- QtSearchView was split into QtSearchView and QtSearchBox.
- QtSearchBox contains all Qt elements and can use them purely
- QtSearchView forwards calls from the SearchController to QtSearchBox
- Searching is initiated with MessageSearch to the SearchController
- Autocompletion is initiated with MessageSearchAutocomplete to the SearchController
- The search field is able to create filter queries by only giving autocompletions for the last token in the query
- For named tokens the search field adds their token ids to the query in the form of "A,25" for faster lookup
bug id = #21
This change refactors the GraphController to use the GraphAccess only with a single call that returns a Graph. The Graph
contains all information for visually representing the active Token with all parent and child nodes and edges. This
reduces coupling of the data and logic tiers.
This change stores each Token name in the separate singleton class Dictionary. The Dictionary gives each saved word an
Id and thereby avoids duplicated names. E.g if the constructor method "Graph::Graph" is stored then the word "Graph"
only appears once in memory.
The class SearchIndex is now responsible for the name hierarchy and is instantiated by the Storage. The SearchIndex
builds the name hierarchy using SearchNodes, each holding a Dictionary string reference of the name it holds. E.g if the
names "math::ceil" and "math::floor" are added to the SearchIndex then 3 nodes get created, the SearchNode "math" will
hold the two childs "ceil" and "floor".
The hierarchical graph creation functionality got split off from Graph into the new subclass StorageGraph. The
StorageGraph creates nodes with a passed SearchNode pointer of the name it represents in the SearchIndex. Thereby the
StorageGraph reuses the hierarchical information in the SearchIndex and can create nodes much quicker by avoiding node
searches and name comparisions.
The name information is now stored in the Nodes via the TokenComponentName class, which is subclassed into
TokenComponentNameReferenced and TokenComponentNameCached. The StorageClass creates nodes with the component
TokenComponentNameReferenced, which holds a pointer to the SearchNode instance holding the name. This allows for
retrieving the full name of the node, without using other Nodes int the Graph, which might not be present. If the Node
is copied then the component changes to a TokenComponentNameCached, which holds the full name as a string, so the
memory in the Storage doesn't have to be accessed anymore.
TokenComponentSignature is now only holding an Id of the signature string saved in the Dictionary, which speeds up the
signature comparison. A follow-up will change saving the whole signature as string to reusing the wordIds it is
consisting of.
Lastly the SearchIndex holds basic fuzzy search functionality. A passed query gets compared down the SearchNode
hierarchy as long as matches for each letter are found. Matches must contain all letters of the query. The search is
case-insensitive. If letters are found in front positions, next to each other or written in uppercase they are weighed
higher in the match ranking. The character ':' is also interpreted and found, although the '::' delimiter is not stored.
E.g. the query "m:l" used on the example above will return both "math::floor" and "math::ceil", but "floor" is ranked
higher because the 'l' appears closer to the start.
This change allows the user to use a simple query syntax in the search field to filter the results using names,
operators and predefined filters.
The class QueryTree is capable of parsing a simple query language consiting of QueryNodes, subclassed as
QueryOperators, QueryCommands and QueryTokens. QueryToken identifies a Token by name. QueryCommand represents predefined
filters. QueryOperator defines the syntactic relationship.
QueryNode examples:
"A" -> QueryToken that identifies the Token named A
class -> QueryCommand that filters all Nodes that are classes
. -> QueryOperator that concatenates filters
QueryExamples:
"A".class -> All Tokens named A that are classes
method.(private|protected) -> All methods that are private or protected
"A".member -> All members of A
"A":field.!const -> All fields of A that are not const (':' can be used instead of '.member')
QueryTree checks for correct operator precedence and validity of the query.
All QueryCommands can be found in data/query/QueryCommand.cpp
All QueryOperators can be found in data/query/QueryOperator.cpp
The parsed QueryTree is then passed to the class GraphFilterConductor, which is capable of applying the query on a
Graph. Each part of the query gets assigned a GraphFilter that filters the input Graph in order of node precedence. The
class SubGraph is used as intermediate container, only holding bare pointers to Nodes and Edges. The output Graph holds
all Nodes that match the query.
MessageActivateTokens is used to show all of the results in the CodeView. The GraphView currently only shows the first
Node of the results.
bug id = #6
This change switches function and method parsing to use the structure ParseFunction, which holds all important values of
the function. This structure is then used also in call and usage parsing to allow distinction between signatures.
This change saves only the short names in the node, getName() no only gives the short version. The fullName is assembled
from the parent nodes on demand using the getFullName() method.
This change moves the EdgeComponent system to Token and redefines them as TokenComponents.
* Performance tests on the side showed that using typeid() in getComponent() is faster than dynamically casting every
pointer.
* The hasComponent() method was deliberately left out in the implementation, because it used to look up the right
component and discard it again. This check for presence can also be achieved by just using getComponent().
* The method addComponent() is protected in Token, so that Node and Edge can guard which Component is set on which type
by implementing specific setters for each Component.
* The GraphTestSuite was extended to include tests for the Component implementation (and missing tests for locationIds
in Token and copying of Node and Edge were added).
Function and method nodes are now distincted by their signature when added to the graph. A different node for each
overloaded declaration is created with the signature as member field.
fortune cookie message = Deine harte Arbeit wird schnell belohnt werden.
This change adds a MenuBar to QtMainWindow and adds an action for opening a different project settings XML file. The
message MessageLoadProject is thereon used to load a new project in the Application and clearing the Storage.
This change adds graph data structures Token, Node, Edge and Graph for holding code semantics in the Storage and fills
them with the parsed data. Things to note:
* Token is the base for the classes Edge and Node and holds a unique id.
* Graph is the owner of all Nodes and Edges.
* A Node saves the Edges it is endpoint of.
* An Edge saves the Nodes that are endpoints.
* Nodes and Edges are not subclassed. The different Node and Edge types are distinguished by the enums NodeType and
EdgeType. Type specific fields are therefore available for all types, a check when assigning them prohibits faulty
usage.
* Storage uses the Graph::getNodeHierarchy method, which is not strict about knowing type names beforehand. E.g. if A::B
is passed in and A doesn't exist so far, then node A gets created as NODE_UNDEFINED with B as it's child.
* The Graph methods addNodeAsPlainCopy and addEdgeAsPlainCopy can be used to create a sub-graph without reference to the
original, but with the same ids.
review id = 19
fortune cookie message = Your cheerful outlook is one of your best assets.
Eberhard Graether