1 Low-Level Instructions and Macro Instructions
2 =============================================
3
4 Instructions manipulated through the temporary infrastructure are reused and can be
5 corrupted by set_source and set_target operations. It might be better if such operations
6 cloned the instructions or if new instructions were explicitly created.
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 Return Value Optimisations
21 ==========================
22
23 Attempt to use result data directly instead of transferring it explicitly to the working
24 registers and then using it.
25
26 Perhaps store resultant working values on Block instances in order to better track active
27 values.
28
29 Class and Module Attribute Assignment
30 =====================================
31
32 Verify that the context information is correctly set, particularly for the unoptimised
33 cases.
34
35 Update docs/assignment.txt.
36
37 Prevent assignments within classes, such as method aliasing, from causing the source of an
38 assignment from being automatically generated. Instead, only external references should be
39 registered.
40
41 Prevent "from <module> import ..." statements from registering references to such local
42 aliases such that they cause the source of each alias to be automatically generated.
43
44 Consider attribute assignment observations, along with the possibility of class and module
45 attribute assignment.
46
47 (Note direct assignments as usual, indirect assignments via the attribute usage
48 mechanism. During attribute collection and inference, add assigned values to all
49 inferred targets.)
50
51 (Since class attributes can be assigned, StoreAttrIndex would no longer need to reject
52 static attributes, although this might still be necessary where attribute usage analysis
53 has not been performed.)
54
55 Potentially consider changing static attribute details to use object-relative offsets in
56 order to simplify the instruction implementations. This might allow us to eliminate the
57 static attribute flag for attributes in the object table, at least at run-time.
58
59 Dynamic Attribute Access
60 ========================
61
62 Consider explicit accessor initialisation:
63
64 attr = accessor("attr")
65 getattr(C, attr)
66
67 Attribute Usage
68 ===============
69
70 Loop entry points and other places where usage becomes more specific might be used as
71 places to impose guards. See tests/attribute_access_type_restriction_loop_list.py for an
72 example.
73
74 Consider attribute usage observations being suspended inside blocks where AttributeError
75 may be caught (although this doesn't anticipate such exceptions being caught outside a
76 function altogether).
77
78 Consider type deduction and its consequences where types belong to the same hierarchy
79 and where a guard could be generated for the most general type.
80
81 Consider permitting multiple class alternatives where the attributes are all identical.
82
83 Support class attribute positioning similar to instance attribute positioning, potentially
84 (for both) based on usage observations. For example, if __iter__ is used on two classes,
85 the class attribute could be exposed at a similar relative position to the class (and
86 potentially accessible using a LoadAttr-style instruction).
87
88 **** Constant attribute users need not maintain usage since they are already resolved. ****
89
90 Consider handling CallFunc in micropython.inspect in order to produce instances of specific classes.
91 Then, consider adding support for guard removal/verification where known instances are involved.
92 Consider handling branches of values within namespaces in order to support more precise value usage.
93
94 Frame Optimisations
95 ===================
96
97 Stack frame replacement where a local frame is unused after a call, such as in a tail call
98 situation.
99
100 Local assignment detection plus frame re-use. Example: slice.__init__ calls
101 xrange.__init__ with the same arguments which are unchanged in xrange.__init__. There is
102 therefore no need to build a new frame for this call, although in some cases the locals
103 frame might need expanding.
104
105 Inlining
106 ========
107
108 Where a function or method call can always be determined, the body of the target could be
109 inlined - copied into place - within the caller. If the target is only ever called by a
110 single caller it could be moved into place.
111
112 Function Specialisation
113 =======================
114
115 Specialisation of certain functions, such as isinstance(x, cls) where cls is a known
116 constant.
117
118 Structure and Object Table Optimisations
119 ========================================
120
121 Fix object table entries for attributes not provided by any known object, or provide an
122 error, potentially overridden by options. For example, the augmented assignment methods
123 are not supported by the built-in objects and thus the operator module functions cause
124 the compilation to fail. Alternatively, just supply the methods since something has to do
125 so in the builtins.
126
127 Consider attribute merging where many attributes are just aliases for the same underlying
128 definition.
129
130 Consider references to defaults as occurring only within the context of a particular
131 function, thus eliminating default value classes if such functions are not themselves
132 invoked.
133
134 Scope Handling
135 ==============
136
137 Consider merging the InspectedModule.store tests with the scope conflict handling.
138
139 Consider labelling _scope on assignments and dealing with the assignment of removed
140 attributes, possibly removing the entire assignment, and distinguishing between such cases
141 and unknown names.
142
143 Check name origin where multiple branches could yield multiple scope interpretations:
144
145 ----
146 try:
147 set # built-in name
148 except NameError:
149 from sets import Set as set # local definition of name
150
151 set # could be confused by the local definition at run-time
152 ----
153
154 Object Coverage
155 ===============
156
157 Support __init__ traversal (and other implicit names) more effectively.
158
159 Other
160 =====
161
162 Support self attribute visualisation in the reports and/or provide a function or
163 annotations which can provide the eventual optimisation directly to such components.
164
165 Check context_value initialisation (avoiding or handling None effectively).
166
167 Consider better "macro" support where new expressions need to be generated and processed.
168
169 Detect TestIdentity results involving constants, potentially optimising status-affected
170 instructions:
171
172 TestIdentity(x, y) # where x is always y
173 JumpIfFalse(...) # would be removed (never false)
174 JumpIfTrue(...) # changed to Jump(...)
175
176 Status-affected blocks could be optimised away for such constant-related results.
