111
|
1 ------------------------------------------------------------------------------
|
|
2 -- --
|
|
3 -- GNAT COMPILER COMPONENTS --
|
|
4 -- --
|
|
5 -- E X P A N D E R --
|
|
6 -- --
|
|
7 -- S p e c --
|
|
8 -- --
|
131
|
9 -- Copyright (C) 1992-2018, Free Software Foundation, Inc. --
|
111
|
10 -- --
|
|
11 -- GNAT is free software; you can redistribute it and/or modify it under --
|
|
12 -- terms of the GNU General Public License as published by the Free Soft- --
|
|
13 -- ware Foundation; either version 3, or (at your option) any later ver- --
|
|
14 -- sion. GNAT is distributed in the hope that it will be useful, but WITH- --
|
|
15 -- OUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY --
|
|
16 -- or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License --
|
|
17 -- for more details. You should have received a copy of the GNU General --
|
|
18 -- Public License distributed with GNAT; see file COPYING3. If not, go to --
|
|
19 -- http://www.gnu.org/licenses for a complete copy of the license. --
|
|
20 -- --
|
|
21 -- GNAT was originally developed by the GNAT team at New York University. --
|
|
22 -- Extensive contributions were provided by Ada Core Technologies Inc. --
|
|
23 -- --
|
|
24 ------------------------------------------------------------------------------
|
|
25
|
|
26 -- This procedure performs any required expansion for the specified node.
|
|
27 -- The argument is the node that is a candidate for possible expansion.
|
|
28 -- If no expansion is required, then Expand returns without doing anything.
|
|
29
|
|
30 -- If the node does need expansion, then the subtree is replaced by the
|
|
31 -- tree corresponding to the required rewriting. This tree is a syntactic
|
|
32 -- tree, except that all Entity fields must be correctly set on all
|
|
33 -- direct names, since the expander presumably knows what it wants, and in
|
|
34 -- any case it doesn't work to have the semantic analyzer perform visibility
|
|
35 -- analysis on these trees (they may have references to non-visible runtime
|
|
36 -- routines etc.) There are a few exceptions to this rule in special cases,
|
|
37 -- but they must be documented clearly.
|
|
38
|
|
39 -- Expand is called in two different situations:
|
|
40
|
|
41 -- Nodes that are not subexpressions (Nkind not in N_Subexpr)
|
|
42
|
|
43 -- In this case, Expand is called from the body of Sem, immediately
|
|
44 -- after completing semantic analysis by calling the corresponding
|
|
45 -- Analyze_N_xxx procedure. If expansion occurs, the given node must
|
|
46 -- be replaced with another node that is also not a subexpression.
|
|
47 -- This seems naturally to be the case, since it is hard to imagine any
|
|
48 -- situation in which it would make sense to replace a non-expression
|
|
49 -- subtree with an expression. Once the substitution is completed, the
|
|
50 -- Expand routine must call Analyze on the resulting node to do any
|
|
51 -- required semantic analysis. Note that references to children copied
|
|
52 -- from the old tree won't be reanalyzed, since their Analyzed flag
|
|
53 -- is set.
|
|
54
|
|
55 -- Nodes that are subexpressions (Nkind in N_Subexpr)
|
|
56
|
|
57 -- In this case, Expand is called from Sem_Res.Resolve after completing
|
|
58 -- the resolution of the subexpression (this means that the expander sees
|
|
59 -- the fully typed subtree). If expansion occurs, the given node must be
|
|
60 -- replaced by a node that is also a subexpression. Again it is hard
|
|
61 -- to see how this restriction could possibly be violated. Once the
|
|
62 -- substitution is completed, the Expand routine must first call Analyze
|
|
63 -- on the resulting node to do any required semantic analysis, and then
|
|
64 -- call Resolve on the node to set the type (typically the type will be
|
|
65 -- the same as the original type of the input node, but this is not
|
|
66 -- always the case).
|
|
67
|
|
68 -- In both these cases, Replace or Rewrite must be used to achieve the
|
|
69 -- expansion of the node, since the Expander routine is only passed the
|
|
70 -- Node_Id of the node to be expanded, and the resulting expanded Node_Id
|
|
71 -- must be the same (the parameter to Expand is mode in, not mode in-out).
|
|
72
|
|
73 -- For nodes other than subexpressions, it is not necessary to preserve the
|
|
74 -- original tree in the Expand routines, unlike the case for modifications
|
|
75 -- to the tree made in the semantic analyzer. This is because anyone who is
|
|
76 -- interested in working with the original tree (like ASIS) is required to
|
|
77 -- compile in semantics checks only mode. Thus Replace may be freely used
|
|
78 -- in such instances.
|
|
79
|
|
80 -- For subexpressions, preservation of the original tree is required because
|
|
81 -- of the need for conformance checking of default expressions, which occurs
|
|
82 -- on expanded trees. This means that Replace should not ever be used on
|
|
83 -- on subexpression nodes. Instead use Rewrite.
|
|
84
|
|
85 -- Note: the front end avoids calls to any of the expand routines if code
|
|
86 -- is not being generated. This is done for three reasons:
|
|
87
|
|
88 -- 1. Make sure tree does not get mucked up by the expander if no
|
|
89 -- code is being generated, and is thus usable by ASIS etc.
|
|
90
|
|
91 -- 2. Save time, since expansion is not needed if a compilation is
|
|
92 -- being done only to check the semantics, or if code generation
|
|
93 -- has been canceled due to previously detected errors.
|
|
94
|
|
95 -- 3. Allow the expand routines to assume that the tree is error free.
|
|
96 -- This results from the fact that code generation mode is always
|
|
97 -- cancelled when any error occurs.
|
|
98
|
|
99 -- If we ever decide to implement a feature allowing object modules to be
|
|
100 -- generated even if errors have been detected, then point 3 will no longer
|
|
101 -- hold, and the expand routines will have to be modified to operate properly
|
|
102 -- in the presence of errors (for many reasons this is not currently true).
|
|
103
|
|
104 -- Note: a consequence of this approach is that error messages must never
|
|
105 -- be generated in the expander, since this would mean that such error
|
|
106 -- messages are not generated when the expander is not being called.
|
|
107
|
|
108 -- Expansion is the last stage of analyzing a node, so Expand sets the
|
|
109 -- Analyzed flag of the node being analyzed as its last action. This is
|
|
110 -- done even if expansion is off (in this case, the only effect of the
|
|
111 -- call to Expand is to set the Analyzed flag to True).
|
|
112
|
|
113 with Types; use Types;
|
|
114
|
|
115 package Expander is
|
|
116
|
|
117 -- The flag Opt.Expander_Active controls whether expansion is active
|
|
118 -- (True) or deactivated (False). When expansion is deactivated all
|
|
119 -- calls to expander routines have no effect. To temporarily disable
|
|
120 -- expansion, always call the routines defined below, do NOT change
|
|
121 -- Expander_Active directly.
|
|
122 --
|
|
123 -- You should not use this flag to test if you are currently processing
|
|
124 -- a generic spec or body. Use the flag Inside_A_Generic instead (see
|
|
125 -- the spec of package Sem).
|
|
126 --
|
|
127 -- There is no good reason for permanently changing the value of this flag
|
|
128 -- except after detecting a syntactic or semantic error. In this event
|
|
129 -- this flag is set to False to disable all subsequent expansion activity.
|
|
130 --
|
|
131 -- In general this flag should be used as a read only value. The only
|
|
132 -- exceptions where it makes sense to temporarily change its value are:
|
|
133 --
|
|
134 -- (a) when starting/completing the processing of a generic definition
|
|
135 -- or declaration (see routines Start_Generic_Processing and
|
|
136 -- End_Generic_Processing in Sem_Ch12)
|
|
137 --
|
131
|
138 -- (b) when starting/completing the preanalysis of an expression
|
|
139 -- (see the spec of package Sem for more info on preanalysis.)
|
111
|
140 --
|
|
141 -- Note that when processing a spec expression (In_Spec_Expression
|
|
142 -- is True) or performing semantic analysis of a generic spec or body
|
131
|
143 -- (Inside_A_Generic) or when performing preanalysis (Full_Analysis is
|
111
|
144 -- False) the Expander_Active flag is False.
|
|
145
|
|
146 procedure Expand (N : Node_Id);
|
|
147 -- Expand node N, as described above
|
|
148
|
|
149 procedure Expander_Mode_Save_And_Set (Status : Boolean);
|
|
150 -- Saves the current setting of the Expander_Active flag on an internal
|
|
151 -- stack and then sets the flag to the given value.
|
|
152 --
|
|
153 -- Note: this routine has no effect in ASIS and GNATprove modes. In ASIS
|
|
154 -- mode, all expansion activity is always off, since we want the original
|
|
155 -- semantic tree for ASIS purposes without any expansion. In GNATprove
|
|
156 -- mode, a very light expansion is performed on specific nodes. Both are
|
|
157 -- achieved by setting Expander_Active False in ASIS and GNATprove modes.
|
|
158 -- In situations such as the call to Instantiate_Bodies in Frontend,
|
|
159 -- Expander_Mode_Save_And_Set may be called to temporarily turn the
|
|
160 -- expander on, but this will have no effect in ASIS and GNATprove modes.
|
|
161
|
|
162 procedure Expander_Mode_Restore;
|
|
163 -- Restores the setting of the Expander_Active flag using the top entry
|
|
164 -- pushed onto the stack by Expander_Mode_Save_And_Reset, popping the
|
|
165 -- stack, except that if any errors have been detected, then the state of
|
|
166 -- the flag is left set to False. Disabled for ASIS and GNATprove modes
|
|
167 -- (see above).
|
|
168
|
|
169 end Expander;
|