This bug provides further information on the problem reported in this tweet:
  https://twitter.com/tomwpayne/status/492404097883336704

Situation: user code that worked reliably for at least 16 months (and maybe as long as several years) breaks with 3.17, but works with 3.16. The console stack trace (in Chrome) is:

Uncaught TypeError: Cannot read property 'remove' of undefined %7Bmain,geometry,places%7D.js:formatted:2600
(anonymous function) %7Bmain,geometry,places%7D.js:formatted:2600
Ge %7Bmain,geometry,places%7D.js:formatted:553
Q.(anonymous function).bindTo %7Bmain,geometry,places%7D.js:formatted:583
(anonymous function) xcplanner:159
y xcplanner:16
Z.lc xcplanner:159
Z.setPath xcplanner:167
Fg xcplanner:190
Eg xcplanner:175
Rg xcplanner:198
(anonymous function) xcplanner:200
(anonymous function) VM236:5
(anonymous function) VM236:3
VP VM236:1
(anonymous function) VM236:3
(anonymous function) VM234:82
(anonymous function) VM234:65
(anonymous function) VM234:82
(anonymous function) VM234:85
Au.d VM234:77
(anonymous function) AuthenticationService.Authenticate?1shttp%3A%2F%2Flocalhost%3A8080%2Fxcplanner&5e1&callback=_xdc_._…:1

Possibly pertinent information:
- The user application is compiled with the Closure Compiler, the code *does* work with 3.17 in uncompiled mode, and only fails in compiled mode.
- The code is a user implementation of an editable PolyLine, and seems to be related to updating a PolyLine's path (represented as a MVCArray of LatLng objects).
- The user code uses a custom map type.

There's not much information in the changelog regarding 3.16 -> 3.17. Is there possibly an internal change, maybe related to inheritance, that could have caused this?

On my side, I'll continue to investigate this. It's entirely possible that my code was inadvertently relying on an undocumented feature. Please feel free to close this bug immediately - I can always re-open it if I become more confident that it's a regression in the Google Maps API.