it to a default value.. This is done to prevent undeclared inputs from being used in the build process. If for example the PATH contained /usr/bin, then you might accidentally use /usr/bin/gcc. So the first step is to set up the environment. This is done by calling the setup script of the standard environment. The environment variable stdenv points to the location of the standard environment being used. (It wasn't specified explicitly as an attribute in , but mkDerivation adds it automatically.) Since Hello needs Perl, we have to make sure that Perl is in the PATH. The perl environment variable points to the location of the Perl package (since it was passed in as an attribute to the derivation), so $perl/bin is the directory containing the Perl interpreter. Now we have to unpack the sources. The src attribute was bound to the result of fetching the Hello source tarball from the network, so the src environment variable points to the location in the Nix store to which the tarball was downloaded. After unpacking, we cd to the resulting source directory. The whole build is performed in a temporary directory created in /tmp, by the way. This directory is removed after the builder finishes, so there is no need to clean up the sources afterwards. Also, the temporary directory is always newly created, so you don't have to worry about files from previous builds interfering with the current build. GNU Hello is a typical Autoconf-based package, so we first have to run its configure script. In Nix every package is stored in a separate location in the Nix store, for instance /nix/store/9a54ba97fb71b65fda531012d0443ce2-hello-2.1.1. Nix computes this path by cryptographically hashing all attributes of the derivation. The path is passed to the builder through the out environment variable. So here we give configure the parameter --prefix=$out to cause Hello to be installed in the expected location. Finally we build Hello (make) and install it into the location specified by out (make install). If you are wondering about the absence of error checking on the result of various commands called in the builder: this is because the shell script is evaluated with Bash's -e option, which causes the script to be aborted if any command fails without an error check.

... rec { hello = (import ../applications/misc/hello/ex-1 ) { inherit fetchurl stdenv perl; }; perl = (import ../development/interpreters/perl) { inherit fetchurl stdenv; }; fetchurl = (import ../build-support/fetchurl) { inherit stdenv; ... }; stdenv = ...; } The Nix expression in is a function; it is missing some arguments that have to be filled in somewhere. In the Nix Packages collection this is done in the file pkgs/top-level/all-packages.nix, where all Nix expressions for packages are imported and called with the appropriate arguments. shows some fragments of all-packages.nix. This file defines a set of attributes, all of which are concrete derivations (i.e., not functions). In fact, we define a mutually recursive set of attributes. That is, the attributes can refer to each other. This is precisely what we want since we want to plug the various packages into each other. Here we import the Nix expression for GNU Hello. The import operation just loads and returns the specified Nix expression. In fact, we could just have put the contents of in all-packages.nix at this point. That would be completely equivalent, but it would make the file rather bulky. Note that we refer to ../applications/misc/hello/ex-1, not ../applications/misc/hello/ex-1/default.nix. When you try to import a directory, Nix automatically appends /default.nix to the file name. This is where the actual composition takes place. Here we call the function imported from ../applications/misc/hello/ex-1 with an attribute set containing the things that the function expects, namely fetchurl, stdenv, and perl. We use inherit again to use the attributes defined in the surrounding scope (we could also have written fetchurl = fetchurl;, etc.). The result of this function call is an actual derivation that can be built by Nix (since when we fill in the arguments of the function, what we get is its body, which is the call to stdenv.mkDerivation in ). Likewise, we have to instantiate Perl, fetchurl, and the standard environment.

You can now try to build Hello. Of course, you could do nix-env -f pkgs/top-level/all-packages.nix -i hello, but you may not want to install a possibly broken package just yet. The best way to test the package is by using the command nix-build, which builds a Nix expression and creates a symlink named result in the current directory: $ nix-build pkgs/top-level/all-packages.nix -A hello building path `/nix/store/632d2b22514d...-hello-2.1.1' hello-2.1.1/ hello-2.1.1/intl/ hello-2.1.1/intl/ChangeLog ... $ ls -l result lrwxrwxrwx ... 2006-09-29 10:43 result -> /nix/store/632d2b22514d...-hello-2.1.1 $ ./result/bin/hello Hello, world! The -A option selects the hello attribute from all-packages.nix. This is faster than using the symbolic package name specified by the name attribute (which also happens to be hello) and is unambiguous (there can be multiple packages with the symbolic name hello, but there can be only one attribute in a set named hello). nix-build registers the ./result symlink as a garbage collection root, so unless and until you delete the ./result symlink, the output of the build will be safely kept on your system. You can use nix-build’s -o switch to give the symlink another name. Nix has a transactional semantics. Once a build finishes successfully, Nix makes a note of this in its database: it registers that the path denoted by out is now valid. If you try to build the derivation again, Nix will see that the path is already valid and finish immediately. If a build fails, either because it returns a non-zero exit code, because Nix or the builder are killed, or because the machine crashes, then the output path will not be registered as valid. If you try to build the derivation again, Nix will remove the output path if it exists (e.g., because the builder died half-way through make install) and try again. Note that there is no negative caching: Nix doesn't remember that a build failed, and so a failed build can always be repeated. This is because Nix cannot distinguish between permanent failures (e.g., a compiler error due to a syntax error in the source) and transient failures (e.g., a disk full condition). Nix also performs locking. If you run multiple Nix builds simultaneously, and they try to build the same derivation, the first Nix instance that gets there will perform the build, while the others block (or perform other derivations if available) until the build finishes: $ nix-build pkgs/top-level/all-packages.nix -A hello waiting for lock on `/nix/store/0h5b7hp8d4hqfrw8igvx97x1xawrjnac-hello-2.1.1x' So it is always safe to run multiple instances of Nix in parallel (which isn’t the case with, say, make). If you have a system with multiple CPUs, you may want to have Nix build different derivations in parallel (insofar as possible). Just pass the option -j N, where N is the maximum number of jobs to be run in parallel, or set. Typically this should be the number of CPUs.

Recall from that the builder looked something like this: PATH=$perl/bin:$PATH tar xvfz $src cd hello-* ./configure --prefix=$out make make install The builders for almost all Unix packages look like this — set up some environment variables, unpack the sources, configure, build, and install. For this reason the standard environment provides some Bash functions that automate the build process. A builder using the generic build facilities in shown in . buildInputs="$perl" source $stdenv/setup genericBuild The buildInputs variable tells setup to use the indicated packages as inputs. This means that if a package provides a bin subdirectory, it's added to PATH; if it has a include subdirectory, it's added to GCC's header search path; and so on.How does it work? setup tries to source the file pkg/nix-support/setup-hook of all dependencies. These “setup hooks” can then set up whatever environment variables they want; for instance, the setup hook for Perl sets the PERL5LIB environment variable to contain the lib/site_perl directories of all inputs. The function genericBuild is defined in the file $stdenv/setup. The final step calls the shell function genericBuild, which performs the steps that were done explicitly in . The generic builder is smart enough to figure out whether to unpack the sources using gzip, bzip2, etc. It can be customised in many ways; see . Discerning readers will note that the buildInputs could just as well have been set in the Nix expression, like this: buildInputs = [perl]; The perl attribute can then be removed, and the builder becomes even shorter: source $stdenv/setup genericBuild In fact, mkDerivation provides a default builder that looks exactly like that, so it is actually possible to omit the builder for Hello entirely.

The Nix expression language is a pure, lazy, functional language. Purity means that operations in the language don't have side-effects (for instance, there is no variable assignment). Laziness means that arguments to functions are evaluated only when they are needed. Functional means that functions are normal values that can be passed around and manipulated in interesting ways. The language is not a full-featured, general purpose language. It's main job is to describe packages, compositions of packages, and the variability within packages. This section presents the various features of the language.

Nix has the following basic data types: Strings can be written in three ways. The most common way is to enclose the string between double quotes, e.g., "foo bar". Strings can span multiple lines. The special characters " and \ and the character sequence ${ must be escaped by prefixing them with a backslash (\). Newlines, carriage returns and tabs can be written as \n, \r and \t, respectively. You can include the result of an expression into a string by enclosing it in ${...}, a feature known as antiquotation. The enclosed expression must evaluate to something that can be coerced into a string (meaning that it must be a string, a path, or a derivation). For instance, rather than writing "--with-freetype2-library=" + freetype + "/lib" (where freetype is a derivation), you can instead write the more natural "--with-freetype2-library=${freetype}/lib" The latter is automatically translated to the former. A more complicated example (from the Nix expression for Qt): configureFlags = " -system-zlib -system-libpng -system-libjpeg ${if openglSupport then "-dlopen-opengl -L${mesa}/lib -I${mesa}/include -L${libXmu}/lib -I${libXmu}/include" else ""} ${if threadSupport then "-thread" else "-no-thread"} "; Note that Nix expressions and strings can be arbitrarily nested; in this case the outer string contains various antiquotations that themselves contain strings (e.g., "-thread"), some of which in turn contain expressions (e.g., ${mesa}). The second way to write string literals is as an indented string, which is enclosed between pairs of double single-quotes, like so: '' This is the first line. This is the second line. This is the third line. '' This kind of string literal intelligently strips indentation from the start of each line. To be precise, it strips from each line a number of spaces equal to the minimal indentation of the string as a whole (disregarding the indentation of empty lines). For instance, the first and second line are indented two space, while the third line is indented three spaces. Thus, two spaces are stripped from each line, so the resulting string is "This is the first line.\nThis is the second line.\n This is the third line.\n" Note that the whitespace and newline following the opening '' is ignored if there is no non-whitespace text on the initial line. Antiquotation (${expr}}) is supported in indented strings. Since ${ and '' have special meaning in indented strings, you need a way to quote them. ${ can be escaped by prefixing it with '', i.e., ''${. '' can be escaped by prefixing it with ', i.e., '''. Finally, linefeed, carriage-return and tab characters can be writted as ''\n, ''\r, ''\t. Indented strings are primarily useful in that they allow multi-line string literals to follow the indentation of the enclosing Nix expression, and that less escaping is typically necessary for strings representing languages such as shell scripts and configuration files because '' is much less common than ". Example: stdenv.mkDerivation { ... postInstall = '' mkdir $out/bin $out/etc cp foo $out/bin echo "Hello World" > $out/etc/foo.conf ${if enableBar then "cp bar $out/bin" else ""} ''; ... } Finally, as a convenience, URIs as defined in appendix B of RFC 2396 can be written as is, without quotes. For instance, the string "http://example.org/foo.tar.bz2" can also be written as http://example.org/foo.tar.bz2. Integers, e.g., 123. Paths, e.g., /bin/sh or ./builder.sh. A path must contain at least one slash to be recognised as such; for instance, builder.sh is not a pathIt's parsed as an expression that selects the attribute sh from the variable builder.. If the file name is relative, i.e., if it does not begin with a slash, it is made absolute at parse time relative to the directory of the Nix expression that contained it. For instance, if a Nix expression in /foo/bar/bla.nix refers to ../xyzzy/fnord.nix, the absolutised path is /foo/xyzzy/fnord.nix. Booleans with values true and false. Lists are formed by enclosing a whitespace-separated list of values between square brackets. For example, [ 123 ./foo.nix "abc" (f {x=y;}) ] defines a list of four elements, the last being the result of a call to the function f. Note that function calls have to be enclosed in parentheses. If they had been omitted, e.g., [ 123 ./foo.nix "abc" f {x=y;} ] the result would be a list of five elements, the fourth one being a function and the fifth being an attribute set. Attribute sets are really the core of the language, since ultimately it's all about creating derivations, which are really just sets of attributes to be passed to build scripts. Attribute sets are just a list of name/value pairs enclosed in curly brackets, where each value is an arbitrary expression terminated by a semicolon. For example: { x = 123; text = "Hello"; y = f { bla = 456; }; } This defines an attribute set with attributes named x, test, y. The order of the attributes is irrelevant. An attribute name may only occur once. Attributes can be selected from an attribute set using the . operator. For instance, { a = "Foo"; b = "Bar"; }.a evaluates to "Foo".

Recursive attribute sets are just normal attribute sets, but the attributes can refer to each other. For example, rec { x = y; y = 123; }.x evaluates to 123. Note that without rec the binding x = y; would refer to the variable y in the surrounding scope, if one exists, and would be invalid if no such variable exists. That is, in a normal (non-recursive) attribute set, attributes are not added to the lexical scope; in a recursive set, they are. Recursive attribute sets of course introduce the danger of infinite recursion. For example, rec { x = y; y = x; }.x does not terminateActually, Nix detects infinite recursion in this case and aborts (infinite recursion encountered).. A let-expression allows you define local variables for an expression. For instance, let x = "foo"; y = "bar"; in x + y evaluates to "foobar". There is also an obsolete form of let-expression, let { attrs }, which is translated to rec { attrs }.body. That is, the body of the let-expression is the body attribute of the attribute set. When defining an attribute set it is often convenient to copy variables from the surrounding lexical scope (e.g., when you want to propagate attributes). This can be shortened using the inherit keyword. For instance, let x = 123; in { inherit x; y = 456; } evaluates to {x = 123; y = 456;}. (Note that this works because x is added to the lexical scope by the let construct.) It is also possible to inherit attributes from another attribute set. For instance, in this fragment from all-packages.nix, graphviz = (import ../tools/graphics/graphviz) { inherit fetchurl stdenv libpng libjpeg expat x11 yacc; inherit (xlibs) libXaw; }; xlibs = { libX11 = ...; libXaw = ...; ... } libpng = ...; libjpg = ...; ... the attribute set used in the function call to the function defined in ../tools/graphics/graphviz inherits a number of variables from the surrounding scope (fetchurl ... yacc), but also inherits libXaw (the X Athena Widgets) from the xlibs (X11 client-side libraries) attribute set. Functions have the following form: pattern: body The pattern specifies what the argument of the function must look like, and binds variables in the body to (parts of) the argument. There are three kinds of patterns: If a pattern is a single identifier, then the function matches any argument. Example: let negate = x: !x; concat = x: y: x + y; in if negate true then concat "foo" "bar" else "" Note that concat is a function that takes one argument and returns a function that takes another argument. This allows partial parameterisation (i.e., only filling some of the arguments of a function); e.g., map (concat "foo") ["bar" "bla" "abc"] evaluates to ["foobar" "foobla" "fooabc"]. An attribute set pattern of the form {name1, name2, …, nameN} matches an attribute set containing the listed attributes, and binds the values of those attributes to variables in the function body. For example, the function {x, y, z}: z + y + x can only be called with a set containing exactly the attributes x, y and z. No other attributes are allowed. If you want to allow additional arguments, you can use an ellipsis (...): {x, y, z, ....}: z + y + x This works on any set that contains at least the three named attributes. It is possible to provide default values for attributes, in which case they are allowed to be missing. A default value is specified by writing name ? e, where e is an arbitrary expression. For example, {x, y ? "foo", z ? "bar"}: z + y + x specifies a function that only requires an attribute named x, but optionally accepts y and z. An @-pattern requires that the argument matches with the patterns on the left- and right-hand side of the @-sign. For example: args@{x, y, z, ...}: z + y + x + args.a Here args is bound to the entire argument, which is further matches against the pattern {x, y, z, ...}. Note that functions do not have names. If you want to give them a name, you can bind them to an attribute, e.g., let concat = {x, y}: x + y; in concat {x = "foo"; y = "bar";} Conditionals look like this: if e1 then e2 else e3 where e1 is an expression that should evaluate to a Boolean value (true or false). Assertions are generally used to check that certain requirements on or between features and dependencies hold. They look like this: assert e1; e2 where e1 is an expression that should evaluate to a Boolean value. If it evaluates to true, e2 is returned; otherwise expression evaluation is aborted and a backtrace is printed. { localServer ? false , httpServer ? false , sslSupport ? false , pythonBindings ? false , javaSwigBindings ? false , javahlBindings ? false , stdenv, fetchurl , openssl ? null, httpd ? null, db4 ? null, expat, swig ? null, j2sdk ? null }: assert localServer -> db4 != null; assert httpServer -> httpd != null && httpd.expat == expat; assert sslSupport -> openssl != null && (httpServer -> httpd.openssl == openssl); assert pythonBindings -> swig != null && swig.pythonSupport; assert javaSwigBindings -> swig != null && swig.javaSupport; assert javahlBindings -> j2sdk != null; stdenv.mkDerivation { name = "subversion-1.1.1"; ... openssl = if sslSupport then openssl else null; ... } show how assertions are used in the Nix expression for Subversion. This assertion states that if Subversion is to have support for local repositories, then Berkeley DB is needed. So if the Subversion function is called with the localServer argument set to true but the db4 argument set to null, then the evaluation fails. This is a more subtle condition: if Subversion is built with Apache (httpServer) support, then the Expat library (an XML library) used by Subversion should be same as the one used by Apache. This is because in this configuration Subversion code ends up being linked with Apache code, and if the Expat libraries do not match, a build- or runtime link error or incompatibility might occur. This assertion says that in order for Subversion to have SSL support (so that it can access https URLs), an OpenSSL library must be passed. Additionally, it says that if Apache support is enabled, then Apache's OpenSSL should match Subversion's. (Note that if Apache support is not enabled, we don't care about Apache's OpenSSL.) The conditional here is not really related to assertions, but is worth pointing out: it ensures that if SSL support is disabled, then the Subversion derivation is not dependent on OpenSSL, even if a non-null value was passed. This prevents an unnecessary rebuild of Subversion if OpenSSL changes. A with-expression, with e1; e2 introduces the attribute set e1 into the lexical scope of the expression e2. For instance, let as = {x = "foo"; y = "bar";}; in with as; x + y evaluates to "foobar" since the with adds the x and y attributes of as to the lexical scope in the expression x + y. The most common use of with is in conjunction with the import function. E.g., with (import ./definitions.nix); ... makes all attributes defined in the file definitions.nix available as if they were defined locally in a rec-expression. Comments can be single-line, started with a # character, or inline/multi-line, enclosed within /* ... */.

lists the operators in the Nix expression language, in order of precedence (from strongest to weakest binding). Syntax Associativity Description e . id none Select attribute named id from attribute set e. Abort evaluation if the attribute doesn’t exist. e1 e2 left Call function e1 with argument e2. e ? id none Test whether attribute set e contains an attribute named id; return true or false. e1 ++ e2 right List concatenation. e1 + e2 left String or path concatenation. ! e left Boolean negation. e1 // e2 right Return an attribute set consisting of the attributes in e1 and e2 (with the latter taking precedence over the former in case of equally named attributes). e1 == e2 none Equality. e1 != e2 none Inequality. e1 && e2 left Logical AND. e1 || e2 left Logical OR. e1 -> e2 none Logical implication (equivalent to !e1 || e2).

The most important built-in function is derivation, which is used to describe a single derivation (a build action). It takes as input an attribute set, the attributes of which specify the inputs of the build. There must be an attribute named system whose value must be a string specifying a Nix platform identifier, such as "i686-linux" or "powerpc-darwin"To figure out your platform identifier, look at the line Checking for the canonical Nix system name in the output of Nix's configure script. The build can only be performed on a machine and operating system matching the platform identifier. (Nix can automatically forward builds for other platforms by forwarding them to other machines; see .) There must be an attribute named name whose value must be a string. This is used as a symbolic name for the package by nix-env, and it is appended to the hash in the output path of the derivation. There must be an attribute named builder that identifies the program that is executed to perform the build. It can be either a derivation or a source (a local file reference, e.g., ./builder.sh). Every attribute is passed as an environment variable to the builder. Attribute values are translated to environment variables as follows: Strings, URIs, and integers are just passed verbatim. A path (e.g., ../foo/sources.tar) causes the referenced file to be copied to the store; its location in the store is put in the environment variable. The idea is that all sources should reside in the Nix store, since all inputs to a derivation should reside in the Nix store. A derivation causes that derivation to be built prior to the present derivation; the output path is put in the environment variable. Lists of the previous types are also allowed. They are simply concatenated, separated by spaces. true is passed as the string 1, false and null are passed as an empty string. The optional attribute args specifies command-line arguments to be passed to the builder. It should be a list. (Note that mkDerivation in the standard environment is a wrapper around derivation that adds a default value for system and always uses Bash as the builder, to which the supplied builder is passed as a command-line argument. See .) The builder is executed as follows: A temporary directory is created under the directory specified by TMPDIR (default /tmp) where the build will take place. The current directory is changed to this directory. The environment is cleared and set to the derivation attributes, as specified above. In addition, the following variables are set: NIX_BUILD_TOP contains the path of the temporary directory for this build. Also, TMPDIR, TEMPDIR, TMP, TEMP are set to point to the temporary directory. This is to prevent the builder from accidentally writing temporary files anywhere else. Doing so might cause interference by other processes. PATH is set to /path-not-set to prevent shells from initialising it to their built-in default value. HOME is set to /homeless-shelter to prevent programs from using /etc/passwd or the like to find the user's home directory, which could cause impurity. Usually, when HOME is set, it is used as the location of the home directory, even if it points to a non-existent path. NIX_STORE is set to the path of the top-level Nix store directory (typically, /nix/store). out is set to point to the output path of the derivation, which is a subdirectory of the Nix store. The output path is a concatenation of the cryptographic hash of all build inputs, and the name attribute. If the output path already exists, it is removed. Also, locks are acquired to prevent multiple Nix instances from performing the same build at the same time. A log of the combined standard output and error is written to /nix/var/log/nix. The builder is executed with the arguments specified by the attribute args. If it exits with exit code 0, it is considered to have succeeded. The temporary directory is removed (unless the -K option was specified). If the build was successful, Nix scans the output for references to the paths of the inputs. These so-called retained dependencies could be used when the output of the derivation is used (e.g., when it's executed or used as input to another derivation), so if we deploy the derivation, we should copy the retained dependencies as well. The scan is performed by looking for the hash parts of file names of the inputs. After the build, Nix sets the last-modified timestamp on all files in the build result to 1 (00:00:01 1/1/1970 UTC), sets the group to the default group, and sets the mode of the file to 0444 or 0555 (i.e., read-only, with execute permission enabled if the file was originally executable). Note that possible setuid and setgid bits are cleared. Setuid and setgid programs are not currently supported by Nix. This is because the Nix archives used in deployment have no concept of ownership information, and because it makes the build result dependent on the user performing the build.

Derivations can declare some infrequently used optional attributes. allowedReferences The optional attribute allowedReferences specifies a list of legal references (dependencies) of the output of the builder. For example, allowedReferences = []; enforces that the output of a derivation cannot have any runtime dependencies on its inputs. This is used in NixOS to check that generated files such as initial ramdisks for booting Linux don’t have accidental dependencies on other paths in the Nix store. exportReferencesGraph This attribute allows builders access to the references graph of their inputs. The attribute is a list of inputs in the Nix store whose references graph the builder needs to know. The value of this attribute should be a list of pairs [name1 path1 name2 path2 ...]. The references graph of each pathN will be stored in a text file nameN in the temporary build directory. The text files have the format used by nix-store --register-validity (with the deriver fields left empty). For example, when the following derivation is built: derivation { ... exportReferencesGraph = ["libfoo-graph" libfoo]; }; the references graph of libfoo is placed in the file libfoo-graph in the temporary build directory. exportReferencesGraph is useful for builders that want to do something with the closure of a store path. Examples include the builders in NixOS that generate the initial ramdisk for booting Linux (a cpio archive containing the closure of the boot script) and the ISO-9660 image for the installation CD (which is populated with a Nix store containing the closure of a bootable NixOS configuration). outputHash outputHashAlgo outputHashMode These attributes declare that the derivation is a so-called fixed-output derivation, which means that a cryptographic hash of the output is already known in advance. When the build of a fixed-output derivation finishes, Nix computes the cryptographic hash of the output and compares it to the hash declared with these attributes. If there is a mismatch, the build fails. The rationale for fixed-output derivations is derivations such as those produced by the fetchurl function. This function downloads a file from a given URL. To ensure that the downloaded file has not been modified, the caller must also specify a cryptographic hash of the file. For example, fetchurl { url = http://ftp.gnu.org/pub/gnu/hello/hello-2.1.1.tar.gz; md5 = "70c9ccf9fac07f762c24f2df2290784d"; } It sometimes happens that the URL of the file changes, e.g., because servers are reorganised or no longer available. We then must update the call to fetchurl, e.g., fetchurl { url = ftp://ftp.nluug.nl/pub/gnu/hello/hello-2.1.1.tar.gz; md5 = "70c9ccf9fac07f762c24f2df2290784d"; } If a fetchurl derivation was treated like a normal derivation, the output paths of the derivation and all derivations depending on it would change. For instance, if we were to change the URL of the Glibc source distribution in Nixpkgs (a package on which almost all other packages depend) massive rebuilds would be needed. This is unfortunate for a change which we know cannot have a real effect as it propagates upwards through the dependency graph. For fixed-output derivations, on the other hand, the name of the output path only depends on the outputHash* and name attributes, while all other attributes are ignored for the purpose of computing the output path. (The name attribute is included because it is part of the path.) As an example, here is the (simplified) Nix expression for fetchurl: {stdenv, curl}: # The curl program is used for downloading. {url, md5}: stdenv.mkDerivation { name = baseNameOf (toString url); builder = ./builder.sh; buildInputs = [curl]; # This is a fixed-output derivation; the output must be a regular # file with MD5 hash md5. outputHashMode = "flat"; outputHashAlgo = "md5"; outputHash = md5; inherit url; } The outputHashAlgo attribute specifies the hash algorithm used to compute the hash. It can currently be "md5", "sha1" or "sha256". The outputHashMode attribute determines how the hash is computed. It must be one of the following two values: "flat" The output must be a non-executable regular file. If it isn’t, the build fails. The hash is simply computed over the contents of that file (so it’s equal to what Unix commands like md5sum or sha1sum produce). This is the default. "recursive" The hash is computed over the NAR archive dump of the output (i.e., the result of nix-store --dump). In this case, the output can be anything, including a directory tree. The outputHash attribute, finally, must be a string containing the hash in either hexadecimal or base-32 notation. (See the nix-hash command for information about converting to and from base-32 notation.) impureEnvVars This attribute allows you to specify a list of environment variables that should be passed from the environment of the calling user to the builder. Usually, the environment is cleared completely when the builder is executed, but with this attribute you can allow specific environment variables to be passed unmodified. For example, fetchurl in Nixpkgs has the line impureEnvVars = ["http_proxy" "https_proxy" ...]; to make it use the proxy server configuration specified by the user in the environment variables http_proxy and friends. This attribute is only allowed in fixed-output derivations, where impurities such as these are okay since (the hash of) the output is known in advance. It is ignored for all other derivations.

The standard environment is used by passing it as an input called stdenv to the derivation, and then doing source $stdenv/setup at the top of the builder. Apart from adding the aforementioned commands to the PATH, setup also does the following: All input packages specified in the buildInputs environment variable have their /bin subdirectory added to PATH, their /include subdirectory added to the C/C++ header file search path, and their /lib subdirectory added to the linker search path. This can be extended. For instance, when the pkgconfig package is used, the subdirectory /lib/pkgconfig of each input is added to the PKG_CONFIG_PATH environment variable. The environment variable NIX_CFLAGS_STRIP is set so that the compiler strips debug information from object files. This can be disabled by setting NIX_STRIP_DEBUG to 0. The setup script also exports a function called genericBuild that knows how to build typical Autoconf-style packages. It can be customised to perform builds for any type of package. It is advisable to use genericBuild since it provides facilities that are almost always useful such as unpacking of sources, patching of sources, nested logging, etc. The definitive, up-to-date documentation of the generic builder is the source itself, which resides in pkgs/stdenv/generic/setup.sh.

The operation of the generic builder can be modified in many places by setting certain variables. These hook variables are typically set to the name of some shell function defined by you. For instance, to perform some additional steps after make install you would set the postInstall variable: postInstall=myPostInstall myPostInstall() { mkdir $out/share/extra cp extrafiles/* $out/share/extra }

At the beginning of each phase, the set of all shell variables is written to the file env-vars at the top-level build directory. This is useful for debugging: it allows you to recreate the environment in which a build was performed. For instance, if a build fails, then assuming you used the -K flag, you can go to the output directory and switch to the environment of the builder: $ nix-build -K ./foo.nix ... fails, keeping build directory `/tmp/nix-1234-0' $ cd /tmp/nix-1234-0 $ source env-vars (edit some files...) $ make (execution continues with the same GCC, make, etc.)