v6.2.0 - Fix DST handling and improve price lookup logic

This release fixes a critical bug related to price display during Daylight Saving Time transitions, implementing timestamp-based lookups for accurate price retrieval. It also includes new and updated functions for better handling of price indices and LED control.
This commit is contained in:
2026-03-29 17:29:43 +02:00
committed by GitHub
parent eb5ef5954b
commit 8543493ac0
+95 -137
View File
@@ -4,6 +4,82 @@ All notable changes to this project are documented here.
---
## v6.2.0 DST (Daylight Saving Time) Handling Fixed (20260329)
**Summary**
This release fixes a critical bug that caused incorrect price display on DST switch days. On March 29, 2026 (the spring forward day), the ticker at 12:41 showed prices for hours 13:00, 14:00, and 15:00 instead of the correct 12:00, 13:00, and 14:00.
### Problem: Arithmetic-Based Index Calculation
**Root Cause (v6.0.0 v6.1.2):**
The code assumed every day has exactly 96 price entries:
```cpp
int startIndex = hourIndex * 4; // e.g., hour 12 → index 48
```
This assumption breaks on DST switch days:
| Day Type | Hours | Price Entries | Example |
|----------|-------|---------------|---------|
| Normal day | 24 | 96 | Array indices 0-95 |
| Spring forward (March) | 23 | 92 | Index 48 points to wrong time |
| Fall back (October) | 25 | 100 | Index 48 points to wrong time |
At 12:41 on March 29, 2026:
- ESP32 correctly reported `timeinfo.tm_hour = 12`
- Old code calculated `12 × 4 = 48`
- But array only had 92 entries (no index 48 that maps to local hour 12)
- Result: Displayed prices for 13:00, 14:00, 15:00 instead of 12:00, 13:00, 14:00
### Solution: Timestamp-Based Lookups
**New approach (v6.2.0):**
All price lookups now search the `unix_seconds` array using actual timestamps:
```cpp
// Find index by searching for matching hour in timestamps
int findPriceIndexForHour(const JsonArray& unixSeconds, int targetHour) {
for (size_t i = 0; i < unixSeconds.size(); i++) {
unsigned long unixTime = unixSeconds[i].as<unsigned long>();
time_t t = (time_t)unixTime;
struct tm* ptm = localtime(&t);
if (ptm != NULL && ptm->tm_hour == targetHour) {
return (int)i; // Found correct index for this hour
}
}
return -1;
}
```
**New functions added:**
- `findPriceIndexForHour()` Finds first price index for a given hour
- `findCurrentPriceIndex()` Finds current 15-minute slot using Unix timestamp
- `getHourFromPriceIndex()` Gets hour from price array index
**Updated functions:**
- `getHourlyAverage()` Now uses timestamp lookup instead of `hourIndex * 4`
- `display15MinuteDetails()` Timestamp-based with hour verification in loop
- `displayPriceRow()` Timestamp-based data index lookup
- `displayPrimaryList()` Timestamp-based current hour detection
- `updateLeds()` Timestamp-based current interval lookup
### Why This Is Future-Proof
| Scenario | Code Behavior |
|----------|--------------|
| Normal days (96 entries) | Works as before |
| DST spring forward (92 entries) | Timestamp lookup finds correct indices |
| DST fall back (100 entries) | Timestamp lookup finds correct indices |
| EU cancels DST | Only update `TZ_CET_CEST` string; code works unchanged |
If EU parliament ever cancels DST switching, you only need to update the `TZ_CET_CEST` line (one line of code). The price lookup logic requires no changes.
---
## v6.1.2 LED Indicator Restored (ESP32 PWM Fix) (20260311)
**Summary**
@@ -15,11 +91,11 @@ This release fixes a regression introduced in **v6.1.1** where the **white LED p
**Problem (v6.1.1):**
- The sketch uses `analogWrite()` to drive the LED with PWM for breathe/blink patterns.
- However, in some LED OFF branches the code used `digitalWrite(LOW)` on the same `whiteLedPin`.
- However, in some "LED OFF" branches the code used `digitalWrite(LOW)` on the same `whiteLedPin`.
- On ESP32 (LEDC), once PWM is attached to a pin, `digitalWrite(LOW)` may **not fully disable** PWM output.
- Result:
- LED could remain **faintly on** (dim glow) when LEDs were supposed to be off (e.g., presence timeout / backlight off).
- Some patterns could appear stuck or inconsistent.
- LED could remain **faintly on** (dim glow) when LEDs were supposed to be off.
- Some patterns could appear "stuck" or inconsistent.
**Solution (v6.1.2):**
@@ -27,7 +103,7 @@ This release fixes a regression introduced in **v6.1.1** where the **white LED p
- Use `analogWrite(whiteLedPin, 0)` instead of `digitalWrite(whiteLedPin, LOW)`.
- Use `analogWrite(whiteLedPin, 255)` instead of `digitalWrite(whiteLedPin, HIGH)`.
- Blink/doubleblink toggles now switch between PWM **0** and **255**.
- This ensures the LED is **truly off** whenever LED output is gated off (no data, no time sync, or presence timeout).
- This ensures the LED is **truly off** whenever LED output is gated off.
---
@@ -35,31 +111,25 @@ This release fixes a regression introduced in **v6.1.1** where the **white LED p
**Summary**
This release fixes a bug where the **daily lowest / highest hourly price marker** ignored negative prices (and also ignored 0.0), which could cause the ticker to incorrectly mark the **lowest positive[...]
This release fixes a bug where the **daily lowest / highest hourly price marker** ignored negative prices (and also ignored 0.0), which could cause the ticker to incorrectly mark the **lowest positive** price as the daily minimum.
### Fixed: Daily Low/High Marker Ignored Negative & Zero Prices
**Problem (v6.1.0):**
- In `processJsonData()` the daily min/max scan used:
- `if (hourlyAvg > 0) { ... }`
```cpp
if (hourlyAvg > 0) { ... }
```
- This had two side effects:
1. **Negative** hourly averages were completely skipped.
2. A true price of **0.0** was also skipped (even though 0 can be a valid market price).
- Additionally, the daily average was computed as `sum / 24.0` even though hours were being skipped from `sum`, making the daily average incorrect whenever any hour was excluded.
**Solution (v6.1.1):**
- The min/max and average scan now:
- Treats an hour as valid based on **data availability** (having all 4×15minute entries), not based on value sign.
- Includes **all values** (negative, zero, positive) when computing:
- `lowestPriceIndex`
- `highestPriceIndex`
- `averagePrice`
- Computes `averagePrice` using the number of valid hours (normally 24).
- Includes **all values** (negative, zero, positive).
---
@@ -67,112 +137,21 @@ This release fixes a bug where the **daily lowest / highest hourly price marker*
**Summary**
This release fixes a bug where the ticker could remain indefinitely on the **No data for today** screen after midnight, even though the API was already returning fresh data. It also refines the [...]
This release fixes a bug where the ticker could remain indefinitely on the **"No data for today"** screen after midnight, even though the API was already returning fresh data.
### Fixed: Stuck on NO_DATA_OFFSET After Midnight
**Problem (v6.0.0):**
- After midnight, the device:
- Detected day rollover and entered a “no data” state.
- Scheduled an immediate API fetch.
- However:
- The API can continue to serve **yesterdays market day** for some time after local midnight.
- The code only checked the **first** `unix_seconds` entry against the current local day.
- HTTP + JSON success always incremented `apiSuccessCount`, even if `processJsonData()` subsequently decided the dataset was “not for today”.
- The scheduler treated such fetches as **successful**, pushed `nextScheduledFetchTime` 24 hours into the future, and never retried.
- Result: the ticker stayed on **NO_DATA_OFFSET** forever, until:
- A manual longpress triggered a fresh fetch at a time when the API finally returned recognized “today” data, or
- The device was rebooted later in the day.
- The API can continue to serve **yesterday's market day** for some time after local midnight.
- HTTP + JSON success always incremented `apiSuccessCount`, even if the data was "not for today".
- The scheduler treated such fetches as **successful** and never retried.
**Solution (v6.1.0):**
1. **Market Day Detection**:
- `processJsonData()` now determines the “market day” using the **last** `unix_seconds` timestamp from the APIs dataset (assumed to cover one full day in 15minute steps).
- It compares that calendar date (local time) against the current local date.
- If they differ:
- The dataset is treated as **“not for today”**.
- `isTodayDataAvailable = false`.
- The function returns **false**, and a new flag `lastProcessJsonAcceptedToday` remains `false`.
2. **Logical Failure vs HTTP/JSON Failure**:
- New global flag:
- `lastProcessJsonAcceptedToday` `true` only when `processJsonData()` accepts the dataset as “todays” data.
- In `handleDataFetching()`:
- A scheduled fetch is considered a **real success** only if:
- HTTP + JSON succeeded, **and**
- `lastProcessJsonAcceptedToday == true`.
- In all other cases (including “HTTP 200 + parse OK but data still for yesterday”):
- The fetch is treated as **failure** for scheduling purposes.
- If `midnightPhaseActive == true`, `scheduleAfterMidnightFailure()` is invoked to plan a retry.
3. **Midnight Phase Cleanup**:
- When a fetch finally provides a dataset for today:
- `isTodayDataAvailable = true`.
- `lastProcessJsonAcceptedToday = true`.
- `midnightPhaseActive` is cleared; `midnightRetryCount` reset to 0.
- `nextScheduledFetchTime` is set to 24 hours ahead.
As a result, the ticker will **keep retrying** after midnight until a correct marketday dataset appears, instead of giving up after the first HTTP 200.
---
### Changed: AfterMidnight Retry Schedule
In `scheduleAfterMidnightFailure()` the retry strategy when `midnightPhaseActive == true` has been tuned.
**Old behavior (v6.0.0, conceptual):**
- First few failures after midnight:
- Retried every 10 minutes up to 5 attempts.
- Afterwards:
- Switched to hourly retries (top of each hour).
**New behavior (v6.1.0 + user configuration):**
- Fast retry phase fully contained within the **first hour after midnight**.
- You configured:
```cpp
if (midnightRetryCount < 2) {
// Retry every 20 minutes for first 2 attempts (~1 hour window)
midnightRetryCount++;
nextScheduledFetchTime = now + 1200; // 20 minutes
debugPrint(2, "Midnight retry " + String(midnightRetryCount) + "/2 in 20 minutes");
} else {
// After that, retry only at top of each hour
...
}
```
- Timeline:
- 00:00 initial attempt (triggered by day rollover).
- If dataset is still for previous day:
- 1st retry at ~00:20.
- 2nd retry at ~00:40.
- After that:
- Retries only at the **top of the next hours** (01:00, 02:00, …),
until a dataset with market day = today is accepted.
This configuration significantly reduces overnight API load while still ensuring the ticker picks up the new day as soon as the API publishes it.
---
### Other Behavior (Retained From v6.0.0)
- **NVS caching** of daily data:
- On boot, if NVS contains a dataset whose date matches todays local date:
- The JSON is deserialized and processed.
- A new API call is **skipped**.
- After each accepted “today” fetch:
- Raw JSON payload, date, and a timestamp are stored in NVS.
- **Manual longpress refresh**:
- Still triggers immediate fetch via `nextScheduledFetchTime = now;`.
- Now also respects the improved “today” detection; “yesterdays” data is not accepted as today.
- **CET/CEST time handling**:
- Unchanged, still uses `TZ_CET_CEST` with `configTzTime`.
- **Secondary menu and NVS status lines**:
- Retained from v6.0.0; updated only for version string and minor wording.
- `processJsonData()` now determines the "market day" using the **last** `unix_seconds` timestamp.
- A scheduled fetch is only successful if `lastProcessJsonAcceptedToday == true`.
- Midnight retry logic keeps trying until valid "today" data is received.
---
@@ -184,30 +163,9 @@ First major redesign focused on reducing API traffic and improving resilience us
### New
- NVS namespace `"my-ticker"` introduced with keys:
- `ssid`, `pass` WiFi credentials.
- `data_day`, `data_mon`, `data_year` stored data calendar day.
- `data_prc` raw JSON string from API.
- `data_last_store` Unix time when data was stored.
- Boot behavior:
- Try to load and validate NVS data.
- If date matches today → reuse it and **skip** initial API call.
- Aftermidnight behavior:
- NVS is overwritten with each successful newday dataset.
- InRAM data for yesterday is invalidated at day rollover.
### UI / Menu
- Primary list:
- 15minute detail for current hour.
- Hourly averages for current + next 2 hours.
- Secondary list:
- Expanded to 20 lines to include:
- Time/date, last update, daily average.
- WiFi RSSI, IP address.
- API success rate, uptime.
- NVS status block.
- Credits & version line.
- NVS namespace `"my-ticker"` with WiFi credentials and daily price data caching.
- Boot behavior: Try to load and validate NVS data; if date matches today → reuse it and **skip** initial API call.
- Aftermidnight: NVS is overwritten with each successful newday dataset.
---