about summary refs log tree commit diff
path: root/lib/derivations.nix
blob: 6867458f9e87c9146152636e6a27aabfb8ab3541 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
{ lib }:

let
  inherit (lib)
    genAttrs
    isString
    throwIfNot
    ;

  showMaybeAttrPosPre = prefix: attrName: v:
    let pos = builtins.unsafeGetAttrPos attrName v;
    in if pos == null then "" else "${prefix}${pos.file}:${toString pos.line}:${toString pos.column}";

  showMaybePackagePosPre = prefix: pkg:
    if pkg?meta.position && isString pkg.meta.position
    then "${prefix}${pkg.meta.position}"
    else "";
in
{
  /*
    Restrict a derivation to a predictable set of attribute names, so
    that the returned attrset is not strict in the actual derivation,
    saving a lot of computation when the derivation is non-trivial.

    This is useful in situations where a derivation might only be used for its
    passthru attributes, improving evaluation performance.

    The returned attribute set is lazy in `derivation`. Specifically, this
    means that the derivation will not be evaluated in at least the
    situations below.

    For illustration and/or testing, we define derivation such that its
    evaluation is very noticeable.

        let derivation = throw "This won't be evaluated.";

    In the following expressions, `derivation` will _not_ be evaluated:

        (lazyDerivation { inherit derivation; }).type

        attrNames (lazyDerivation { inherit derivation; })

        (lazyDerivation { inherit derivation; } // { foo = true; }).foo

        (lazyDerivation { inherit derivation; meta.foo = true; }).meta

    In these expressions, `derivation` _will_ be evaluated:

        "${lazyDerivation { inherit derivation }}"

        (lazyDerivation { inherit derivation }).outPath

        (lazyDerivation { inherit derivation }).meta

    And the following expressions are not valid, because the refer to
    implementation details and/or attributes that may not be present on
    some derivations:

        (lazyDerivation { inherit derivation }).buildInputs

        (lazyDerivation { inherit derivation }).passthru

        (lazyDerivation { inherit derivation }).pythonPath

  */
  lazyDerivation =
    args@{
      # The derivation to be wrapped.
      derivation
    , # Optional meta attribute.
      #
      # While this function is primarily about derivations, it can improve
      # the `meta` package attribute, which is usually specified through
      # `mkDerivation`.
      meta ? null
    , # Optional extra values to add to the returned attrset.
      #
      # This can be used for adding package attributes, such as `tests`.
      passthru ? { }
    , # Optional list of assumed outputs. Default: ["out"]
      #
      # This must match the set of outputs that the returned derivation has.
      # You must use this when the derivation has multiple outputs.
      outputs ? [ "out" ]
    }:
    let
      # These checks are strict in `drv` and some `drv` attributes, but the
      # attrset spine returned by lazyDerivation does not depend on it.
      # Instead, the individual derivation attributes do depend on it.
      checked =
        throwIfNot (derivation.type or null == "derivation")
          "lazyDerivation: input must be a derivation."
          throwIfNot
          # NOTE: Technically we could require our outputs to be a subset of the
          # actual ones, or even leave them unchecked and fail on a lazy basis.
          # However, consider the case where an output is added in the underlying
          # derivation, such as dev. lazyDerivation would remove it and cause it
          # to fail as a buildInputs item, without any indication as to what
          # happened. Hence the more stringent condition. We could consider
          # adding a flag to control this behavior if there's a valid case for it,
          # but the documentation must have a note like this.
          (derivation.outputs == outputs)
          ''
            lib.lazyDerivation: The derivation ${derivation.name or "<unknown>"} has outputs that don't match the assumed outputs.

            Assumed outputs passed to lazyDerivation${showMaybeAttrPosPre ",\n    at " "outputs" args}:
                ${lib.generators.toPretty { multiline = false; } outputs};

            Actual outputs of the derivation${showMaybePackagePosPre ",\n    defined at " derivation}:
                ${lib.generators.toPretty { multiline = false; } derivation.outputs}

            If the outputs are known ahead of evaluating the derivation,
            then update the lazyDerivation call to match the actual outputs, in the same order.
            If lazyDerivation is passed a literal value, just change it to the actual outputs.
            As a result it will work as before / as intended.

            Otherwise, when the outputs are dynamic and can't be known ahead of time, it won't
            be possible to add laziness, but lib.lazyDerivation may still be useful for trimming
            the attributes.
            If you want to keep trimming the attributes, make sure that the package is in a
            variable (don't evaluate it twice!) and pass the variable and its outputs attribute
            to lib.lazyDerivation. This largely defeats laziness, but keeps the trimming.
            If none of the above works for you, replace the lib.lazyDerivation call by the
            expression in the derivation argument.
          ''
          derivation;
    in
    {
      # Hardcoded `type`
      #
      # `lazyDerivation` requires its `derivation` argument to be a derivation,
      # so if it is not, that is a programming error by the caller and not
      # something that `lazyDerivation` consumers should be able to correct
      # for after the fact.
      # So, to improve laziness, we assume correctness here and check it only
      # when actual derivation values are accessed later.
      type = "derivation";

      # A fixed set of derivation values, so that `lazyDerivation` can return
      # its attrset before evaluating `derivation`.
      # This must only list attributes that are available on _all_ derivations.
      inherit (checked) outPath outputName drvPath name system;
      inherit outputs;

      # The meta attribute can either be taken from the derivation, or if the
      # `lazyDerivation` caller knew a shortcut, be taken from there.
      meta = args.meta or checked.meta;
    }
    // genAttrs outputs (outputName: checked.${outputName})
    // passthru;

  /* Conditionally set a derivation attribute.

     Because `mkDerivation` sets `__ignoreNulls = true`, a derivation
     attribute set to `null` will not impact the derivation output hash.
     Thus, this function passes through its `value` argument if the `cond`
     is `true`, but returns `null` if not.

     Type: optionalDrvAttr :: Bool -> a -> a | Null

     Example:
       (stdenv.mkDerivation {
         name = "foo";
         x = optionalDrvAttr true 1;
         y = optionalDrvAttr false 1;
       }).drvPath == (stdenv.mkDerivation {
         name = "foo";
         x = 1;
       }).drvPath
       => true
  */
  optionalDrvAttr =
    # Condition
    cond:
    # Attribute value
    value: if cond then value else null;
}