Label truncation performance improvement

[ugurdogrusoz]

In order to improve the performance due to label truncation (when Fit labels to nodes is true), which seems to be an expensive operation, we could truncate them just once at the beginning of the element instantiation, save it in a new data field, and then use that field all the time.

[ugurdogrusoz]

Let's revisit this issue in line with what we discussed please. Briefly the goal here is to set the text-wrap style different than user specifies with the application for performance reasons as follows (how it is in the application per map -> how it should be in Cytoscape.js per node):

• none -> none in this case new data property from which the label is taken is set as the same as the original property from which the label is coming
• wrap -> needs wrapping ? wrap : none in this case, if the label needs wrapping, we calculate the locations of the newline characters and create a new data property that contains the label with new line characters and set the Cytoscape.js wrap style as "wrap". if the label doesn't need wrapping (it's short anyway), we create the new data property to have the same text as the original one and set the Cytoscape.js wrap style as "none" (to avoid expensive computations).
• ellipses -> none in this case, we create a new data property which has the same text as original text (in case it's short anyway) or a different text with ellipses (when it's not short) and we set the Cytoscape.js wrap style as "none" to avoid expensive computations repeatedly.

[canbax]

If we use a simple function to truncate the node labels and let text-wrapping be empty, I see a small performance improvement If we use a complex function (from the source of cytoscape.js) and do the same, I see a huge performance decrease. Currently, we rely on setting text-wrapping: wrap or text-wrapping: ellipsis. I think it is fine because when we provide a simple function for truncation, our function would be not robust for style changes in app_description.json. e.g. if the user sets a very big font-size for some type of node, our simple method doesn't consider font-size or font-family and gives bad results.

[ugurdogrusoz]

Let's try to use a moderate function (as in Newt) to truncate or wrap (in case of ellipses and wrap, resp.) and see if the performance improves.

[ugurdogrusoz]

@hasanbalci Please review and advise suitability for Newt.

[hasanbalci]

For queries, I couldn't see an improvement for "wrap" and "ellipses" options, only new "none" option provides a better performance. During layout, I think text-wrap type doesn't have an effect on simple graphs. It may have an effect on compound graphs but I am not certain on it.

[ugurdogrusoz]

Looks like we ve done something useless and also complicated things. I suppose we should roll back and stick to core Cytoscape.js functionality. It might be nice to keep the related option in setting as radio buttons of three available choices though. Let's talk.

[canbax]

I did some performance tests to see the difference. I see that the pre-calculation of labels performs better than setting text-wrap: ellipsis when I load data first.

When I do layout after the first load I didn't see much difference.

In addition, I tried to add classes to nodes by checking if the label needs to be truncated. For example if a node label is already short does not need to be truncated, don't touch it. This requires checking each node in for loop and adding class if necessary. Since there is for loop, this gave worse results. Since the results are a lot worse, I didn't make tests with this case.

The details of tests are below https://docs.google.com/document/d/1jSw1fHaqxgTWX1ArUzQG9pHnM-MCIdUiU47u8qdEALk/edit?usp=sharing

[ugurdogrusoz]

Results look good. Let's just do a few similar tests on a different machine (Macbook?) as well and then decide.

[canbax]

I made tests with Macbook. Results are similar to previous ones. On the first loads, manual labeling performs better.

[ugurdogrusoz]

Let's close this at least until somebody brings this up as a performance issue.