Separate AST into direct AST representation of script and an internal representation #49869
Assignees
Labels
Comments
stu-elastic
added
the
:Core/Infra/Scripting
|
Pinging @elastic/es-core-infra (:Core/Infra/Scripting) |
|
Migrated from: #31009 |
This was referenced Feb 3, 2020
jdconrad
added a commit
that referenced
this issue
Feb 11, 2020
This change moves almost all of the customized code required to decorate the execute method into an external phase. This external phase operates only on the ir tree after the user tree has generated the initial ir tree from user specified code. The external phase uses ir nodes to create the necessary customized code instead of writing asm directly. This is a first example of modifying the ir tree to customize it specifically for Elasticsearch and leaves the ir nodes in a more generic state. Another change required for this was to remove the notion of auto-return from the ir tree completely. The user tree is now responsible for generating appropriate ir tree nodes to support auto-return. This is the first example of divergent user and ir trees as the user tree is intended to be higher level while the ir tree is supposed to provide lower level translation into asm. Relates to #49869 Relates to #51841
This was referenced Feb 25, 2020
jdconrad
added a commit
that referenced
this issue
Mar 3, 2020
This change removes the ECast node from the user tree, so that there are no structural changes to the user tree making it structurally immutable. All generated casts are now built directly into the ir tree instead. This also removes the isNull member variable from AExpression because now an instanceof operator can be used instead of having to pass this information between potentially multiple levels of user tree nodes. Relates to #49869
jdconrad
added a commit
that referenced
this issue
Mar 9, 2020
…53075) This is a step toward the long-term goal of making the "user" trees nodes immutable. This change isolates the mutable data for expression nodes in the "user" tree during the semantic (analysis) phase by moving the mutable data into Input and Output objects. These objects are created locally during the semantic phase to share information required for semantic checking between parent-child nodes. Note that for this change, Input and Output are still stored in a mutable way on the expression nodes. This will not be the case once the semantic (analysis) phase and ir generation (write) phase are combined in a future change. Relates #49869
jdconrad
added a commit
that referenced
this issue
Mar 13, 2020
…53348) This is the follow up to #53075, isolating the mutable data on the statement nodes as the referenced change did the expression nodes. This is a step toward the long-term goal of making the "user" trees nodes immutable. This change isolates the mutable data for statement nodes in the "user" tree during the semantic (analysis) phase by moving the mutable data into Input and Output objects. These objects are created locally during the semantic phase to share information required for semantic checking between parent-child nodes. Note that for this change, Input and Output are still stored in a mutable way on the statement nodes. This will not be the case once the semantic (analysis) phase and ir generation (write) phase are combined in a future change. Relates #49869
jdconrad
added a commit
that referenced
this issue
Mar 23, 2020
This is the next step in producing immutable state in the Painless "user" tree. This combines the analyze phase and the write phase so that both semantic checking and the ir tree are generated in the same phase. This allows the removal of the Input/Output objects created in the previous PRs from being mutable state on the nodes. Instead they are now local state for the analyze phase. Relates: #53561 Relates: #49869
|
Closing in favor of #53702 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Painless parses the script into an AST and mutates while processing. Correlation with user provided script is difficult as we don't have the raw representation. Keeping the original AST around without mutating it will enable better error messages and other IDE-like functionality.
The text was updated successfully, but these errors were encountered: