[renderTextLayer] Decouple computation and DOM operations

Currently renderTextLayer will compute the different properties (left, top, font-style, scale…) using canvas.measureText() and build the result view using vanilla DOM operations.

I’m working on an Angular PDF viewer based on pdf.js engine and not just a dummy wrapper. I would like to know if it possible to either :

1. abstract those DOM operations inside renderTextLayer
2. to extract the computation part so we could build our own renderTextLayer.

1) would most certainly result in an API break, while 2) could be contained on internal stuff and new API.

I can see at least 2 benefits :

• We could avoid the DOM rebuild and just update the style of existing nodes on certain cases (when we zoom on the page for example).
• We should* be able to make this work with server side rendering. (* still not sure about how to handle measureText() on node).

I will propose one or several implementation if you approve the idea.

[ghetolay]

Should I just start the PR ?

[Snuffleupagus]

I really don’t think that we want to complicate and/or possibly slow down the general textLayer building by introducing e.g. a callback function, provided from the outside, for handling text measurement. Furthermore, with somewhat recent changes, note that we’re now also relying even more on the browser to obtain information about fonts used for the textLayer; see e.g. https://github.com/mozilla/pdf.js/blob/d285580f486178a41a316d615f3157ec7e1afcb3/src/display/text_layer.js#L51-L116

---

This is probably the sort of thing that belongs in a custom implementation/fork of the PDF.js library, given the somewhat specialized use-case outlined in https://github.com/mozilla/pdf.js/issues/8575#issue-238494962. All-in-all, I do believe that this issue as filed should now be closed as WONTFIX.