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 Ambiguous/Multiple Class/Function Definitions
27 =============================================
28
29 Classes and functions are not supposed to have multiple definitions, where one code path
30 may define one form of a class or function with a given name and another code path may
31 define another form with that name. Currently, such multiple definitions are treated like
32 "unions" in the object table.
33
34 Consider functions as well as classes which are supported using "shadow" names for the
35 second and subsequent definitions of classes in the same namespace.
36
37 Class and Module Attribute Assignment
38 =====================================
39
40 Allow unrestricted class and module assignment (but not new external binding of
41 attributes), eliminating run-time checks on object types in instructions like
42 StoreAttrIndex. This may involve less specific objects being identified during inspection.
43
44 Potentially re-evaluate class bases in order to see if they are non-constant.
45
46 Verify that the context information is correctly set, particularly for the unoptimised
47 cases.
48
49 Update docs/assignment.txt.
50
51 Prevent assignments within classes, such as method aliasing, from causing the source of an
52 assignment from being automatically generated. Instead, only external references should be
53 registered.
54
55 Prevent "from <module> import ..." statements from registering references to such local
56 aliases such that they cause the source of each alias to be automatically generated.
57
58 Consider attribute assignment observations, along with the possibility of class and module
59 attribute assignment.
60
61 (Note direct assignments as usual, indirect assignments via the attribute usage
62 mechanism. During attribute collection and inference, add assigned values to all
63 inferred targets.)
64
65 (Since class attributes can be assigned, StoreAttrIndex would no longer need to reject
66 static attributes, although this might still be necessary where attribute usage analysis
67 has not been performed.)
68
69 Potentially consider changing static attribute details to use object-relative offsets in
70 order to simplify the instruction implementations. This might allow us to eliminate the
71 static attribute flag for attributes in the object table, at least at run-time.
72
73 Dynamic Attribute Access
74 ========================
75
76 Consider explicit accessor initialisation:
77
78 attr = accessor("attr")
79 getattr(C, attr)
80
81 Attribute Usage
82 ===============
83
84 To consider: is it useful to distinguish between attribute name sets when the same names
85 are mentioned, but where one path through the code sets different values on attributes
86 than another? The _attrtypes collapses observations in order to make a list of object
87 types for a name, and the final set of names leading to such type deductions might be a
88 useful annotation to be added alongside _attrcombined.
89
90 (Update the reports to group identical sets of attribute names.)
91
92 Attribute usage on attributes might be possible if one can show that the expression of an
93 attribute access is constant and that the attribute target is also constant or only refers
94 to a single type. For example, in the sys module:
95
96 stderr = file()
97
98 If no work is done to associate the result of the invocation with the stderr name, then
99 one could instead at least attempt to determine whether stderr is assigned only once. If
100 so, it might be possible to record attribute usage on references to the name. For example:
101
102 sys.stderr.write(...) # sys.stderr supports write -> {file, ...}
103
104 Interface/Type Generalisation
105 -----------------------------
106
107 Consolidate interface observations by taking all cached table accesses and determining
108 which usage patterns lead to the same types. For example, if full usage of {a, b} and
109 {a, b, c} leads to A and B in both cases, either {a, b} can be considered as partial usage
110 of the complete interface {a, b, c}, or the latter can be considered as an
111 overspecification of the former.
112
113 Consider type deduction and its consequences where types belong to the same hierarchy
114 and where a guard could be generated for the most general type.
115
116 Consider permitting multiple class alternatives where the attributes are all identical.
117
118 Support class attribute positioning similar to instance attribute positioning, potentially
119 (for both) based on usage observations. For example, if __iter__ is used on two classes,
120 the class attribute could be exposed at a similar relative position to the class (and
121 potentially accessible using a LoadAttr-style instruction).
122
123 **** Constant attribute users need not maintain usage since they are already resolved. ****
124
125 Self-Related Usage
126 ------------------
127
128 Perform attribute usage on attributes of self as names, potentially combining observations
129 across methods.
130
131 Additional Guards
132 -----------------
133
134 Consider handling branches of values within namespaces in order to support more precise value usage.
135
136 Loop entry points and other places where usage becomes more specific might be used as
137 places to impose guards. See tests/attribute_access_type_restriction_loop_list.py for an
138 example. (Such information is already shown in the reports.)
139
140 Strict Interfaces/Types
141 -----------------------
142
143 Make the gathering of usage parameterisable according to the optimisation level so that a
144 choice can be made between control-flow-dependent observations and the simple collection
145 of all attributes used with a name (producing a more static interface observation).
146
147 AttributeError
148 --------------
149
150 Consider attribute usage observations being suspended or optional inside blocks where
151 AttributeError may be caught (although this doesn't anticipate such exceptions being
152 caught outside a function altogether). For example:
153
154 y = a.y
155 try:
156 z = a.z # z is an optional attribute
157 except AttributeError:
158 z = None
159
160 Frame Optimisations
161 ===================
162
163 Stack frame replacement where a local frame is unused after a call, such as in a tail call
164 situation.
165
166 Local assignment detection plus frame re-use. Example: slice.__init__ calls
167 xrange.__init__ with the same arguments which are unchanged in xrange.__init__. There is
168 therefore no need to build a new frame for this call, although in some cases the locals
169 frame might need expanding.
170
171 Reference tracking where objects associated with names are assigned to attributes of other
172 objects may assist in allocation optimisations. Recording whether an object referenced by
173 a name is assigned to an attribute, propagated to another name and assigned to an
174 attribute, or passed to another function or method might, if such observations were
175 combined, allow frame-based or temporary allocation to occur.
176
177 Instantiation Deduction
178 =======================
179
180 Consider handling Const, List and Tuple in micropython.inspect in order to produce
181 instances of specific classes. Then, consider adding support for guard
182 removal/verification where known instances are involved. For example:
183
184 l = []
185 l.append(123) # type deductions are filtered using instantiation knowledge
186
187 Currently, this is done only for Const values in the context of attribute accesses during
188 inspection.
189
190 Handling CallFunc in a similar way is more challenging. Consider the definitions in the sys module:
191
192 stderr = file()
193
194 It must first be established that file only ever refers to the built-in file class, and
195 only then can the assumption be made that stderr in this case refers to instances of file.
196 If file can also refer to other objects, potential filtering operations are more severely
197 limited.
198
199 Invocation-Related Deduction
200 ============================
201
202 Where an attribute access (either in conjunction with usage observations or independently)
203 could refer to a number of different targets, but where the resulting attribute is then
204 used in an invocation, filtering of the targets could be done to eliminate any targets
205 that are not callable. Guards would need introducing to prevent inappropriate operations
206 from occurring at run-time.
207
208 Inlining
209 ========
210
211 Where a function or method call can always be determined, the body of the target could be
212 inlined - copied into place - within the caller. If the target is only ever called by a
213 single caller it could be moved into place. This could enhance deductions based on
214 attribute usage since observations from the inlined function would be merged into the
215 caller.
216
217 Function Specialisation
218 =======================
219
220 Specialisation of certain functions, such as isinstance(x, cls) where cls is a known
221 constant.
222
223 Structure and Object Table Optimisations
224 ========================================
225
226 Fix object table entries for attributes not provided by any known object, or provide an
227 error, potentially overridden by options. For example, the augmented assignment methods
228 are not supported by the built-in objects and thus the operator module functions cause
229 the compilation to fail. Alternatively, just supply the methods since something has to do
230 so in the builtins.
231
232 Consider attribute merging where many attributes are just aliases for the same underlying
233 definition.
234
235 Consider references to defaults as occurring only within the context of a particular
236 function, thus eliminating default value classes if such functions are not themselves
237 invoked.
238
239 Scope Handling
240 ==============
241
242 Consider merging the InspectedModule.store tests with the scope conflict handling.
243
244 Consider labelling _scope on assignments and dealing with the assignment of removed
245 attributes, possibly removing the entire assignment, and distinguishing between such cases
246 and unknown names.
247
248 Check name origin where multiple branches could yield multiple scope interpretations:
249
250 try:
251 set # built-in name
252 except NameError:
253 from sets import Set as set # local definition of name
254
255 set # could be confused by the local definition at run-time
256
257 Object Coverage
258 ===============
259
260 Support __init__ traversal (and other implicit names) more effectively.
261
262 Importing Modules
263 =================
264
265 Consider supporting relative imports, even though this is arguably a misfeature.
266
267 Other
268 =====
269
270 Consider a separate annotation phase where deductions are added to the AST for the
271 benefit of both the reporting and code generation phases.
272
273 Check context_value initialisation (avoiding or handling None effectively).
274
275 Consider better "macro" support where new expressions need to be generated and processed.
276
277 Detect TestIdentity results involving constants, potentially optimising status-affected
278 instructions:
279
280 TestIdentity(x, y) # where x is always y
281 JumpIfFalse(...) # would be removed (never false)
282 JumpIfTrue(...) # changed to Jump(...)
283
284 Status-affected blocks could be optimised away for such constant-related results.