* Network: force array order when sorting hierarchical levels
Fixes #340r34.
If coordinates are not available to sort within a hierarchical level, sort to array order instead.
The previous fix on this issue was not good enough to circumvent this quirk in the chromium sorting. This should bury it.
* Added TimSort for sorting in DirectionStrategy
* Added TimSort to LayoutEngine
* Fixed typo
* Add template for document generation with 'jsdoc'.
In essence, it defines the subdirectory `docs` as a `jsdoc` template. Benefits:
- allows the usage of partials, in order to DRY common parts of the html files.
- makes available the jsdoc-comments, for addition into the documentation.
- enables extraction of data from the source code. For example, the list of edge endpoints `['arrow', 'bar', 'circle']` can now be extracted from the source and inserted into the documentation on generation.
In this initial version, the only file that has been changed is `docs/data/dataset.html`.
In here, partials have been added to illustrate how common page elements can be DRY'd.
The template has been set up in such a way, that resource files will be copied and that html files can pass through unchanged if no special template tags (`<?js...?>`) are used. This allows for a gradual transition of the html files to templates.
**Usage:** `gulp docs`
- The result files are placed in subdirectory `gen/docs/`.
**NOTE:** The release procedure will have to be adjusted by adding this gulp command.
The docs-files will then have to be taken from `gen/docs`.
* Edits to docs/README
* Adjusted layout of README.md
* Further edits to README.md
* Removed pre tags again in README.md - don't work in code block
* Linted the gulpfile
* Added proof of concept for docs generation from source
* Network: DRY code in Label for parsing markup
This gets rid of a major eyesore for me. The accumulator object was identical for HTML and Markdown.
In addition, the parsing has been refactored. Common elements have been DRY'd and the logic of the parsing has been made more comprehensible.
* Added suggestion @mbroad wrt regexp precompile
* Fixed linting
* Network: preload images in options for all shape types
Fixes#3532
If the images option is set, preload it, even if the node shape is not `image` or `circularImage`.
This needs to be done because the user can switch to an image shape later, as is happening in the issue.
Option `image` is only mandatory for the image shapes.
There is no unit test for this, because it would need a mock object for Image and I didn't succeed in creating/finding one.
* Network: Handle null data gracefully
During testing I discovered that passing `null` for network data leads to a thrown error.
This adds a guard to prevent it, plus unit tests for regression.
Fixes#3532
If the images option is set, preload it, even if the node shape is not `image` or `circularImage`.
This needs to be done because the user can switch to an image shape later, as is happening in the issue.
Option `image` is only mandatory for the image shapes.
There is no unit test for this, because it would need a mock object for Image and I didn't succeed in creating/finding one.
* Network: Prevent crash when dataChanged is triggered during initial setting of options
Fixes#3562.
Options `hidden` and `physics` can emit a `_dataChanged` event within `setOptions()`.
If this happens when setting options during the initialization of the `Network` instance, this leads
to an error thrown, because there is no DataSet instance connected yet to the instance.
This bug was introduced in `v4.21.0`. Unit tests have been added for this case.
* Edited comment
Previous condition was too strict: if *any* node for the cluster was already clustered, the clustering would abort.
Current fix scans for already clustered nodes and proceeds with what is left.
* Network: Retain constraint values in label font handling
Fixes#3517.
Due to changed logic in the label font handling, the option values for `widthConstraint` and `heightConstraint` were overwritten.
The fix is in effect a reversal of two code lines: parsing constraint options should come *after* parsing (multi)font options.
Further changes:
- Additional 1-liner fix: constraint values were not copied for edge-instance specific options.
- Small refactoriing of `Label#constrain()` in order to separate concerns
- Added unit test for regression testing of this issue.
This leads to the curious observation that, while the actual change is two lines of source code, this resulted in a +-150 line regression test.
* Made unit test more linear, removed tabs
* Made 'enhanced subset' of unit test
* Removed TODO from comment
* Culled redundant nodes from unit test
* Network: Add extra check on null value during label handling
This fixes an oversight in #3486. Unit tests added, not only for null labels, but for all weird label values I could thing of.
* Enhanced unit tests, adjusted label check
* Small refactoring in Label._drawBackground()
* Added size calculation to Label, basic framework for detecting click on label.
* First fully working version of label click.
* Put in extra checks, refactored visibility of labels for more general usage.
* Final fixes to code; added to example and in docs
* Adressed review comments
* Add next attempt to fix Travis unit test bug
* Addressed review issues
Fixes#3483.
If `shape: 'image'` is defined as option, this was used for initializing hidden cluster nodes during layout initialization.
However, because no image is being passed, this led to an exception.
This fix prevens shape `image` from being used for hidden cluster nodes. In addition, some extra fields for these nodes are
now overridden for some performance improvament.
* The next fix on Travis unit test failure
This is the next escalation on the war against the Travis unit tests failing (which came into being by yours truly).
By accident, I could recreate the unit test failure on my development machine. This led to a more directed effort to
squash the bug.
The insight here is that test `(window === undefined)` fails, but `(typeof window === 'undefined`)` succeeds. This undoubtedly has to do with the special status `window` has as a global object.
Changes:
- Added check on presence of `window` in `Canvas._requestNextFrame()`, fixed local source errors.
- Added catch clause in `CanvasRendered._determinePixelRatio()`
- small fix: raised timeout for the network `worldCup2014` unit test
* Preliminary refactoring in utils.js
* Added unit tests for extend routines, commenting and small fixes
* More unit tests for extend routines
* - Completed unit tests for extend routines in
- Small fixes and cleanup in `util.js`
- Removed `util.protoExtend()`, not used anywhere
* Added unit tests for known font options
* Interim save before trying out another proto chain strategy
* Fixed problem in first example #3408
* Removed silly file that shouldn't be there
* Added unit test for multi-fonts
* Comment edits
* Verufy unit tests, small adjustments for groups
* Further work on getting unit tests to work. PARTS NEED TO BE CLEANED UP!
* Further tweaks to get unit tests working
* All unit tests passing
* Fixes due to linting
* Small edits
* Removed prototype handling from font pile
* Fixes during testing examples of #3408
* Added unit test for edge labels, small fixes
* Added unit tests for shorthand string fonts; some tests still failing
* All unit tests pass
* Removed Label.parseOptions()
* Completed shorthand font tests, code cleanup, fixed choosify for edges
* Addressed review comments
* Addressed review comments, cleanup
* Added unit test for Array.prototype mangling - first passing version
* Enhanced unit test for prototype stressing to catch more illegal for-in loops
* Fixed all for-in linting violations
* Add endpoint 'bar' to Network
Fixes#3412
- Adds class `Bar` and handling to `EndPoints.js`
- Adjusted `options.js` for endpoint, minor code changes
- Added to example `network/edgeTypes/arrowTypes`.
- Added to edges documentation
* Grumbl fix travis unit test yet again. Go die already!
* Network: Fixes sorting on Layout sorting, refactoring.
Fix for #3403
Chromium has divergent behaviour on sorting of `undefined` values, a check has been added for these.
The fix itself is small, it adds a check on `undefined` in the sorting function(s).
In addition:
- Fixed off-by-one error on tree index
- Refactored away checks on visibility by replacing `_isVisible()` with a Strategy pattern for direction
The latter removes a long-standing eyesore in `LayoutManager`; it's probably also faster.
There is some hope that these fixes will improve the hierarchical layout initialization, which has had
problems for quite a while.
* reformat code block
* Add next attempt to fix Travis unit test bug
* Fix unit test - timeout
* Adjustments for comment reviews
* Moved direction strategies to separate module; added fix for 'window is undefined'
This is the next escalation on the war against the Travis unit tests failing (which came into being by yours truly).
By accident, I could recreate the unit test failure on my development machine. This led to a more directed effort to
squash the bug.
The insight here is that test `(window === undefined)` fails, but `(typeof window === 'undefined`)` succeeds. This undoubtedly has to do with the special status `window` has as a global object.
Changes:
- Added check on presence of `window` in `Canvas._requestNextFrame()`, fixed local source errors.
- Added catch clause in `CanvasRendered._determinePixelRatio()`
- small fix: raised timeout for the network `worldCup2014` unit test
* Moved endpoint-specific code to module EndPoints
* Further decomposed functionality for endpoints
* Added commenting, notably for typedef's
* Add next attempt at fixing travis bug
* Added default sizes for nodes without labels
* Next attempt at fixing travis problem
* Fix on calculation of multifont labels - the height was calculated too large.
* added hexagon shape to the node.
* Updated function for hexagon shape.js
* Modifid the shape login for hexagon
* Remove this.translate from the Shape.js for hexagon
* updated hexagon draw logic
* Fixed code review comments and update the branch
* Updated changes in test.js
* Corrects usage of @static jsdoc annotation
* Correct unresolvable types
* Correct types in jsdoc and remove extraneous @class & @constructor jsdoc
* Remove incorrect @static jsdoc
* Adds missing jsdoc for param copyFromGlobals
* correct jsdoc in util
* Corrects casing on jsdocs
* Swaps @inheritDoc to @ignore for constructors where constructor args are documented in the class
* Instantiates Errors with `new`
* Addresses improperly defined @callback tags.
* Split callbacks out to separate jsdoc
* Moves constructor jsdocs back to constructor and drop @ignore.
* First fix for opening clusters, added unit tests
* Added opening of child cluster to parent cluster
* Added more tests for multi-level clusters
* Commenting fixes
* Added unit tests for option allowSingleNodeCluster, fixes due to that
* Added test for allowSingleNodeCluster with nested clusters
* Fixes due to linting
* Removed TODO from code
A recurring problem during travis tests is that global variable `window` can get reset while `Network` tests using a mock canvas object are running.
This issue has been adressed several times, but it still happens. This PR reduces the possibility of it happening again to a minimum.
PR's affected by this issue should be merged with this fix and re-submitted. At time of writing, these are:
- #3402
- #3405
- #3399
Chances are it will occur sporadically for other PR's as well.
Labelling this `High Priority` because it indrectly affects unrelated PR's.
Fix for #3367
Changes:
- Clustered nodes with shared connections are added to the same cluster
- joinCondition value `null` handled as `undefined`
- Fixed bug on adding clusters if joinCondition present and returns true
Unit tests have been added to `Network.test.js` for these changes.
* Added commenting for options, refactored choosify()
* Refactored updateLabelModule()
* added comment at node creation
* Completed fix for #3350
* parseOptions() different handling for bridged/extended; unit tests for reversal of options
* Fixes on linting and unit testing
* Enables require MethodDefinition for require-jsdoc lint rule and corrects 66% of missing Method Definitions
* Adds jsdoc for all remaining methods and corrects @class/@constructor documentation
* Define values more accurately
* Correct bugs that prevent jsdoc generation
* Add support for edge style
* Update for TODO: support of solid/dotted/dashed edges (attr = 'style')
* Edge styles are defined as var edgeStyles in parseAttributeList().
* Add example for edge styles
Add an editable example for playing DOT edge styles
* Correct typo and remove TODO description
* Correct typo of filename from 'dotEdgeSytles.html' to
'dotEdgeStyles.html'.
* Remove TODO description of the issue for edge style
* Enable unit testing for module Network
Adds mocks for certain components, in order to let module `Network` be run in unit tests.
Changes:
- Create a mock object for `hammer.js` when running browserless. The inspiration is taken from [here](https://github.com/uber/deck.gl/pull/658)
- Create a mock function for `window.requestAnimationFrame()` when running browserless in `network/modules/CanvasRenderer.js`
- Added unit test for `Network` to show that unit testing for it now works
- Fixed naming of container in `test/Graph3d.test.js`
Since `hammer.js` is also used in other modules, this fix is potentially an enabler for full-module unit tests for those as well.
* Cleanup unit test Network
* Fix for unit test
* Added example networks to unit test
* Fix error in loading disassemblerExample
* Network unit test final fixes
* Adjusted comments
* First working version of updating clustered edge
* Added fix for #1315 as well
* Enable unit testing for module Network
Adds mocks for certain components, in order to let module `Network` be run in unit tests.
Changes:
- Create a mock object for `hammer.js` when running browserless. The inspiration is taken from [here](https://github.com/uber/deck.gl/pull/658)
- Create a mock function for `window.requestAnimationFrame()` when running browserless in `network/modules/CanvasRenderer.js`
- Added unit test for `Network` to show that unit testing for it now works
- Fixed naming of container in `test/Graph3d.test.js`
Since `hammer.js` is also used in other modules, this fix is potentially an enabler for full-module unit tests for those as well.
* Cleanup unit test Network
* Added unit test for fix issue #1218
* Adding test for #1315 - Interim save
* Completed unit test and fixes for #1315
* Added fixes for #1291
* Added unit test for #1219
* Added header comment for Clustering.js, small fixes
* Fix for unit test
* Added example networks to unit test
* Fix error in loading disassemblerExample
* Network unit test final fixes
* Fixes for linting
* Fixed essential typo
* Fixed linting error
* Fixed unit test
* Fixed unit test again
* Makes network eslint compliant
Unused variables in private functions, who were never called internally with those arguments were removed.
lint rule was disabled for public functions, and in private functions which were called internally with those unused arguments
* Disables linting for unused args passed into DataSet and DataView
* Levels of direct hierarchical network only incremented
* Cleaned up old code
* Quick fix on presence globaOptions in mergeOptions()
* Revert fix, doesn't work
* Network: Block recalculation of level in LayoutEngine._determineLevelsDirected()
Fix for #2311.
Nodes with bidirectional edges got their levels shifted due to the handling of both edge directions.
This fix adds a check on bidirectionality and blocks any subsequent level adjustment.
Pure tree layouts are unaffected by this change.
* Code cleanup, for better understanding
* Further refactoring; text processing to blocks in separate method
* Added unit test for labels - tests standard text and html tags
* Labels added unit tests for markdown
* Further refactoring; made multi and regular handling more congruent
* Interim save, not there yet
* Unit tests done, first working version
* Added test case with two big words
* Code cleanup
* Break up huge words into lines.
* Restore unrelated code change
* Update hierarchy when node level changes
Fix for #3220
On change of data of an existing node, the level is checked for changes.
If changed, for a recalculation of the hierarchical layout.
The fix does not explicitly check for hierarchical layout; this should not be a problem.
The added code can be used to add further node fields which may trigger recalculation of layout.
* Changes due to review
* First interim commit
* Fixes during testing
* Allow multiple edges to be hidden by a clustered edge.
Fix for #3245.
This fix adjusts the clustering edges so that theyi can refer to multiple edges instead of just one.
This API method is now insufficient, since multiple base edges can be returned.
- Added replacing method `clustering.getBaseEdges()`
- Adjusted example `changingClusteredEdgeNodes` for the new method. This is the *only* place where `getBaseEdge()` was used
- Adjusted documentation for new method and deprecation old method.
Method `getbaseEdge()` should now be considered `deprecated`, and in due time should be removed.
* Edits of method name in example
* Edits of method name in example
* adjusted deprecation method getBaseEdge() in docs
* Adjusted deprecation method for getBaseEdge() in docs
wimrijnders
justinharrell
Ondřej Hruška
macleodbroad-wf
Ravindra Gupta
Yasufumi Ogawa