Difference between revisions of "Semantics"
m (s/LibertyEiffel/Liberty Eiffel/) |
|||
(One intermediate revision by one other user not shown) | |||
Line 3: | Line 3: | ||
''Note:'' this page is still being written. Take the information with a grain of salt. |
''Note:'' this page is still being written. Take the information with a grain of salt. |
||
− | Semantic analysis is strongly linked with the most important aspect of the |
+ | Semantic analysis is strongly linked with the most important aspect of the Liberty Eiffel tools: '''code specialization'''. Liberty Eiffel does not use a syntax tree; but a ''semantics'' tree. Each node of that tree is a reflection of the syntax tree in the light of a particular type (think generic classes, inheritance). The original tree is provided by the parser. |
All the specialization phases are one pass over the whole code specialized by earlier phases. The usual algorithm is the following: |
All the specialization phases are one pass over the whole code specialized by earlier phases. The usual algorithm is the following: |
Latest revision as of 18:12, 14 June 2016
Overview
Note: this page is still being written. Take the information with a grain of salt.
Semantic analysis is strongly linked with the most important aspect of the Liberty Eiffel tools: code specialization. Liberty Eiffel does not use a syntax tree; but a semantics tree. Each node of that tree is a reflection of the syntax tree in the light of a particular type (think generic classes, inheritance). The original tree is provided by the parser.
All the specialization phases are one pass over the whole code specialized by earlier phases. The usual algorithm is the following:
- try to specialize every attribute
- if any changed, twin Current and assign the specialized attributes
- return either Current (if nothing changed) or the twin (i.e. specialized)
Most specialization phases can run simultaneously for different types. Some phases (viz. collect) may even call again earlier phases. The only barrier is at the end of collect: at that time, all the types are ready to use (i.e. the whole system is known).
The specialization phases are:
- specialize_in
- specialize_thru
- specialize_and_check
- collect
- simplify
- adapt
Also of note:
- resolve_in (not available before, including, specialize_2)
- FEATURE_ACCUMULATOR
- ANONYMOUS_FEATURE_MIXER
Specialization phases
specialize_in
- Type building
The code to specialize comes directly from the parser. This code is used every time a new TYPE must be created: all the code coming from parents is specialized when the type of the target (i.e. Current) is the type being created.
For instance, the code coming from ANY is specialized in every type, either directly of indirectly (barring redefines, of course).
At the end of this phase, a TYPE is partially defined; only the features coming from its parents are specialized.
- Feature collect
At the end of feature collect (see below), the signatures of the features that are not inherited are specialized in the context of their target type.
specialize_thru
- Feature collect
At the end of feature collect (see below), the signatures of the inherited features are specialized in the context of their target type, using their parent type.
specialize_and_check
When a type uses other types, those must have been collected first. In this phase, all the types used by the target type have been collected.
Most semantics checks happen here. Code transformation may also happen (see e.g. issue:78).
collect
Feature collect is the phase where a type collects all its known features. To do so, it is helped by the FEATURE_ACCUMULATOR.
At the end of this phase, a TYPE is considered completely built.
During this phase, new types may be created, which will also be specialized.
When no new type is created, the system is deemed whole.
simplify
This phase, only run in boost mode, tries to simplify the code: inlining, etc.
adapt
This phase transforms the code from the 2.x SmartEiffel core classes to the old 1.x SmartEiffel classes which are used by the generation back-end.