This is an interim release. Do not be alarmed if issues with installation are encountered.
- New toolchain version.
- Make wiring.c have the functions referred to in the doc.
- Split the three ugly functions that constituted 60% of wiring.c into a separate file, because finding anything else with those #ifdef-hell-spawned abominations is awful.
- General cleanup of wiring.c and timers.h for maintainability.
- Fix some of the constants for timers so that all timers can still get uniform codes specifying the portmux and (for non-TCA's) which pin within the mux it is, which matters for the other core.
- Documentation improvements.
- Study of the state of 20 MHz timekeeping is ongoing. By my calculations it resulted in time passing 1/51th too fast
- Corrected critical bug in the TCD0 millis option (the default on 1-series parts) (#1243)
- Corrected bug sometimes encountered when using serial under atypical cases the causes of which remain mysterious), where it would complain about
__poll_dre_done. (#1226) - Correct bug with Comparator (#1236)
- Correct numerous issues with the microchip board defs. Clearly nobody is using them - since 2 of them didn't compile, and none of the optiboot ones would ever be able to successfully be programmed over the bootloader because selfprogramming wasn't enabled (BOOTEND = 0).
- Major Bugfix: Correct issue #1164 thanks to John Vasileff. This would cause Serial.flush() to hang in one-wire mode.
- Major Bugfix: Correct buffered PWM on all parts.
- Major Bugfix: Correct PWM on non-default pins on all parts; digitalPinHasPWM() macro.
- Fix bug with CCL clock source selection options.
- Correct some ambiguities with the clock specification defines for RTC Millis using external clock; we don't support it,
- Enhance documentation
- Correct critical problem with timekeeping on parts with a TCD. (#1243)
- Removed example of exactly the sort of boneheaded and strictly-worse code I rail against in a trivial part of Wire. While it was merely a bad example at the time it was written, it is now known to be a bad, dangerous example, because it featured code that demonstrated Strictly Worse access to registers, by loading a vport register to a pointer register, and using that - not only is it stupid and just a less-featureful PORT when accessed that way, avoiding pointer access to the I/O space (64 addresses, - 28 VPORTs and 4 GPIORs in low I/O and SP, RAMPZ (if applicable), CCP (the register used for
__PROTECTED_WRITE()}), and of course the SREG) is necessary due to an erratum effecting all AVRxm parts. And indeed, none of those are registers that anyone should ever access using a pointer, even if it didn't wasn't safe against an incredibly scary bug - the low I/O is all about the bitwise instructions, which need both the register and bit compile time known.VPORT_t*is not a type you should ever declare - if you have had to declare it, you get no benefit from faster access to it (like you do if you use the compile-time known VPORT's and access them without pointing. ) - Enhance error messages from wrong-part on upload.