Detection of caller graphs is slow on first use

[eric-laporte]

In the toolbar of the FSGraph window, there is an icon (with three arrows pointing to a rectangle) to list all the graphs that call the graph displayed in the window. On the first use of this button in a given language, the detection takes more time to complete. From the second use in the same language, the duration of the detection is short and stable, even on another graph or after closing and relaunching Unitex.

What steps will reproduce the problem?

1. Open a language where the icon for detection of caller graphs has never been used
2. Open a graph
3. Click on the icon
4. Open another graph
5. Click on the icon

What is the expected output?

The detection should take roughly the same time to complete in steps 3 and 5.

What do you see instead?

The detection takes more time in step 3 than in step 5. The difference is significant if there are many files in the directory and subdirectories of the language in my workspace. In Korean (4310 files), it takes 45 s on the first time and 5 s on the second time. In Arabic (6884 files), it takes 31 s on the first time and 6 s on the second time. In Modern Greek (59 files), it takes less than 1 s on both times.

It is usual to have thousands of graphs in languages with complex morphology. Users may conclude that the button doesn't work on the day they discover and try it.

More info

• Unitex/GramLab IDE version: 3.2.56-alpha
• UnitexToolLogger version: 3.2.57-alpha
• Did this work before?: the symptom is the same with version 3.1.4314.