This completes re-work of all JSON buffer memory management to use HAPOut.
Next up: Stress test with Max Accessories to see how many Accessories we can now handle.
This removes last of large-sized JSON buffers - everything has now been converted to using hapOut.
Next up: Must add prettyPrint to hapOut to support 'd' command
Hashing of entire database is now done piecemeal within HapOut, preventing the need to load the entire database into a large buffer just to obtain a unique hash.
Modified updateDatabase() to use HapOut.getHash()
New class-level method SpanPoint::setEncryption(boolean encrypt). Default is to use encryption unless method is called with encrypt=false.
With encryption, 7 connections max; without encryption, 20 connections max.
Serial Monitor provides warnings if connections are exceeded.
Stylistic only. Cannot remove .get() when a TempBuffer is being re-cast. Also can't remove .get() when cast can't be determined, such as when used in Serial.printf() [potential to do: explore if this can be addressed, though this is stylistic only and has no impact on code]
New Operator now uses PSRAM if available, else uses Internal RAM. Obviates the need for the ps_new() macro, though will keep definition since it may be useful for any class that does not already override malloc() to use PSRAM.
Reduces NVS consumption by from 3 to only 1 records when storing numerical values.
Also: Fixed memory problem in getCharacteristics by replacing dynamically-sized stack buffers with TempBuffer (must do this everywhere), which was causing stack probllem when combination number of Accessories and number of Characteristics got very large.
When using PSRAM, changing the number of Accessories no longer creates any change to heap usage on internal RAM. The entire Accessory will be stored in PSRAM.
Forces all heap requests to be drawn from PSRAM, unless the device does not contain PSRAM.
To do: Provide easy-to-use NEW() macro; add custom allocator to unordered sets; auto-shrink vectors after updateDatabase()
Significantly reduces memory usage - unordered_sets take up much more memory and using std::find from <algorithm> is as efficient as using find() method of an unordered_set.
To do: Remove all uses of unordered_set across HomeSpan wherever possible
An optional callback used to track the number of times the device reboots, where a reboot is defined as any restart where the uptime was less than a specified number of milliseconds (default=5000)
Low Memory Watermark is compared to DEFAULT_LOW_MEM_THRESHOLD (currently set at 80,000 bytes) after HAP initialization but before WIFI connection. If Low Memory Watermark is below DEFAULT_LOW_MEM_THRESHOLD, a WARNING is issued.
Note: memory is based on heap with MALLOC_CAP_DEFAULT capabilities
Adds a second type of WiFi Callback that is called every time WiFi is established OR re-established after a disconnect. Passes the number of times WiFi has been connected as an argument.
Changed fixed-size array `struct Controller[MAX_CONTROLLERS]` to a dynamic linked-list of Controllers. Re-coded all related functions. MAX_CONTROLLERS no sets the size of any buffers but is only used to limit the ultimate size of the linked-list.
Saved about 1K of RAM since most of the time there are only 2 Controllers defined (as opposed to the 16 allowed).
* Added support for zero-length TLV
* Added SEPARATOR as a formal kTLVType (and updated listControllers() to use)
* Added `uint8_t *buf(tagType tag, uint8_t *src, int len);` to load buffer needing external memcpy (and updated listControllers() to use)
Optimized use of heap memory instead of stack memory for temporary variables, and makes code easier to read, provided all heap usage is free() at end of each function.
Also removed char *hostName as a member variable of homeSpan. It's never actually used since MDNS constructs its own copy of hostName as needed.
Acquiring time from a timeserver is now spawned as a separate task running in the background. This avoids blocking HomeSpan while setting the time.
Default wait time has been extended from 10 seconds to 2 minutes, since there are no problems with blocking.
Gregg