1 Name usage types: as parameters, as base classes, as callables. This potentially restricts
2 attribute usage effects because names mentioned as base classes are not propagated and
3 made freely available for use in attribute accesses.
4
5 Low-Level Instructions and Macro Instructions
6 =============================================
7
8 Have contexts and values stored separately in memory. This involves eliminating DataValue
9 and storing attributes using two words.
10
11 Migrate macro instructions such as the *Index instructions to library code implemented
12 using low-level instructions.
13
14 Consider introducing classic machine level instructions (word addition, subtraction, and
15 so on) in order to implement all current RSVP instructions.
16
17 Move common code sequences to a library routine, such as the context checking that occurs
18 in functions and methods.
19
20 Dataflow Optimisations
21 ======================
22
23 Assignments, particularly now that no result register exists, may cause StoreTemp/LoadTemp
24 instruction pairs to be produced and these could be eliminated.
25
26 Class and Module Attribute Assignment
27 =====================================
28
29 Verify that the context information is correctly set, particularly for the unoptimised
30 cases.
31
32 Update docs/assignment.txt.
33
34 Prevent assignments within classes, such as method aliasing, from causing the source of an
35 assignment from being automatically generated. Instead, only external references should be
36 registered.
37
38 Prevent "from <module> import ..." statements from registering references to such local
39 aliases such that they cause the source of each alias to be automatically generated.
40
41 Consider attribute assignment observations, along with the possibility of class and module
42 attribute assignment.
43
44 (Note direct assignments as usual, indirect assignments via the attribute usage
45 mechanism. During attribute collection and inference, add assigned values to all
46 inferred targets.)
47
48 (Since class attributes can be assigned, StoreAttrIndex would no longer need to reject
49 static attributes, although this might still be necessary where attribute usage analysis
50 has not been performed.)
51
52 Potentially consider changing static attribute details to use object-relative offsets in
53 order to simplify the instruction implementations. This might allow us to eliminate the
54 static attribute flag for attributes in the object table, at least at run-time.
55
56 Dynamic Attribute Access
57 ========================
58
59 Consider explicit accessor initialisation:
60
61 attr = accessor("attr")
62 getattr(C, attr)
63
64 Attribute Usage
65 ===============
66
67 To consider: is it useful to distinguish between attribute name sets when the same names
68 are mentioned, but where one path through the code sets different values on attributes
69 than another? The _attrtypes collapses observations in order to make a list of object
70 types for a name, and the final set of names leading to such type deductions might be a
71 useful annotation to be added alongside _attrcombined.
72
73 Interface/Type Generalisation
74 -----------------------------
75
76 Consolidate interface observations by taking all cached table accesses and determining
77 which usage patterns lead to the same types. For example, if full usage of {a, b} and
78 {a, b, c} leads to A and B in both cases, either {a, b} can be considered as partial usage
79 of the complete interface {a, b, c}, or the latter can be considered as an
80 overspecification of the former.
81
82 Consider type deduction and its consequences where types belong to the same hierarchy
83 and where a guard could be generated for the most general type.
84
85 Consider permitting multiple class alternatives where the attributes are all identical.
86
87 Support class attribute positioning similar to instance attribute positioning, potentially
88 (for both) based on usage observations. For example, if __iter__ is used on two classes,
89 the class attribute could be exposed at a similar relative position to the class (and
90 potentially accessible using a LoadAttr-style instruction).
91
92 **** Constant attribute users need not maintain usage since they are already resolved. ****
93
94 Self-related Usage
95 ------------------
96
97 Perform attribute usage on attributes of self as names, potentially combining observations
98 across methods.
99
100 Additional Guards
101 -----------------
102
103 Consider handling branches of values within namespaces in order to support more precise value usage.
104
105 Loop entry points and other places where usage becomes more specific might be used as
106 places to impose guards. See tests/attribute_access_type_restriction_loop_list.py for an
107 example. (Such information is already shown in the reports.)
108
109 Strict Interfaces/Types
110 -----------------------
111
112 Make the gathering of usage parameterisable according to the optimisation level so that a
113 choice can be made between control-flow-dependent observations and the simple collection
114 of all attributes used with a name (producing a more static interface observation).
115
116 AttributeError
117 --------------
118
119 Consider attribute usage observations being suspended or optional inside blocks where
120 AttributeError may be caught (although this doesn't anticipate such exceptions being
121 caught outside a function altogether). For example:
122
123 y = a.y
124 try:
125 z = a.z # z is an optional attribute
126 except AttributeError:
127 z = None
128
129 Instantiation Deduction
130 -----------------------
131
132 Consider handling CallFunc in micropython.inspect in order to produce instances of specific classes.
133 Then, consider adding support for guard removal/verification where known instances are involved. For
134 example:
135
136 l = []
137 l.append(123) # type deductions are filtered using instantiation knowledge
138
139 Frame Optimisations
140 ===================
141
142 Stack frame replacement where a local frame is unused after a call, such as in a tail call
143 situation.
144
145 Local assignment detection plus frame re-use. Example: slice.__init__ calls
146 xrange.__init__ with the same arguments which are unchanged in xrange.__init__. There is
147 therefore no need to build a new frame for this call, although in some cases the locals
148 frame might need expanding.
149
150 Reference tracking where objects associated with names are assigned to attributes of other
151 objects may assist in allocation optimisations. Recording whether an object referenced by
152 a name is assigned to an attribute, propagated to another name and assigned to an
153 attribute, or passed to another function or method might, if such observations were
154 combined, allow frame-based or temporary allocation to occur.
155
156 Instantiation
157 =============
158
159 Specific instances could be produced, providing type information and acting somewhat like
160 classes during inspection.
161
162 Inlining
163 ========
164
165 Where a function or method call can always be determined, the body of the target could be
166 inlined - copied into place - within the caller. If the target is only ever called by a
167 single caller it could be moved into place. This could enhance deductions based on
168 attribute usage since observations from the inlined function would be merged into the
169 caller.
170
171 Function Specialisation
172 =======================
173
174 Specialisation of certain functions, such as isinstance(x, cls) where cls is a known
175 constant.
176
177 Structure and Object Table Optimisations
178 ========================================
179
180 Fix object table entries for attributes not provided by any known object, or provide an
181 error, potentially overridden by options. For example, the augmented assignment methods
182 are not supported by the built-in objects and thus the operator module functions cause
183 the compilation to fail. Alternatively, just supply the methods since something has to do
184 so in the builtins.
185
186 Consider attribute merging where many attributes are just aliases for the same underlying
187 definition.
188
189 Consider references to defaults as occurring only within the context of a particular
190 function, thus eliminating default value classes if such functions are not themselves
191 invoked.
192
193 Scope Handling
194 ==============
195
196 Consider merging the InspectedModule.store tests with the scope conflict handling.
197
198 Consider labelling _scope on assignments and dealing with the assignment of removed
199 attributes, possibly removing the entire assignment, and distinguishing between such cases
200 and unknown names.
201
202 Check name origin where multiple branches could yield multiple scope interpretations:
203
204 try:
205 set # built-in name
206 except NameError:
207 from sets import Set as set # local definition of name
208
209 set # could be confused by the local definition at run-time
210
211 Object Coverage
212 ===============
213
214 Support __init__ traversal (and other implicit names) more effectively.
215
216 Importing Modules
217 =================
218
219 Consider supporting relative imports, even though this is arguably a misfeature.
220
221 Other
222 =====
223
224 Consider a separate annotation phase where deductions are added to the AST for the
225 benefit of both the reporting and code generation phases.
226
227 Support self attribute visualisation in the reports and/or provide a function or
228 annotations which can provide the eventual optimisation directly to such components.
229
230 Check context_value initialisation (avoiding or handling None effectively).
231
232 Consider better "macro" support where new expressions need to be generated and processed.
233
234 Detect TestIdentity results involving constants, potentially optimising status-affected
235 instructions:
236
237 TestIdentity(x, y) # where x is always y
238 JumpIfFalse(...) # would be removed (never false)
239 JumpIfTrue(...) # changed to Jump(...)
240
241 Status-affected blocks could be optimised away for such constant-related results.
