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 Allow unrestricted class and module assignment (but not new external binding of
30 attributes), eliminating run-time checks on object types in instructions like
31 StoreAttrIndex. This may involve less specific objects being identified during inspection.
32
33 Potentially re-evaluate class bases in order to see if they are non-constant.
34
35 Verify that the context information is correctly set, particularly for the unoptimised
36 cases.
37
38 Update docs/assignment.txt.
39
40 Prevent assignments within classes, such as method aliasing, from causing the source of an
41 assignment from being automatically generated. Instead, only external references should be
42 registered.
43
44 Prevent "from <module> import ..." statements from registering references to such local
45 aliases such that they cause the source of each alias to be automatically generated.
46
47 Consider attribute assignment observations, along with the possibility of class and module
48 attribute assignment.
49
50 (Note direct assignments as usual, indirect assignments via the attribute usage
51 mechanism. During attribute collection and inference, add assigned values to all
52 inferred targets.)
53
54 (Since class attributes can be assigned, StoreAttrIndex would no longer need to reject
55 static attributes, although this might still be necessary where attribute usage analysis
56 has not been performed.)
57
58 Potentially consider changing static attribute details to use object-relative offsets in
59 order to simplify the instruction implementations. This might allow us to eliminate the
60 static attribute flag for attributes in the object table, at least at run-time.
61
62 Dynamic Attribute Access
63 ========================
64
65 Consider explicit accessor initialisation:
66
67 attr = accessor("attr")
68 getattr(C, attr)
69
70 Attribute Usage
71 ===============
72
73 To consider: is it useful to distinguish between attribute name sets when the same names
74 are mentioned, but where one path through the code sets different values on attributes
75 than another? The _attrtypes collapses observations in order to make a list of object
76 types for a name, and the final set of names leading to such type deductions might be a
77 useful annotation to be added alongside _attrcombined.
78
79 Interface/Type Generalisation
80 -----------------------------
81
82 Consolidate interface observations by taking all cached table accesses and determining
83 which usage patterns lead to the same types. For example, if full usage of {a, b} and
84 {a, b, c} leads to A and B in both cases, either {a, b} can be considered as partial usage
85 of the complete interface {a, b, c}, or the latter can be considered as an
86 overspecification of the former.
87
88 Consider type deduction and its consequences where types belong to the same hierarchy
89 and where a guard could be generated for the most general type.
90
91 Consider permitting multiple class alternatives where the attributes are all identical.
92
93 Support class attribute positioning similar to instance attribute positioning, potentially
94 (for both) based on usage observations. For example, if __iter__ is used on two classes,
95 the class attribute could be exposed at a similar relative position to the class (and
96 potentially accessible using a LoadAttr-style instruction).
97
98 **** Constant attribute users need not maintain usage since they are already resolved. ****
99
100 Self-related Usage
101 ------------------
102
103 Perform attribute usage on attributes of self as names, potentially combining observations
104 across methods.
105
106 Additional Guards
107 -----------------
108
109 Consider handling branches of values within namespaces in order to support more precise value usage.
110
111 Loop entry points and other places where usage becomes more specific might be used as
112 places to impose guards. See tests/attribute_access_type_restriction_loop_list.py for an
113 example. (Such information is already shown in the reports.)
114
115 Strict Interfaces/Types
116 -----------------------
117
118 Make the gathering of usage parameterisable according to the optimisation level so that a
119 choice can be made between control-flow-dependent observations and the simple collection
120 of all attributes used with a name (producing a more static interface observation).
121
122 AttributeError
123 --------------
124
125 Consider attribute usage observations being suspended or optional inside blocks where
126 AttributeError may be caught (although this doesn't anticipate such exceptions being
127 caught outside a function altogether). For example:
128
129 y = a.y
130 try:
131 z = a.z # z is an optional attribute
132 except AttributeError:
133 z = None
134
135 Instantiation Deduction
136 -----------------------
137
138 Consider handling CallFunc in micropython.inspect in order to produce instances of specific classes.
139 Then, consider adding support for guard removal/verification where known instances are involved. For
140 example:
141
142 l = []
143 l.append(123) # type deductions are filtered using instantiation knowledge
144
145 Frame Optimisations
146 ===================
147
148 Stack frame replacement where a local frame is unused after a call, such as in a tail call
149 situation.
150
151 Local assignment detection plus frame re-use. Example: slice.__init__ calls
152 xrange.__init__ with the same arguments which are unchanged in xrange.__init__. There is
153 therefore no need to build a new frame for this call, although in some cases the locals
154 frame might need expanding.
155
156 Reference tracking where objects associated with names are assigned to attributes of other
157 objects may assist in allocation optimisations. Recording whether an object referenced by
158 a name is assigned to an attribute, propagated to another name and assigned to an
159 attribute, or passed to another function or method might, if such observations were
160 combined, allow frame-based or temporary allocation to occur.
161
162 Instantiation
163 =============
164
165 Specific instances could be produced, providing type information and acting somewhat like
166 classes during inspection.
167
168 Inlining
169 ========
170
171 Where a function or method call can always be determined, the body of the target could be
172 inlined - copied into place - within the caller. If the target is only ever called by a
173 single caller it could be moved into place. This could enhance deductions based on
174 attribute usage since observations from the inlined function would be merged into the
175 caller.
176
177 Function Specialisation
178 =======================
179
180 Specialisation of certain functions, such as isinstance(x, cls) where cls is a known
181 constant.
182
183 Structure and Object Table Optimisations
184 ========================================
185
186 Fix object table entries for attributes not provided by any known object, or provide an
187 error, potentially overridden by options. For example, the augmented assignment methods
188 are not supported by the built-in objects and thus the operator module functions cause
189 the compilation to fail. Alternatively, just supply the methods since something has to do
190 so in the builtins.
191
192 Consider attribute merging where many attributes are just aliases for the same underlying
193 definition.
194
195 Consider references to defaults as occurring only within the context of a particular
196 function, thus eliminating default value classes if such functions are not themselves
197 invoked.
198
199 Scope Handling
200 ==============
201
202 Consider merging the InspectedModule.store tests with the scope conflict handling.
203
204 Consider labelling _scope on assignments and dealing with the assignment of removed
205 attributes, possibly removing the entire assignment, and distinguishing between such cases
206 and unknown names.
207
208 Check name origin where multiple branches could yield multiple scope interpretations:
209
210 try:
211 set # built-in name
212 except NameError:
213 from sets import Set as set # local definition of name
214
215 set # could be confused by the local definition at run-time
216
217 Object Coverage
218 ===============
219
220 Support __init__ traversal (and other implicit names) more effectively.
221
222 Importing Modules
223 =================
224
225 Consider supporting relative imports, even though this is arguably a misfeature.
226
227 Other
228 =====
229
230 Consider a separate annotation phase where deductions are added to the AST for the
231 benefit of both the reporting and code generation phases.
232
233 Support self attribute visualisation in the reports and/or provide a function or
234 annotations which can provide the eventual optimisation directly to such components.
235
236 Check context_value initialisation (avoiding or handling None effectively).
237
238 Consider better "macro" support where new expressions need to be generated and processed.
239
240 Detect TestIdentity results involving constants, potentially optimising status-affected
241 instructions:
242
243 TestIdentity(x, y) # where x is always y
244 JumpIfFalse(...) # would be removed (never false)
245 JumpIfTrue(...) # changed to Jump(...)
246
247 Status-affected blocks could be optimised away for such constant-related results.