---
layout: documentation
title: Query language
---
<h1>The Bazel Query Reference</h1>


This page is the reference manual for the <em>Bazel Query Language</em> used
when you use <code>bazel query</code> to analyze build dependencies. It also
describes the output formats <code>bazel query</code> supports.

<p>
  If you were looking for a how-to, please see the <a href='query-how-to.md'>Bazel Query How-To</a>.
</p>
<h3> bazel cquery </h3>
<p>Traditional Bazel query runs on the post-loading phase target graph and therefore
  has no concept of configurations and their related concepts. Notably, this means it does not
  correctly resolve <a href = "be/functions.html#select">select statements</a>
  and rather returns all possible resolutions of selects. A newer query environment, cquery,
  properly handles configurations but doesn't provide all of the functionality of this original
  query. For more information, see the <a href="cquery.html">Bazel cquery reference</a>.
</p>
<h2 id='examples'>Examples</h2>

<p>
  How do people use <code>bazel query</code>?  Here are typical examples:
</p>

<p>
  Why does the <code>//foo</code> tree depend on <code>//bar/baz</code>?
  Show a path:</p>
  <pre>somepath(foo/..., //bar/baz:all)</pre>


<p>
  What C++ libraries do all the <code>foo</code> tests depend on that
  the <code>foo_bin</code> target does not?</p>
  <pre>kind("cc_library", deps(kind(".*test rule", foo/...)) except deps(//foo:foo_bin))</pre>


<h2 id='tokens'>Tokens: The lexical syntax</h2>

<p>
  Expressions in the query language are composed of the following
  tokens:</p>
  <ul>
    <li>
      <p>
        <b>Keywords</b>, such as <code>let</code>. Keywords are the reserved words of the
        language, and each of them is described below. The complete set
        of keywords is:
      </p>

<code><!-- keep this alphabetically sorted -->
<a href="#set-operations">except</a><br/>
<a href="#variables">in</a><br/>
<a href="#set-operations">intersect</a><br/>
<a href="#variables">let</a><br/>
<a href="#set">set</a><br/>
<a href="#set-operations">union</a><br/>
</code>
    </li>

    <li>
      <p>
        <b>Words</b>, such as <code>foo/...</code> or
        <code>".*test rule"</code> or <code>//bar/baz:all</code>. If a
        character sequence is "quoted" (begins and ends with a single-quote
        <code>'</code> or begins and ends with a double-quote
        <code>&quot;</code>), it is a word. If a character sequence is not
        quoted, it may still be parsed as a word. Unquoted words are sequences
        of characters drawn from the alphabet characters <code>A-Za-z</code>,
        the numerals <code>0-9</code>, and the special characters <code>*/@.-_:$~</code>
        (asterisk, forward slash, at, period, hyphen, underscore, colon, dollar sign, tilde).
      </p>
      <p>
        Exceptions:

      <ul>
        <li>An unquoted word may not start with a hyphen <code>-</code></li>.
        <li>
          An unquoted word may not include the characters plus sign <code>+</code> or equals sign
          <code>=</code> at all.
        </li>
      </ul>
      <p>
        These three characters are permitted in <a href="build-ref.html#name">target names</a> but
        they are also used for operators in the query language. For example,
      </p>
      <pre>
        //foo:bar+wiz    # WRONG: scanned as //foo:bar + wiz.
        //foo:bar=wiz    # WRONG: scanned as //foo:bar = wiz.
        "//foo:bar+wiz"  # OK.
        "//foo:bar=wiz"  # OK.
      </pre>

      Concrete examples, for the first argument to the <a href='#kind'>kind</a> function:

      <pre>
        kind("blah rule", //a:b)  # The first argument contains a space character, so it needs to be quoted.
        kind(.blah, //a:b)        # It's fine for the first argument to be unquoted and start with a period.
        kind(".+blah", //a:b)     # The first argument contains a plus character, so it needs to be quoted.
        kind(*blah, //a:b)        # It's fine for the first argument to be unquoted and start with an asterisk. (But this is a silly example, because the first argument is an invalid regex).
      </pre>

      <p>
        Therefore, it is recommended to quote strings when producing query expressions programmatically from arbitrary user-supplied values.
      <p>
      <p>
        Additionally, your shell may have its own quoting requirements, e.g.
      </p>
      <pre>
        bazel query ' "//foo:bar=wiz" '  # single-quotes for shell, double-quotes for Bazel.
      </pre>

      <p>
        Keywords, when quoted, are treated as ordinary words. For example,
        <code>set</code> is a keyword and <code>"set"</code> in quotes is treated as a word.
        Both <code>foo</code> and <code>"foo"</code> are words.
      </p>

      <p>
        However, be careful when using single or double quotes in target names. When quoting one or
        more target names, use only one type of quotes (either all single or all double quotes).

      </p>
      <pre>
        The following are examples of what the Java query string will be:

        'a"'a'         # WRONG: Error message: unclosed quotation.
        "a'"a"         # WRONG: Error message: unclosed quotation.
        '"a" + 'a''    # WRONG: Error message: unexpected token 'a' after query expression '"a" + '
        "'a' + "a""    # WRONG: Error message: unexpected token 'a' after query expression ''a' + '
        "a'a"          # OK.
        'a"a'          # OK.
        '"a" + "a"'    # OK
        "'a' + 'a'"    # OK
      </pre>

    <li><b>Punctuation</b>, such as parens <code>()</code>, period
    <code>.</code> and comma <code>,</code>, etc. Words containing
    punctuation (other than the exceptions listed above in <b>Words</b>) must be quoted.
  </ul>

<p>
  Whitespace characters outside of a quoted word are ignored.
</p>

<h2 id='concepts'>Bazel query language concepts</h2>
<p>
  The Bazel query language is a language of expressions. Every
  expression evaluates to a <b>partially-ordered set</b> of targets,
  or equivalently, a <b>graph</b> (DAG) of targets. This is the only
  datatype.
</p>
<p>
  Set and graph refer to the same datatype, but emphasize different
  aspects of it, for example:
</p>
<ul>
  <li><b>Set:</b> The partial order of the targets is not interesting.</li>
  <li><b>Graph:</b> The partial order of targets is significant.</li>
</ul>

<h3>Cycles in the dependency graph</h3>
<p>
  Build dependency graphs should be acyclic.

  The algorithms used by the query language are intended for use in
  acyclic graphs, but are robust against cycles. The details of how
  cycles are treated are not specified and should not be relied upon.
</p>

<h3 id='implicit_deps'>Implicit dependencies</h3>

<p>
  In addition to build dependencies that are defined explicitly in BUILD files,
  Bazel adds additional <em>implicit</em> dependencies to rules. For example
  every Java rule implicitly depends on the JavaBuilder. Implicit dependencies
  are established using attributes that start with <code>$</code> and they
  cannot be overridden in BUILD files.

</p>

<p>
  Per default <code>bazel query</code> takes implicit dependencies into account
  when computing the query result. This behavior can be changed with
  the <code>--[no]implicit_deps</code> option.
</p>

<h3 id='soundness'>Soundness</h3>

<p>
  Bazel query language expressions operate over the build
  dependency graph, which is the graph implicitly defined by all
  rule declarations in all BUILD files. It is important to understand
  that this graph is somewhat abstract, and does not constitute a
  complete description of how to perform all the steps of a build. In
  order to perform a build, a <em>configuration</em> is required too;
  see the <a href='user-manual.html#configurations'>configurations</a>
  section of the User's Guide for more detail.
</p>

<p>
  The result of evaluating an expression in the Bazel query language
  is true <em>for all configurations</em>, which means that it may be
  a conservative over-approximation, and not exactly precise. If you
  use the query tool to compute the set of all source files needed
  during a build, it may report more than are actually necessary
  because, for example, the query tool will include all the files
  needed to support message translation, even though you don't intend
  to use that feature in your build.
</p>

<h3 id='graph-order'>On the preservation of graph order</h3>

<p>
  Operations preserve any ordering
  constraints inherited from their subexpressions. You can think of
  this as "the law of conservation of partial order". Consider an
  example: if you issue a query to determine the transitive closure of
  dependencies of a particular target, the resulting set is ordered
  according to the dependency graph. If you filter that set to
  include only the targets of <code>file</code> kind, the same
  <em>transitive</em> partial ordering relation holds between every
  pair of targets in the resulting subset&mdash;even though none of
  these pairs is actually directly connected in the original graph.
  (There are no file&ndash;file edges in the build dependency graph).
</p>

<p>
  However, while all operators <em>preserve</em> order, some
  operations, such as the <a href='#set-operations'>set operations</a>
  don't <em>introduce</em> any ordering constraints of their own.
  Consider this expression:
</p>

<pre>deps(x) union y</pre>

<p>
  The order of the final result set is guaranteed to preserve all the
  ordering constraints of its subexpressions, namely, that all the
  transitive dependencies of <code>x</code> are correctly ordered with
  respect to each other. However, the query guarantees nothing about
  the ordering of the targets in <code>y</code>, nor about the
  ordering of the targets in <code>deps(x)</code> relative to those in
  <code>y</code> (except for those targets in
  <code>y</code> that also happen to be in <code>deps(x)</code>).
</p>

<p>
  Operators that introduce ordering constraints include:
  <code>allpaths</code>,
  <code>deps</code>,
  <code>rdeps</code>,
  <code>somepath</code>,
  and the target pattern wildcards
  <code>package:*</code>,
  <code>dir/...</code>, etc.
</p>

<h3 id='sky-query'>Sky query</h3>

<p>
  Query has two different implementations, with slightly different features. The alternative one is
  called "Sky Query", and is activated by passing the following two flags:
  (<code>--universe_scope</code> or <code>--infer_universe_scope</code>) and
  <code>--order_output=no</code>.
  <code>--universe_scope=&lt;target_pattern1&gt;,...,&lt;target_patternN&gt;</code> tells query to
  preload the transitive closure of the target pattern specified by the target patterns, which can
  be both additive and subtractive. All queries are then evaluated in this "scope". In particular,
  the <a href='#allrdeps'><code>allrdeps</code></a> and
  <a href='#rbuildfiles'><code>rbuildfiles</code></a> operators only return results from this scope.
  <code>--infer_universe_scope</code> tells Bazel to infer a value for <code>--universe_scope</code>
  from the query expression. This inferred value is simply the list of unique target patterns in the
  query expression, but this might not be what you want. For example:
</p>
<pre>
   bazel query --infer_universe_scope --order_output=no "allrdeps(//my:target)"
</pre>
<p>
  The list of unique target patterns in this query expression is <code>["//my:target"]</code>, so
  Bazel will treat this the same as the invocation
</p>
<pre>
   bazel query --universe_scope=//my:target --order_output=no "allrdeps(//my:target)"
</pre>
<p>
  But the result of that query with that --universe_scope will just be <code>//my:target</code>;
  none of the rdeps of <code>//my:target</code> will be in the universe, by construction! On the
  other hand, consider:
<pre>
   bazel query --infer_universe_scope --order_output=no "tests(//a/... + b/...) intersect allrdeps(siblings(rbuildfiles(my/starlark/file.bzl)))"
</pre>
<p>
  This is a meaningful query invocation that is trying to compute the test targets in the
  <a href='#tests'><code>tests</code></a> expansion of the targets under some directories that
  transitively depend on targets whose definition uses a certain bzl file. Here,
  <code>--infer_universe_scope</code> is a convenience, especially in the case where the choice of
  <code>--universe_scope</code> would otherwise require you to parse the query expression yourself.
</p>
<p>
  So, for query expressions that use universe-scoped operators like
  <a href='#allrdeps'><code>allrdeps</code></a> and
  <a href='#rbuildfiles'><code>rbuildfiles</code></a> be sure to use
  <code>--infer_universe_scope</code> only if its behavior is what you want.
</p>

<p>
  Sky Query has some advantages and disadvantages compared to the default query. The main
  disadvantage is that it cannot order its output according to graph order, and thus certain
  <a href='#output-formats'>output formats</a> are forbidden. Its advantages are that it provides
  two operators (<a href='#allrdeps'><code>allrdeps</code></a> and
  <a href='#rbuildfiles'><code>rbuildfiles</code></a>) that are not available in the default query.
  As well, Sky Query does its work by introspecting the
  <a href='https://bazel.build/designs/skyframe.html'>Skyframe</a> graph, rather than creating a new
  graph, which is what the default implementation does. Thus, there are some circumstances in which
  it is faster and uses less memory.
</p>

<h2 id='expressions'>Expressions: Syntax and semantics of the grammar</h2>

<p>
  This is the grammar of the Bazel query language, expressed in EBNF
  notation:
</p>


<pre>expr ::= <var>word</var>
       | let <var>name</var> = <var>expr</var> in <var>expr</var>
       | (<var>expr</var>)
       | <var>expr</var> intersect <var>expr</var>
       | <var>expr</var> ^ <var>expr</var>
       | <var>expr</var> union <var>expr</var>
       | <var>expr</var> + <var>expr</var>
       | <var>expr</var> except <var>expr</var>
       | <var>expr</var> - <var>expr</var>
       | set(<var>word</var> *)
       | <var>word</var> '(' <var>int</var> | <var>word</var> | <var>expr</var> ... ')'
</pre>

<p>
  The following sections describe each of the productions of this grammar in order.
</p>

<h3 id="target-patterns">Target patterns</h3>
<pre>expr ::= <var>word</var></pre>
<p>
  Syntactically, a <em>target pattern</em> is just a word. It
  is interpreted as an (unordered) set of targets. The simplest
  target pattern is a label,
  which identifies a single target (file or rule). For example, the
  target pattern <code>//foo:bar</code> evaluates to a set
  containing one element, the target, the <code>bar</code>
  rule.
</p>

<p>
  Target patterns generalize labels to include wildcards over packages
  and targets. For example, <code>foo/...:all</code> (or
  just <code>foo/...</code>) is a target pattern that evaluates to a
  set containing all <em>rules</em> in every package recursively
  beneath the <code>foo</code> directory;
  <code>bar/baz:all</code> is a target pattern that
  evaluates to a set containing all the rules in the
  <code>bar/baz</code> package, but not its subpackages.
</p>

<p>
  Similarly, <code>foo/...:*</code> is a target pattern that evaluates
  to a set containing all <em>targets</em> (rules <em>and</em> files) in
  every package recursively beneath the <code>foo</code> directory;
  <code>bar/baz:*</code> evaluates to a set containing
  all the targets in the
  <code>bar/baz</code> package, but not its subpackages.
</p>

<p>
  Because the <code>:*</code> wildcard matches files as well as rules,
  it is often more useful than <code>:all</code> for queries.
  Conversely, the <code>:all</code> wildcard (implicit in target
  patterns like <code>foo/...</code>) is typically more useful for
  builds.
</p>

<p>
  <code>bazel query</code> target patterns work the same as
  <code>bazel build</code> build targets do;
  refer to <a href='user-manual.html#target-patterns'>Target Patterns</a>
  in the Bazel User Manual for further details, or type <code>bazel
  help target-syntax</code>.

</p>

<p>
  Target patterns may evaluate to a singleton set (in the case of a
  label), to a set containing many elements (as in the case of
  <code>foo/...</code>, which has thousands of elements) or to the
  empty set, if the target pattern matches no targets.
</p>

<p>
  All nodes in the result of a target pattern expression are correctly
  ordered relative to each other according to the dependency relation.
  So, the result of <code>foo:*</code> is not just the set of targets
  in package <code>foo</code>, it is also the <em>graph</em> over
  those targets. (No guarantees are made about the relative ordering
  of the result nodes against other nodes.) See the section
  on <a href='#graph-order'>graph order</a> for more details.
</p>

<h3 id="variables">Variables</h3>
<pre>expr ::= let <var>name</var> = <var>expr</var><sub>1</sub> in <var>expr</var><sub>2</sub>
       | <var>$name</var></pre>
<p>
  The Bazel query language allows definitions of and references to
  variables. The
  result of evaluation of a <code>let</code> expression is the same as
  that of <var>expr</var><sub>2</sub>, with all free occurrences of
  variable <var>name</var> replaced by the value of
  <var>expr</var><sub>1</sub>.
</p>

<p>
  For example, <code>let v = foo/... in allpaths($v, //common)
  intersect $v</code> is equivalent to the <code>allpaths(foo/...,
  //common) intersect foo/...</code>.
</p>

<p>
  An occurrence of a variable reference <code>name</code> other than in
  an enclosing <code>let <var>name</var> = ...</code> expression is an
  error. In other words, toplevel query expressions cannot have free
  variables.
</p>

<p>
  In the above grammar productions, <code>name</code> is like
  <em>word</em>, but with the additional constraint that it be a legal
  identifier in the C programming language. References to the variable
  must be prepended with the "$" character.
</p>

<p>
  Each <code>let</code> expression defines only a single variable,
  but you can nest them.
</p>

<p>
  (Both <a
  href='#target-patterns'>target patterns</a> and variable references
  consist of just a single token, a word, creating a syntactic
  ambiguity. However, there is no semantic ambiguity, because the
  subset of words that are legal variable names is disjoint from the
  subset of words that are legal target patterns.)
</p>

<p>
  (Technically speaking, <code>let</code> expressions do not increase
  the expressiveness of the query language: any query expressible in
  the language can also be expressed without them. However, they
  improve the conciseness of many queries, and may also lead to more
  efficient query evaluation.)
</p>

<h3 id="parentheses">Parenthesized expressions</h3>
<pre>expr ::= (<var>expr</var>)</pre>

<p>
  Parentheses associate subexpressions to force an
  order of evaluation.
  A parenthesized expression evaluates
  to the value of its argument.
</p>

<h3 id="set-operations">Algebraic set operations:
                        intersection, union, set difference</h3>

<pre>expr ::= <var>expr</var> intersect <var>expr</var>
       | <var>expr</var> ^ <var>expr</var>
       | <var>expr</var> union <var>expr</var>
       | <var>expr</var> + <var>expr</var>
       | <var>expr</var> except <var>expr</var>
       | <var>expr</var> - <var>expr</var>
</pre>

<p>
  These three operators compute the usual set operations over their
  arguments. Each operator has two forms, a nominal form such
  as <code>intersect</code> and a symbolic form such
  as <code>^</code>. Both forms are equivalent;
  the symbolic forms are quicker to type. (For clarity, the rest of
  this manual uses the nominal forms.)  For example,
</p>

<pre>foo/... except foo/bar/...</pre>

  evaluates to the set of targets that match
  <code>foo/...</code> but not
  <code>foo/bar/...</code>&nbsp;. Equivalently:

<pre>foo/... - foo/bar/...</pre>

  The <code>intersect</code> (<code>^</code>)
  and <code>union</code> (<code>+</code>) operations are commutative
  (symmetric); <code>except</code> (<code>-</code>) is
  asymmetric. The parser treats all three operators as
  left-associative and of equal precedence, so you might want parentheses.
  For example, the first two of these expressions are
  equivalent, but the third is not:

<pre>x intersect y union z
(x intersect y) union z
x intersect (y union z)</pre>

<p>
  **Recommendation:** Use parentheses where there is
  any danger of ambiguity in reading a query expression.
</p>

<h3 id="set">Read targets from an external source: set</h3>
<pre>expr ::= set(<var>word</var> *) </pre>
<p>
  The <code>set(<var>a</var> <var>b</var> <var>c</var> ...)</code>
  operator computes the union of a set of zero or
  more <a href='#target-patterns'>target patterns</a>, separated by
  whitespace (no commas).
</p>

<p>
  In conjunction with the Bourne shell's <code>$(...)</code>
  feature, <code>set()</code> provides a means of saving the results
  of one query in a regular text file, manipulating that text file
  using other programs (e.g. standard UNIX shell tools), and then
  introducing the result back into the query tool as a value for
  further processing. For example:
</p>
<pre>
  bazel query deps(//my:target) --output=label | grep ... | sed ... | awk ... &gt; foo
  bazel query "kind(cc_binary, set($(&lt;foo)))"
</pre>
<p>
  In the next example, <code>kind(cc_library,
  deps(//some_dir/foo:main, 5))</code> is effectively computed
  by filtering on the <code>maxrank</code> values using
  an <code>awk</code> program.
</p>
<pre>
  bazel query 'deps(//some_dir/foo:main)' --output maxrank |
        awk '($1 &lt; 5) { print $2;} ' &gt; foo
  bazel query "kind(cc_library, set($(&lt;foo)))"
</pre>
<p>
  In these examples, <code>$(&lt;foo)</code> is a shorthand
  for <code>$(cat foo)</code>, but shell commands other
  than <code>cat</code> may be used too&mdash;such as
  the previous <code>awk</code> command.
</p>

<p>
  Note, <code>set()</code> introduces no graph ordering constraints,
  so path information may be lost when saving and reloading sets of
  nodes using it. See the <a href='#graph-order'>graph order</a>
  section below for more detail.
</p>

<h2 id="functions">Functions</h2>
<pre>expr ::= <var>word</var> '(' <var>int</var> | <var>word</var> | <var>expr</var> ... ')'</pre>
<p>
  The query language defines several functions. The name of the function
  determines the number and type of arguments it requires. The following
  functions are available:
</p>

<code><!-- keep this alphabetically sorted -->
<a href="#path-operators">allpaths</a><br/>
<a href="#attr">attr</a><br/>

<a href="#buildfiles">buildfiles</a><br/>
<a href="#rbuildfiles">rbuildfiles</a><br/>

<a href="#deps">deps</a><br/>
<a href="#filter">filter</a><br/>
<a href="#kind">kind</a><br/>
<a href="#labels">labels</a><br/>
<a href="#loadfiles">loadfiles</a><br/>
<a href="#rdeps">rdeps</a><br/>
<a href="#allrdeps">allrdeps</a><br/>
<a href="#same_pkg_direct_rdeps">same_pkg_direct_rdeps</a><br/>
<a href="#siblings">siblings</a><br/>
<a href="#some">some</a><br/>
<a href="#path-operators">somepath</a><br/>
<a href="#tests">tests</a><br/>
<a href="#visible">visible</a><br/>
</code>

<h3 id="deps">Transitive closure of dependencies: deps</h3>
<pre>expr ::= deps(<var>expr</var>)
       | deps(<var>expr</var>, <var>depth</var>)</pre>
<p>
  The <code>deps(<var>x</var>)</code> operator evaluates to the graph
  formed by the transitive closure of dependencies of its argument set
  <var>x</var>. For example, the value of <code>deps(//foo)</code> is
  the dependency graph rooted at the single node <code>foo</code>,
  including all its dependencies. The value of
  <code>deps(foo/...)</code> is the dependency graphs whose roots are
  all rules in every package beneath the <code>foo</code> directory.
  Please note that 'dependencies' means only rule and file targets
  in this context, therefore the BUILD and Starlark files needed to
  create these targets are not included here. For that you should use the
  <a href="#buildfiles"><code>buildfiles</code></a> operator.
</p>

<p>
  The resulting graph is ordered according to the dependency relation.
  See the section on <a href='#graph-order'>graph order</a> for more
  details.
</p>

<p>
  The <code>deps</code> operator accepts an optional second argument,
  which is an integer literal specifying an upper bound on the depth
  of the search. So <code>deps(foo:*, 0)</code> returns all targets
  in the <code>foo</code> package, while <code>deps(foo:*, 1)</code>
  further includes the direct prerequisites of any target in the
  <code>foo</code> package, and <code>deps(foo:*, 2)</code> further
  includes the nodes directly reachable from the nodes in
  <code>deps(foo:*, 1)</code>, and so on.
  (These numbers correspond to the ranks shown in
  the <a href='#output-ranked'><code>minrank</code></a> output
  format.)  If the <var>depth</var> parameter is omitted, the search
  is unbounded, i.e. it computes the reflexive transitive closure of
  prerequsites.
</p>

<h3 id="rdeps">Transitive closure of reverse dependencies: rdeps</h3>
<pre>expr ::= rdeps(<var>expr</var>, <var>expr</var>)
       | rdeps(<var>expr</var>, <var>expr</var>, <var>depth</var>)</pre>
<p>
  The <code>rdeps(<var>u</var>, <var>x</var>)</code> operator evaluates
  to the reverse dependencies of the argument set <var>x</var> within the
  transitive closure of the universe set <var>u</var>.
</p>

<p>
  The resulting graph is ordered according to the dependency relation. See the
  section on <a href='#graph-order'>graph order</a> for more details.
</p>

<p>
  The <code>rdeps</code> operator accepts an optional third argument,
  which is an integer literal specifying an upper bound on the depth of the
  search. The resulting graph will only include nodes within a distance of the
  specified depth from any node in the argument set. So
  <code>rdeps(//foo, //common, 1)</code> evaluates to all nodes in the
  transitive closure of <code>//foo</code> that directly depend on
  <code>//common</code>. (These numbers correspond to the ranks shown in the
  <a href='#output-ranked'><code>minrank</code></a> output format.) If the
  <var>depth</var> parameter is omitted, the search is unbounded.
</p>

<h3 id="allrdeps">Transitive closure of all reverse dependencies: allrdeps</h3>
<pre>expr ::= allrdeps(<var>expr</var>)
       | allrdeps(<var>expr</var>, <var>depth</var>)</pre>
<b>Only available with <a href='#sky-query'>Sky Query</a></b><br/>
<p>
  The <code>allrdeps</code> operator behaves just like the <a href='#rdeps'><code>rdeps</code></a>
  operator, except that the "universe set" is whatever the <code>--universe_scope</code> flag
  evaluated to, instead of being separately specified. Thus, if
  <code>--universe_scope=//foo/...</code> was passed, then <code>allrdeps(//bar)</code> is
  equivalent to <code>rdeps(//foo/..., //bar)</code>.
</p>

<h3 id="same_pkg_direct_rdeps">Direct reverse dependencies in the same package: same_pkg_direct_rdeps</h3>
<pre>expr ::= same_pkg_direct_rdeps(<var>expr</var>)</pre>
<p>
  The <code>same_pkg_direct_rdeps(<var>x</var>)</code> operator evalutes to the full set of targets
  that are in the same package as a target in the argument set, and which directly depend on it.
</p>

<h3 id="siblings">Dealing with a target's package: siblings</h3>
<pre>expr ::= siblings(<var>expr</var>)</pre>
<p>
  The <code>siblings(<var>x</var>)</code> operator evalutes to the full set of targets that are in
  the same package as a target in the argument set.
</p>

<h3 id="some">Arbitrary choice: some</h3>
<pre>expr ::= some(<var>expr</var>)</pre>
<p>
  The <code>some(<var>x</var>)</code> operator selects one target
  arbitrarily from its argument set <var>x</var>, and evaluates to a
  singleton set containing only that target. For example, the
  expression <code>some(//foo:main union //bar:baz)</code>
  evaluates to a set containing either <code>//foo:main</code> or
  <code>//bar:baz</code>&mdash;though which one is not defined.
</p>

<p>
  If the argument is a singleton, then <code>some</code>
  computes the identity function: <code>some(//foo:main)</code> is
  equivalent to <code>//foo:main</code>.

  It is an error if the specified argument set is empty, as in the
  expression <code>some(//foo:main intersect //bar:baz)</code>.
</p>

<h3 id="path-operators">Path operators: somepath, allpaths</h3>
<pre>expr ::= somepath(<var>expr</var>, <var>expr</var>)
       | allpaths(<var>expr</var>, <var>expr</var>)</pre>
<p>
  The <code>somepath(<var>S</var>, <var>E</var>)</code> and
  <code>allpaths(<var>S</var>, <var>E</var>)</code> operators compute
  paths between two sets of targets. Both queries accept two
  arguments, a set <var>S</var> of starting points and a set
  <var>E</var> of ending points. <code>somepath</code> returns the
  graph of nodes on <em>some</em> arbitrary path from a target in
  <var>S</var> to a target in <var>E</var>; <code>allpaths</code>
  returns the graph of nodes on <em>all</em> paths from any target in
  <var>S</var> to any target in <var>E</var>.
</p>

<p>
  The resulting graphs are ordered according to the dependency relation.
  See the section on <a href='#graph-order'>graph order</a> for more
  details.
</p>

<table style='margin: auto'><tr>
<td style='text-align: center'>
<!-- digraph somepath1 {
  graph [size="4,4"]
  node [label="",shape=circle];
  n1;
  n2 [fillcolor="pink",style=filled];
  n3 [fillcolor="pink",style=filled];
  n4 [fillcolor="pink",style=filled,label="E"];
  n5; n6;
  n7 [fillcolor="pink",style=filled,label="S1"];
  n8 [label="S2"];
  n9;
  n10 [fillcolor="pink",style=filled];

  n1 -> n2;
  n2 -> n3;
  n7 -> n5;
  n7 -> n2;
  n5 -> n6;
  n6 -> n4;
  n8 -> n6;
  n6 -> n9;
  n2 -> n10;
  n3 -> n10;
  n10 -> n4;
  n10 -> n11;
} -->
<img src="images/somepath1.svg" alt="somepath1.svg" style="margin-left=10;" />
<p><code>somepath(S1 + S2, E)</code>,<br/>one possible result.</p>
</td>
<td style='padding: 40px; text-align: center'>
<!-- digraph somepath2 {
  graph [size="4,4"]
  node [label="",shape=circle];

  n1; n2; n3;
  n4 [fillcolor="pink",style=filled,label="E"];
  n5;
  n6 [fillcolor="pink",style=filled];
  n7 [label="S1"];
  n8 [fillcolor="pink",style=filled,label="S2"];
  n9; n10;

  n1 -> n2;
  n2 -> n3;
  n7 -> n5;
  n7 -> n2;
  n5 -> n6;
  n6 -> n4;
  n8 -> n6;
  n6 -> n9;
  n2 -> n10;
  n3 -> n10;
  n10 -> n4;
  n10 -> n11;
} -->
<img src="images/somepath2.svg" alt="somepath2.svg" style="margin-left=10;" />
<p><code>somepath(S1 + S2, E)</code>,<br/>another possible result.</p>
</td>
<td style='text-align: center'>
<!-- digraph allpaths {
  graph [size="4,4"]
  node [label="",shape=circle];
  n1;
  n2 [fillcolor="pink",style=filled];
  n3 [fillcolor="pink",style=filled];
  n4 [fillcolor="pink",style=filled,label="E"];
  n5 [fillcolor="pink",style=filled];
  n6 [fillcolor="pink",style=filled];
  n7 [fillcolor="pink",style=filled, label="S1"];
  n8 [fillcolor="pink",style=filled, label="S2"];
  n9;
  n10 [fillcolor="pink",style=filled];

  n1 -> n2;
  n2 -> n3;
  n7 -> n5;
  n7 -> n2;
  n5 -> n6;
  n6 -> n4;
  n8 -> n6;
  n6 -> n9;
  n2 -> n10;
  n3 -> n10;
  n10 -> n4;
  n10 -> n11;
} -->
<img src="images/allpaths.svg" alt="allpaths.svg" style="margin-left=10;" />
<p><code>allpaths(S1 + S2, E)</code>.</p>
</td>
</tr></table>

<h3 id="kind">Target kind filtering: kind</h3>
<pre>expr ::= kind(<var>word</var>, <var>expr</var>) </pre>
<p>
  The <code>kind(<var>pattern</var>, <var>input</var>)</code> operator
  applies a filter to a set of targets, and discards those targets
  that are not of the expected kind. The <var>pattern</var> parameter specifies
  what kind of target to match.
</p>
<ul>
<li><b>file</b> patterns can be one of:
  <ul>
    <li><code>source file</code>
    <li><code>generated file</code>
  </ul>
<li><b>rule</b> patterns can be one of:
  <ul>
    <li><code><var>ruletype</var> rule</code>
    <li><code><var>ruletype</var></code><br>
    Where <var>ruletype</var> is a build rule. The difference between these
    forms is that including "rule" causes the <a href="#regex">regular expression</a> match for
    <var>ruletype</var> to be anchored.
  </ul>
<li><b>package group</b> patterns should simply be:
  <ul>
    <li><code>package group</code>
  </ul>
</ul>
<p>
  For example, the kinds for the four targets defined by the BUILD file
  (for package <code>p</code>) shown below are illustrated in the
  table:
</p>

<table style='margin: auto'><tr><td style='padding-right:10px'>
<pre style='margin-left: 0em;'>
genrule(
    name = "a",
    srcs = ["a.in"],
    outs = ["a.out"],
    cmd = "...",
)
</pre>
</td><td>
   <table class="grid">
     <tr><th>Target</th><th>Kind</th></tr>
     <tr class='tt'><td>//p:a</td><td>genrule rule</td></tr>
     <tr class='tt'><td>//p:a.in</td><td>source file</td></tr>
     <tr class='tt'><td>//p:a.out</td><td>generated file</td></tr>
     <tr class='tt'><td>//p:BUILD</td><td>source file</td></tr>
   </table>
</td></tr></table>

<p>
  Thus, <code>kind("cc_.* rule", foo/...)</code> evaluates to the set
  of all <code>cc_library</code>, <code>cc_binary</code>, etc,
  rule targets beneath
  <code>foo</code>, and <code>kind("source file", deps(//foo))</code>
  evaluates to the set of all source files in the transitive closure
  of dependencies of the <code>//foo</code> target.
</p>

<p>
  Quotation of the <var>pattern</var> argument is often required
  because without it, many <a href="#regex">regular expressions</a>, such as <code>source
  file</code> and <code>.*_test</code>, are not considered words by
  the parser.
</p>

<p>
  When matching for <code>package group</code>, targets ending in
  <code>:all</code> may not yield any results.
  Use <code>:all-targets</code> instead.
</p>

<h3 id="filter">Target name filtering: filter</h3>
<pre>expr ::= filter(<var>word</var>, <var>expr</var>) </pre>
<p>
  The <code>filter(<var>pattern</var>, <var>input</var>)</code> operator
  applies a filter to a set of targets, and discards targets whose
  labels (in absolute form) do not match the pattern; it
  evaluates to a subset of its input.
</p>

<p>
  The first argument, <var>pattern</var> is a word containing a
  <a href="#regex">regular expression</a> over target names. A <code>filter</code> expression
  evaluates to the set containing all targets <var>x</var> such that
  <var>x</var> is a member of the set <var>input</var> and the
  label (in absolute form, e.g. <code>//foo:bar</code>)
  of <var>x</var> contains an (unanchored) match
  for the regular expression <var>pattern</var>. Since all
  target names start with <code>//</code>, it may be used as an alternative
  to the <code>^</code> regular expression anchor.
</p>

<p>
  This operator often provides a much faster and more robust alternative to the
  <code>intersect</code> operator. For example, in order to see all
  <code>bar</code> dependencies of the <code>//foo:foo</code> target, one could
  evaluate
</p>
<pre>deps(//foo) intersect //bar/...</pre>
<p>
  This statement, however, will require parsing of all BUILD files in the
  <code>bar</code> tree, which will be slow and prone to errors in
  irrelevant BUILD files. An alternative would be:
</p>
<pre>filter(//bar, deps(//foo))</pre>
<p>
  which would first calculate the set of <code>//foo</code> dependencies and
  then would filter only targets matching the provided pattern&mdash;in other
  words, targets with names containing <code>//bar</code> as a
  substring.
</p>

<p>
  Another common use of the <code>filter(<var>pattern</var>,
  <var>expr</var>)</code> operator is to filter specific files by their
  name or extension. For example,
</p>
<pre>filter("\.cc$", deps(//foo))</pre>
<p>
  will provide a list of all <code>.cc</code> files used to build
  <code>//foo</code>.
</p>

<h3 id="attr">Rule attribute filtering: attr</h3>
<pre>expr ::= attr(<var>word</var>, <var>word</var>, <var>expr</var>) </pre>
<p>
  The <code>attr(<var>name</var>, <var>pattern</var>, <var>input</var>)</code>
  operator applies a filter to a set of targets, and discards targets that
  are not rules, rule targets that do not have attribute <var>name</var>
  defined or rule targets where the attribute value does not match the provided
  <a href="#regex">regular expression</a> <var>pattern</var>; it evaluates to a subset of its input.
</p>

<p>
  The first argument, <var>name</var> is the name of the rule attribute that
  should be matched against the provided <a href="#regex">regular expression</a> pattern. The second
  argument, <var>pattern</var> is a regular expression over the attribute
  values. An <code>attr</code> expression evaluates to the set containing all
  targets <var>x</var> such that  <var>x</var> is a member of the set
  <var>input</var>, is a rule with the defined attribute <var>name</var> and
  the attribute value contains an (unanchored) match for the regular expression
  <var>pattern</var>. Please note, that if <var>name</var> is an optional
  attribute and rule does not specify it explicitly then default attribute
  value will be used for comparison. For example,
</p>
<pre>attr(linkshared, 0, deps(//foo))</pre>
<p>
  will select all <code>//foo</code> dependencies that are allowed to have a
  linkshared attribute (e.g., <code>cc_binary</code> rule) and have it either
  explicitly set to 0 or do not set it at all but default value is 0 (e.g. for
  <code>cc_binary</code> rules).
</p>

<p>
  List-type attributes (such as <code>srcs</code>, <code>data</code>, etc) are
  converted to strings of the form <code>[value<sub>1</sub>, ..., value<sub>n</sub>]</code>,
  starting with a <code>[</code> bracket, ending with a <code>]</code> bracket
  and using "<code>, </code>" (comma, space) to delimit multiple values.
  Labels are converted to strings by using the absolute form of the
  label. For example, an attribute <code>deps=[":foo",
  "//otherpkg:bar", "wiz"]</code> would be converted to the
  string <code>[//thispkg:foo, //otherpkg:bar, //thispkg:wiz]</code>.
  Brackets
  are always present, so the empty list would use string value <code>[]</code>
  for matching purposes. For example,
</p>
<pre>attr("srcs", "\[\]", deps(//foo))</pre>
<p>
  will select all rules among <code>//foo</code> dependencies that have an
  empty <code>srcs</code> attribute, while
</p>
<pre>attr("data", ".{3,}", deps(//foo))</pre>
<p>
  will select all rules among <code>//foo</code> dependencies that specify at
  least one value in the <code>data</code> attribute (every label is at least
  3 characters long due to the <code>//</code> and <code>:</code>).
</p>

<h3 id="visible">Rule visibility filtering: visible</h3>
<pre>expr ::= visible(<var>expr</var>, <var>expr</var>) </pre>
<p>
  The <code>visible(<var>predicate</var>, <var>input</var>)</code> operator
  applies a filter to a set of targets, and discards targets without the
  required visibility.
</p>

<p>
  The first argument, <var>predicate</var>, is a set of targets that all targets
  in the output must be visible to. A <var>visible</var> expression
  evaluates to the set containing all targets <var>x</var> such that <var>x</var>
  is a member of the set <var>input</var>, and for all targets <var>y</var> in
  <var>predicate</var> <var>x</var> is visible to <var>y</var>. For example:
</p>
<pre>visible(//foo, //bar:*)</pre>
<p>
  will select all targets in the package <code>//bar</code> that <code>//foo</code>
  can depend on without violating visibility restrictions.
</p>

<h3 id="labels">Evaluation of rule attributes of type label: labels</h3>
<pre>expr ::= labels(<var>word</var>, <var>expr</var>) </pre>
<p>
  The <code>labels(<var>attr_name</var>, <var>inputs</var>)</code>
  operator returns the set of targets specified in the
  attribute <var>attr_name</var> of type "label" or "list of label" in
  some rule in set <var>inputs</var>.
</p>

<p>
  For example, <code>labels(srcs, //foo)</code> returns the set of
  targets appearing in the <code>srcs</code> attribute of
  the <code>//foo</code> rule. If there are multiple rules
  with <code>srcs</code> attributes in the <var>inputs</var> set, the
  union of their <code>srcs</code> is returned.
</p>

<h3 id="tests">Expand and filter test_suites: tests</h3>
<pre>expr ::= tests(<var>expr</var>)</pre>
<p>
  The <code>tests(<var>x</var>)</code> operator returns the set of all test
  rules in set <var>x</var>, expanding any <code>test_suite</code> rules into
  the set of individual tests that they refer to, and applying filtering by
  <code>tag</code> and <code>size</code>.

  By default, query evaluation
  ignores any non-test targets in all <code>test_suite</code> rules. This can be
  changed to errors with the <code>--strict_test_suite</code> option.
</p>

<p>
  For example, the query <code>kind(test, foo:*)</code> lists all
  the <code>*_test</code> and <code>test_suite</code> rules
  in the <code>foo</code> package. All the results are (by
  definition) members of the <code>foo</code> package. In contrast,
  the query <code>tests(foo:*)</code> will return all of the
  individual tests that would be executed by <code>bazel test
  foo:*</code>: this may include tests belonging to other packages,
  that are referenced directly or indirectly
  via <code>test_suite</code> rules.
</p>


<h3 id="buildfiles">Package definition files: buildfiles</h3>
<pre>expr ::= buildfiles(<var>expr</var>)</pre>
<p>
  The <code>buildfiles(<var>x</var>)</code> operator returns the set
  of files that define the packages of each target in
  set <var>x</var>; in other words, for each package, its BUILD file,
  plus any files it references via <code>load</code>. Note that this also returns the BUILD files
  of the packages containing these <code>load</code>ed files.
</p>

<p>
  This operator is typically used when determining what files or
  packages are required to build a specified target, often in conjunction with
  the <a href='#output-package'><code>--output package</code></a>
  option, below). For example,
</p>
<pre>bazel query 'buildfiles(deps(//foo))' --output package</pre>
<p>
  returns the set of all packages on which <code>//foo</code> transitively
  depends.
</p>

<p>
  (Note: a naive attempt at the above query would omit
  the <code>buildfiles</code> operator and use only <code>deps</code>,
  but this yields an incorrect result: while the result contains the
  majority of needed packages, those packages that contain only files
  that are <code>load()</code>'ed will be missing.
</p>

<h3 id="rbuildfiles">Package definition files: rbuildfiles</h3>
<pre>expr ::= rbuildfiles(<var>word</var>, ...)</pre>
<b>Only available with <a href='#sky-query'>Sky Query</a></b><br/>
<p>
  The <code>rbuildfiles</code> operator takes a comma-separated list of path fragments and returns
  the set of BUILD files that transitively depend on these path fragments. For instance, if
  <code>//foo</code> is a package, then <code>rbuildfiles(foo/BUILD)</code> will return the
  <code>//foo:BUILD</code> target. If the <code>foo/BUILD</code> file has
  <code>load('//bar:file.bzl'...</code> in it, then <code>rbuildfiles(bar/file.bzl)</code> will
  return the <code>//foo:BUILD</code> target, as well as the targets for any other BUILD files that
  load <code>//bar:file.bzl</code>
</p>

<p>
  The scope of the <scope>rbuildfiles</scope> operator is the universe specified by the
  <code>--universe_scope</code> flag. Files that do not correspond directly to BUILD files and .bzl
  files do not affect the results. For instance, source files (like <code>foo.cc</code>) are ignored,
  even if they are explicitly mentioned in the BUILD file. Symlinks, however, are respected, so that
  if <code>foo/BUILD</code> is a symlink to <code>bar/BUILD</code>, then
  <code>rbuildfiles(bar/BUILD)</code> will include <code>//foo:BUILD</code> in its results.
</p>

<p>
  The <code>rbuildfiles</code> operator is almost morally the inverse of the
  <a href='#buildfiles'><code>buildfiles</code></a> operator. However, this moral inversion
  holds more strongly in one direction: the outputs of <code>rbuildfiles</code> are just like the
  inputs of <code>buildfiles</code>; the former will only contain BUILD file targets in packages,
  and the latter may contain such targets. In the other direction, the correspondence is weaker. The
  outputs of the <code>buildfiles</code> operator are targets corresponding to all packages and .bzl
  files needed by a given input. However, the inputs of the <code>rbuildfiles</code> operator are
  not those targets, but rather the path fragments that correspond to those targets.
</p>

<h3 id="loadfiles">Package definition files: loadfiles</h3>
<pre>expr ::= loadfiles(<var>expr</var>)</pre>
<p>
  The <code>loadfiles(<var>x</var>)</code> operator returns the set of
  Starlark files that are needed to load the packages of each target in
  set <var>x</var>. In other words, for each package, it returns the
  .bzl files that are referenced from its BUILD files.
</p>

<h2 id='output-formats'>Output formats</h2>

<p>
  <code>bazel query</code> generates a graph.
  You specify the content, format, and ordering by which
  <code>bazel query</code> presents this graph
  by means of the <code>--output</code>
  command-line option.
 </p>

<p>
  When running with <a href='#sky-query'>Sky Query</a>, only output formats that are compatible with
  unordered output are allowed. Specifically, <code>graph</code>, <code>minrank</code>, and
  <code>maxrank</code> output formats are forbidden.
</p>

<p>
  Some of the output formats accept additional options. The name of
  each output option is prefixed with the output format to which it
  applies, so <code>--graph:factored</code> applies only
  when <code>--output=graph</code> is being used; it has no effect if
  an output format other than <code>graph</code> is used. Similarly,
  <code>--xml:line_numbers</code> applies only when <code>--output=xml</code>
  is being used.
</p>

<h3 id='result-order'>On the ordering of results</h3>

<p>
  Although query expressions always follow the "<a href='#graph-order'>law of
  conservation of graph order</a>", <i>presenting</i> the results may be done
  in either a dependency-ordered or unordered manner. This does <b>not</b>
  influence the targets in the result set or how the query is computed. It only
  affects how the results are printed to stdout. Moreover, nodes that are
  equivalent in the dependency order may or may not be ordered alphabetically.
  The <code>--order_output</code> flag can be used to control this behavior.
  (The <code>--[no]order_results</code> flag has a subset of the functionality
  of the <code>--order_output</code> flag and is deprecated.)
</p>
<p>
  The default value of this flag is <code>auto</code>, which prints results in <b>lexicographical
  order</b>. However, when <code>somepath(a,b)</code> is used, the results will be printed in
  <code>deps</code> order instead.
</p>
<p>
  When this flag is <code>no</code> and <code>--output</code> is one of
  <code>build</code>, <code>label</code>, <code>label_kind</code>,
  <code>location</code>, <code>package</code>, <code>proto</code>, or
  <code>xml</code>, the outputs will be printed in arbitrary order. <b>This is
  generally the fastest option</b>. It is not supported though when
  <code>--output</code> is one of <code>graph</code>, <code>minrank</code> or
  <code>maxrank</code>: with these formats, bazel will always print results
  ordered by the dependency order or rank.
</p>
<p>
  When this flag is <code>deps</code>, bazel prints results in some topological order&mdash;that is,
  dependencies first. However, nodes that are unordered by the dependency order
  (because there is no path from either one to the other) may be printed in any order.
</p>
<p>
  When this flag is <code>full</code>, bazel prints nodes in a fully deterministic (total) order.
  First, all nodes are sorted alphabetically. Then, each node in the list is used as the start of a
  post-order depth-first search in which outgoing edges to unvisited nodes are traversed in
  alphabetical order of the successor nodes. Finally, nodes are printed in the reverse of the order
  in which they were visited.
</p>
<p>
  Printing nodes in this order may be slower, so it should be used only when determinism is
  important.
</p>

<h3 id="output-build">Print the source form of targets as they would appear in BUILD</h3>
<pre>--output build</pre>
<p>
  With this option, the representation of each target is as if it were
  hand-written in the BUILD language. All variables and function calls
  (e.g. glob, macros) are expanded, which is useful for seeing the effect
  of Starlark macros. Additionally, each effective rule reports a
  <code>generator_name</code> and/or <code>generator_function</code>) value,
  giving the name of the macro that was evaluated to produce the effective
  rule.
</p>
<p>
  Although the output uses the same syntax as BUILD files, it is not
  guaranteed to produce a valid BUILD file.
</p>

<h3 id="output-label">Print the label of each target</h3>
<pre>--output label</pre>
<p>
  With this option, the set of names (or <em>labels</em>) of each target
  in the resulting graph is printed, one label per line, in
  topological order (unless <code>--noorder_results</code> is specified, see
  <a href='#result-order'>notes on the ordering of results</a>).
  (A topological ordering is one in which a graph
  node appears earlier than all of its successors.)  Of course there
  are many possible topological orderings of a graph (<em>reverse
  postorder</em> is just one); which one is chosen is not specified.

  When printing the output of a <code>somepath</code> query, the order
  in which the nodes are printed is the order of the path.
</p>

<p>
  Caveat: in some corner cases, there may be two distinct targets with
  the same label; for example, a <code>sh_binary</code> rule and its
  sole (implicit) <code>srcs</code> file may both be called
  <code>foo.sh</code>. If the result of a query contains both of
  these targets, the output (in <code>label</code> format) will appear
  to contain a duplicate. When using the <code>label_kind</code> (see
  below) format, the distinction becomes clear: the two targets have
  the same name, but one has kind <code>sh_binary rule</code> and the
  other kind <code>source file</code>.
</p>

<h3 id="output-label_kind">Print the label and kind of each target</h3>
<pre>--output label_kind</pre>
<p>
  Like <code>label</code>, this output format prints the labels of
  each target in the resulting graph, in topological order, but it
  additionally precedes the label by
  the <a href='#kind'><em>kind</em></a> of the target.
</p>

<h3 id="output-ranked">Print the label of each target, in rank order</h3>
<pre>--output minrank
--output maxrank</pre>
<p>
  Like <code>label</code>, the <code>minrank</code>
  and <code>maxrank</code> output formats print the labels of each
  target in the resulting graph, but instead of appearing in
  topological order, they appear in rank order, preceded by their
  rank number. These are unaffected by the result ordering
  <code>--[no]order_results</code> flag (see <a href='#result-order'>notes on
  the ordering of results</a>).
</p>

<p>
  There are two variants of this format: <code>minrank</code> ranks
  each node by the length of the shortest path from a root node to it.
  "Root" nodes (those which have no incoming edges) are of rank 0,
  their successors are of rank 1, etc. (As always, edges point from a
  target to its prerequisites: the targets it depends upon.)
</p>

<p>
  <code>maxrank</code> ranks each node by the length of the longest
  path from a root node to it. Again, "roots" have rank 0, all other
  nodes have a rank which is one greater than the maximum rank of all
  their predecessors.
</p>

<p>
  All nodes in a cycle are considered of equal rank. (Most graphs are
  acyclic, but cycles do occur
  simply because BUILD files contain erroneous cycles.)
</p>

<p>
  These output formats are useful for discovering how deep a graph is.
  If used for the result of a <code>deps(x)</code>, <code>rdeps(x)</code>,
  or <code>allpaths</code> query, then the rank number is equal to the
  length of the shortest (with <code>minrank</code>) or longest
  (with <code>maxrank</code>) path from <code>x</code> to a node in
  that rank. <code>maxrank</code> can be used to determine the
  longest sequence of build steps required to build a target.
</p>

<p>
  Please note, the ranked output of a <code>somepath</code> query is
  basically meaningless because <code>somepath</code> doesn't
  guarantee to return either a shortest or a longest path, and it may
  include "transitive" edges from one path node to another that are
  not direct edges in original graph.
</p>

<p>
  For example, the graph on the left yields the outputs on the right
  when <code>--output minrank</code> and <code>--output maxrank</code>
  are specified, respectively.
</p>

<table style='margin: auto'><tr><td>
<!-- digraph mygraph {
  node [shape=box];
"//a:a" -> "//a:a.cc"
"//b:b" -> "//a:a"
"//b:b" -> "//b:b.cc"
"//c:c" -> "//b:b"
"//c:c" -> "//a:a"
} -->
<img src="images/out-ranked.svg" alt="out-ranked.svg" style="margin-left=10;" />
</td><td>
<pre>
minrank

0 //c:c
1 //b:b
1 //a:a
2 //b:b.cc
2 //a:a.cc
</pre>
</td><td>
<pre>
maxrank

0 //c:c
1 //b:b
2 //a:a
2 //b:b.cc
3 //a:a.cc
</pre>
</td></tr></table>

<h3 id="output-location">Print the location of each target</h3>
<pre>--output location</pre>
<p>
  Like <code>label_kind</code>, this option prints out, for each
  target in the result, the target's kind and label, but it is
  prefixed by a string describing the location of that target, as a
  filename and line number. The format resembles the output of
  <code>grep</code>. Thus, tools that can parse the latter (such as Emacs
  or vi) can also use the query output to step through a series of
  matches, allowing the Bazel query tool to be used as a
  dependency-graph-aware "grep for BUILD files".
</p>

<p>
  The location information varies by target kind (see the <a
  href='#kind'>kind</a> operator). For rules, the
  location of the rule's declaration within the BUILD file is printed.
  For source files, the location of line 1 of the actual file is
  printed. For a generated file, the location of the rule that
  generates it is printed. (The query tool does not have sufficient
  information to find the actual location of the generated file, and
  in any case, it might not exist if a build has not yet been
  performed.)
</p>

<h3 id="output-package">Print the set of packages</h3>
<pre>--output package</pre>
<p>
  This option prints the name of all packages to which
  some target in the result set belongs. The names are printed in
  lexicographical order; duplicates are excluded. Formally, this
  is a <em>projection</em> from the set of labels (package, target) onto
  packages.
</p>

<p>
  Packages in external repositories are formatted as
  <code>@repo//foo/bar</code> while packages in the main repository are
  formatted as <code>foo/bar</code>.
</p>
<p>
  In conjunction with the <code>deps(...)</code> query, this output
  option can be used to find the set of packages that must be checked
  out in order to build a given set of targets.
</p>

<h3 id="output-graph">Display a graph of the result</h3>
<pre>--output graph</pre>
<p>
  This option causes the query result to be printed as a directed
  graph in the popular AT&amp;T GraphViz format. Typically the
  result is saved to a file, such as <code>.png</code> or <code>.svg</code>.
  (If the <code>dot</code> program is not installed on your workstation, you
  can install it using the command <code>sudo apt-get install graphviz</code>.)
  See the example section below for a sample invocation.
</p>

<p>
  This output format is particularly useful for <code>allpaths</code>,
  <code>deps</code>, or <code>rdeps</code> queries, where the result
  includes a <em>set of paths</em> that cannot be easily visualized when
  rendered in a linear form, such as with <code>--output label</code>.
</p>

<p>
  By default, the graph is rendered in a <em>factored</em> form. That is,
  topologically-equivalent nodes are merged together into a single
  node with multiple labels. This makes the graph more compact
  and readable, because typical result graphs contain highly
  repetitive patterns. For example, a <code>java_library</code> rule
  may depend on hundreds of Java source files all generated by the
  same <code>genrule</code>; in the factored graph, all these files
  are represented by a single node. This behavior may be disabled
  with the <code>--nograph:factored</code> option.
</p>

<h4><code>--graph:node_limit <var>n</var></code></h4>
<p>
  The option specifies the maximum length of the label string for a
  graph node in the output. Longer labels will be truncated; -1
  disables truncation. Due to the factored form in which graphs are
  usually printed, the node labels may be very long. GraphViz cannot
  handle labels exceeding 1024 characters, which is the default value
  of this option. This option has no effect unless
  <code>--output=graph</code> is being used.
</p>

<h4><code>--[no]graph:factored</code></h4>
<p>
  By default, graphs are displayed in factored form, as explained
  <a href='#output-graph'>above</a>.
  When <code>--nograph:factored</code> is specified, graphs are
  printed without factoring. This makes visualization using GraphViz
  impractical, but the simpler format may ease processing by other
  tools (e.g. grep). This option has no effect
  unless <code>--output=graph</code> is being used.
</p>

<h3 id="output-xml">XML</h3>
<pre>--output xml</pre>
<p>
  This option causes the resulting targets to be printed in an XML
  form. The output starts with an XML header such as this
</p>
<pre>
  &lt;?xml version="1.0" encoding="UTF-8"?&gt;
  &lt;query version="2"&gt;
</pre>
<!-- The docs should continue to document version 2 into perpetuity,
     even if we add new formats, to handle clients synced to old CLs. -->
<p>
  and then continues with an XML element for each target
  in the result graph, in topological order (unless
  <a href='#result-order'>unordered results</a> are requested),
  and then finishes with a terminating
</p>
<pre>
&lt;/query&gt;
</pre>
<p>
  Simple entries are emitted for targets of <code>file</code>
  kind:
</p>
<pre>
  &lt;source-file name='//foo:foo_main.cc' .../&gt;
  &lt;generated-file name='//foo:libfoo.so' .../&gt;
</pre>
<p>
  But for rules, the XML is structured and contains definitions of all
  the attributes of the rule, including those whose value was not
  explicitly specified in the rule's BUILD file.
</p>
<p>
  Additionally, the result includes <code>rule-input</code> and
  <code>rule-output</code> elements so that the topology of the
  dependency graph can be reconstructed without having to know that,
  for example, the elements of the <code>srcs</code> attribute are
  forward dependencies (prerequisites) and the contents of the
  <code>outs</code> attribute are backward dependencies (consumers).

  <code>rule-input</code> elements for <a
  href='#implicit_deps'>implicit dependencies</a> are suppressed if
  <code>--noimplicit_deps</code> is specified.
</p>
<pre>
  &lt;rule class='cc_binary rule' name='//foo:foo' ...&gt;
    &lt;list name='srcs'&gt;
      &lt;label value='//foo:foo_main.cc'/&gt;
      &lt;label value='//foo:bar.cc'/&gt;
      ...
    &lt;/list&gt;
    &lt;list name='deps'&gt;
      &lt;label value='//common:common'/&gt;
      &lt;label value='//collections:collections'/&gt;
      ...
    &lt;/list&gt;
    &lt;list name='data'&gt;
      ...
    &lt;/list&gt;
    &lt;int name='linkstatic' value='0'/&gt;
    &lt;int name='linkshared' value='0'/&gt;
    &lt;list name='licenses'/&gt;
    &lt;list name='distribs'&gt;
      &lt;distribution value="INTERNAL" /&gt;
    &lt;/list&gt;
    &lt;rule-input name="//common:common" /&gt;
    &lt;rule-input name="//collections:collections" /&gt;
    &lt;rule-input name="//foo:foo_main.cc" /&gt;
    &lt;rule-input name="//foo:bar.cc" /&gt;
    ...
  &lt;/rule&gt;
</pre>

<p>
  Every XML element for a target contains a <code>name</code>
  attribute, whose value is the target's label, and
  a <code>location</code> attribute, whose value is the target's
  location as printed by the <a href='output-location'><code>--output
  location</code></a>.
</p>

<h4><code>--[no]xml:line_numbers</code></h4>
<p>
  By default, the locations displayed in the XML output contain line numbers.
  When <code>--noxml:line_numbers</code> is specified, line numbers are not
  printed.
</p>

<h4><code>--[no]xml:default_values</code></h4>
<p>
  By default, XML output does not include rule attribute whose value
  is the default value for that kind of attribute (e.g. because it
  were not specified in the BUILD file, or the default value was
  provided explicitly). This option causes such attribute values to
  be included in the XML output.
</p>


<h3 id="regex">Regular expressions</h3>

Regular expressions in the query language use the Java regex library, so you can use the
full syntax for <a href="https://docs.oracle.com/javase/8/docs/api/java/util/regex/Pattern.html"><code>java.util.regex.Pattern</code></a>.


<h3 id="external-repos">Querying with external repositories</h3>

<p>
  If the build depends on rules from external repositories (defined in the
  WORKSPACE file) then query results will include these dependencies. For
  example, if <code>//foo:bar</code> depends on <code>//external:some-lib</code>
  and <code>//external:some-lib</code> is bound to
  <code>@other-repo//baz:lib</code>, then
  <code>bazel query 'deps(//foo:bar)'</code>
  will list both <code>@other-repo//baz:lib</code> and
  <code>//external:some-lib</code> as dependencies.
</p>

<p>
  External repositories themselves are not dependencies of a build. That is, in
  the example above, <code>//external:other-repo</code> is not a dependency. It
  can be queried for as a member of the <code>//external</code> package, though,
  for example:
<p>

<pre>
  # Querying over all members of //external returns the repository.
  bazel query 'kind(http_archive, //external:*)'
  //external:other-repo

  # ...but the repository is not a dependency.
  bazel query 'kind(http_archive, deps(//foo:bar))'
  INFO: Empty results
</pre>