Performance improvement for QR rendering

[danielpanero]

It takes a lot of time to render a lot of QR Bills since addPath is an expensive function as we need to convert from QR segments to SVG and then to PDF.

We can bypass the conversion from QR to SVG, and directly convert QR segments to PDF.

This leads to huge performance benefits (a4.js):	Number of bills	10	100	1000	10000
Normal	754.651ms	5.426s	50.278s	Node crashed :(
Modified	278.702ms	969.642ms	6.985s	1 min 7.421s

[danielpanero]

Furthermore, the second most expensive function is the getPenaltyScore() found in qr-code-generator.ts, that checks which of the 8 mask patterns is the most effective one...

const qrCode = qrcodegen.QrCode.encodeSegments([eci, ...segments], qrcodegen.QrCode.Ecc.MEDIUM, 1, 40, 0);

If we impose mask 0, this leads to again to huge performance gains:	Number of bills	10	100	1000	10000
Normal	754.651ms	5.426s	50.278s	Node crashed :(
Modified	278.702ms	969.642ms	6.985s	1 min 7.421s
Modified V2	157.408ms	586.346ms	3.679s	34.487s

[schoero]

Thank you for further researching and testing performance bottlenecks. It sounds good, but I'm not sure if it's the best idea to hardcode the masking pattern. After all, the readability of the QR code will suffer in most cases. Do you have a more in-depth knowledge about QR codes and do you know if this is safe to do?

I don't really want to trade readability for performance.

[danielpanero]

Yes, that's the problem... for 10/100 bills the performance gains are not justifiable as I think it's better to lose some time improving the readability of the code. After 1000 I don't know...

[schoero]

I checked the specifications again. There is no mention about the masking patterns. But there are rules to minimum module size and error correction etc, so I would like to play it safe and stick to the expensive calculation for now.

Im going to merge, feel free to open an issue if you want to further discuss this topic.