The following answer is taken from the official blog post shared by Dr. Manuel Kuehner in the comment.
Besides the following improvements
- remove the need to have the data flow through disks
- write the compiler in C++
- use Apple’s Metal framework to write the characters and shapes of the output directly into graphics memory (rather than creating this intermediate representation)
The game changer is that they manage to do incremental typesetting, as seen in the following quote
Even more excitingly the deeper editor/typesetter integration allowed us to move from the traditional typeset model to “Live typeset” in Texpad 1.8. Rather than starting at the beginning of a document, and working all the way to the end with each typeset, TexpadTeX could consult with the editor and the viewer, and typeset only the part of the document between a user’s cursor, and the end of what is visible in the output viewer. By typesetting just what the user needs to preview their work, this incremental model greatly reduces the time required for a typeset. This allowed us to run a typeset with each keystroke, which in most cases gives the user an almost realtime preview of the output. The downside of moving from a typeset once every few minutes, to more than one a second, is that the extra computation made Texpad thirsty for battery power.
-pvc. – Unknown Jun 07 '22 at 20:26xseconds or afteryseconds of editor activity, and show the result in a preview pane. For a small documents,x=1 secondis almost a true live preview. For large documents that take a long time to compile is better sety >> 1 seconds, or just deactivate the automatic compilation, to avoid compilations after every minimal typing pause. – Fran Jun 07 '22 at 21:26