Sign-magnitude is the solution for glitch problem and not linearity. With less accurate resistors the linearity remains less precise even with sign-magnitude, only distribution is different.
Full 8-bit sine with random 10% resistors:
sign magnitude glitch: https://ctrlv.cz/shots/2020/06/07/pugl.png (smaller glitches through digital "change", easier to filter out)
sign magnitude steps detail: https://ctrlv.cz/shots/2020/06/07/PSFz.png
standard R-2R glitch: https://ctrlv.cz/shots/2020/06/07/yl4r.png (huge glitch, problem to filter out)
standard R-2R steps detail: https://ctrlv.cz/shots/2020/06/07/zLWw.png (small glitches are due to small current capability of switches, ignore it, only the one huge through digital "change" is important)
The rule remains ... resistors with 0.1% accuracy is enough only for 10-bit R-2R DAC,
0.0015% for 16-bit (about the limit for a classic R-2R discrete dac including switches resistance),
0,000006% for 24-bit (far away from physical possibilities, so above 20-bits other tricks are used in the R-2R chip dacs)
Full 8-bit sine with random 10% resistors:
sign magnitude glitch: https://ctrlv.cz/shots/2020/06/07/pugl.png (smaller glitches through digital "change", easier to filter out)
sign magnitude steps detail: https://ctrlv.cz/shots/2020/06/07/PSFz.png
standard R-2R glitch: https://ctrlv.cz/shots/2020/06/07/yl4r.png (huge glitch, problem to filter out)
standard R-2R steps detail: https://ctrlv.cz/shots/2020/06/07/zLWw.png (small glitches are due to small current capability of switches, ignore it, only the one huge through digital "change" is important)
The rule remains ... resistors with 0.1% accuracy is enough only for 10-bit R-2R DAC,
0.0015% for 16-bit (about the limit for a classic R-2R discrete dac including switches resistance),
0,000006% for 24-bit (far away from physical possibilities, so above 20-bits other tricks are used in the R-2R chip dacs)