It might help to remove C5 (for faster IO0 rise time).

The assumption is probably that most users prefer a more accurate clock over a faster wakeup, especially when using the internal oscillator.

You're better off skipping calibration in your case.

You could convert the instructions (after untangling the macros) to normal ULP assembly. There's not much to the touch stuff.
hulp_dump_program might be useful as an intermediate.

I had the same problem and used this: https://github.com/espressif/esp-idf/pull/4352
Although if I was doing it today I'd simply put a callback in esp_light_sleep_start (just before it returns). Hopefully IDF will have something like that in future.

According to https://github.com/espressif/esp-idf/blob/8131d6f46d690e9cb60cc5cd457863cc5479351f/components/esp_rom/include/esp32/rom/gpio.h#L228-L242, 0x100 is the magic number to disconnect the I2C signal from the previous pins. eg. // Disconnect previous gpio_matrix_in(GPIO_NUM_18, 0x100, 0, 0); g...

READ_RTC_REG(RTC_CNTL_TIME0_REG+2, 0, 16)

This doesn't look right. Are you trying to read the upper 16 bits?: READ_RTC_REG(RTC_CNTL_TIME0_REG, 16, 16)

There's no (reasonable) way to actually modify register write instructions at runtime, but if this is representative of what you need to do then it might be feasible to include *all* possible instructions and jump to the one you need. eg: // RTCIO 14 move r0 14 // Each RTCIO requires 2 instructions ...

A new member was recently added to i2c_config_t: https://github.com/espressif/esp-idf/blob/c13afea635adec735435961270d0894ff46eef85/components/driver/include/driver/i2c.h#L66 You can initialise to 0 for default behaviour (or, even better, initialise the entire struct; any new additions like this wil...

Possibly addressed by this recent fix? https://github.com/espressif/esp-idf/commit/27aa6c289f90898cb227c3f93c9ba2f068466ab7 That was targeting light sleep, but I suppose if the ULP is waking frequently enough then it could result in a measurable increase in power consumption in deep sleep too (is VS...