* Rewrite HTTPClient chunked transfer-encoding
Fixes#3029
The chunked decoder seems to have had some race conditions resulting in
lost data.
The HTTPClient::getStreamPtr returned the raw WiFiClient which included
all of the chunked block size markers, requiring the user to manually
parse the chunked encoding.
Remove the existing chunked handling, add a HTTPStream as a WiFiClient
subclass. The HTTPClient will return this HTTPStream which will always
properly cut out chunk markers when present (and pass things raw for
non-encoded streams). Use this new HTTPStream class to replace the
existing handling in HTTPClient
Update the fingerprint in the StreamHTTPSClient example because their
cert was updated.
* Astyle
* Add block write passthru
* Add other WiFiClient passthru calls
BT SPP doesn't have backoff and we can receive long
messages. If there's not enough memory to store them
then they will be truncated.
Increase the default space to 1K to help avoid this issue.
Related to #2991, the ADCInput bufferWords needs to be a size that always
has the 1st ADC sample at offset 0:0. For 1 and 2 inputs that's guaranteed.
For 3 inputs we need a multiple of 3 words to ensure things always align
in the case of overflow. For 4 inputs, a multiple of 2 words is needed.
Moves to just released GCC 14.3
Moves to full-fat printf/scanf because Newlib no longer supports all the
calls libstdc++ requires with the older, smaller, nano-formatted-io option.
Moves to latest SDK develop branch with important float acceleration fixes
for the RP2350, among other updates.
Moves to latest pioasm, picotool develop branches
* SDK was refactored, update the includes
* Rebuild PicoSDK and BearSSL libs
* Update certs in BSSL validation example
* Don't spell check the certs!
* Updated tools with full fat++ newlib printf/scanf
When a link comes up, we were only sending a DHCP request if the existing
netif's ipaddress was 0. When a DHCP lease is acquired at first that IP
is changed to the given address, and if we do another ::begin we wouldn't
dhcp_start to send a new request and use the old one (until its lease
expired).
In the case of network changing (i.e. WiFiMulti on different nets or
moving an Ethernet cable to another router) that old address would not
be valid.
Track if an IP address has been manually set and use that to determine
if DHCP needs to be re-requested instead of looking at the old netif's
ipaddress.
Fixes#2974
Newlib has a stack protector fail handler which write(2)s a
message to STDERR which doesn't work here. Override with a
call to panic().
Tell GCC to protect all functions for stack protection instead of
ones that it heuristically decides need protection. Slower but
safer, and only when stack protection is enabled.
Fixes#2935
The PICO_2350A patch flipped the logic of the ADCInput bitmask calcs
causing the 2040 and 2350A to use bitmasks for the 2350B (i.e. wrong
GPIOs) and vice versa.
The SerialPIO(SoftwareSerial) receive path was convoluted and required
a lot of work on the host to get the actual data out. It also wasn't
always sampling on the proper edge leading to errors whenever clocks
or hold times shifted slightly.
Rewrite the SerialPIO RX path to explicitily wait for start bit,
pause 1/2 bit time, then idle for a full bit time for all bits.
Takes more PIO instruction memory but works flawlessly even with
mismatched clocks.
Tested with a loopback from HW UART to SW UART with a 5% clock
mismatch @ 19200 baud without reception errors, whereas the
original code would fail with less than a 0.5% variation.
Fixes#2928
````
SoftwareSerial s(15, -1);
void setup() {
Serial1.setTX(0);
Serial1.begin(19200 + 1920/2, SERIAL_8N1);
s.begin(19200, SERIAL_8N1);
}
void loop() {
Serial.println("---");
Serial1.write("Had we but world enough and time", 32);
uint32_t now = millis();
while (millis() - now < 500) {
while (s.available()) {
auto c = s.read();
Serial.printf("%02x '%c'\n", c, c);
}
}
}
````
Fixes#2914
There are 2 systick units, one per core. Set up and start core1's
systick unit and track each core's epoch separately.
Document a method of preserving 100% user-only code on core1
and add a core1_disable_systick boolean flag that works like the
separate stack one.
* Add PIO.h header verification to CI
Ensure all PIO .pio.h headers match the .pio sources in the tree
* Install all tools for Style check
* Clean up mismatched PIO headers
No functional changes, but the PDM pdm.pio file did not init a data pin
while the pdm.pio.h (the one actually used in the core) did. Correct to
match.
* No need for submodules in the style check
* Convert to SDK RP2350A/B determination
Fixes#2878
The SDK uses `defined(PICO_RP2350) && !PICO_RP2350A` to indicate an RP2350B
chip, not the define PICO_RP2350B.
Match the SDK's usage by converting from `defined(PICO_RP2350B)` to
`defined(PICO_RP2350) && !PICO_RP2350A` and update the chip variants
accordingly.
* Need to explicitly override PICO_RP2350A for all
The *SDK*'s board definition file hardcodes a PICO_RP2350A value for
all boards, but we use the same board file for both A and B variants.
Override the SDK board definition in the variant definition file.
* Generic RP2350 needs 2-stage PICO_RP2350A setting
Also ensure SDK board definition included before pins_arduino.h for
clearing up redefinition errors.
* Factor out undef PICO_RP2350A
* Update Arduino.h
Boards with selectable flash sizes only had a 1MB FS as the smallest
option on 2MB boards. For the normal Pico @ 2MB, though, we supported
filesystems down to 64KB. Add those same options to the 2MB SKUs
of configurable boards.
Windows Python doesn't seem to kill the worker Thread properly when the IDE
is exited, leading to a) multiple Python3 instances on a PC after many uses
and b) errors updating the core when it tries to re-install Python3 while
still having the older version's EXE loaded and in use.
When an exception happens on the input() call under Windows, it seems like it
can still leave a thread running. Add one more flag, caught in a global
exception handler.
Works around https://github.com/arduino/arduino-cli/issues/2867
Supercedes #2815
Move to pico-sdk official develop branch which includes a necessary
IRQ header fix.
200MHz is now default for the Pico, but 133 is still available from
the menus.
* Adjust the PSRAM clock when over/underclock F_SYS
Fixes#2818
* Need to increase PSRAM divider before sysclk increase
Per datasheet, when increasing sysclk speed we need to set the qmi
clocks first and do a dummy transfer to ensure no invalid speed
operations happen on the bus. Handle the logic for this while setting
up the overclock.
* Move to Arduino API 10501
Track upstream Arduino API headers
* IPAddress V4/V6 compatiblity restored
* Fix WiFiUDP includes
* String differences in example
* HardwareSerial using exported
* Add PIO-based SoftwareSPI enabling SPI on any pins
The Raspberry Pi team has a working PIO-based SPI interface. Wrap it
to work like a hardware SPI interface, allowing SPI on any pin
combination.
Tested reading and writing an SD card using unmodified SD library.
* Add W5500 example
Good for testing, shows non-contiguous pin outs.
* Don't set SDFAT_FILE_TYPE, default is OK
Fixes#2772
No need to set SDFAT_FILE_TYPE=3 as that is the defaulr value with upstream
SdFat. Remove it from platform.txt and platform.io build.
* Codespell got all techy
* Replace ESP8266SdFat w/SdFat 2.3.0, add SDIO and ExFAT support
Remove ESP8266SdFat fork and replaces with upstream SdFat to simplify
maintenance.
This 2.3.0 version adds SDIO support and enables exFAT support.
Also upgraded FAT filename support to 256 chars, identical to LittleFS.
* Add SDIO support to SD and SDFS, documentation, and examples
* Update SD examples to all support SPI0, SPI1, or SDIO
* Remove Pico-SDK C++ exception-override new/delete
We support exceptions on and off, but the cxx_options file in the
SDK implemented a single override to new/delete.
Remove it so we will use GCC's build-in operator new/delete
which will be correct for either option (2 different libstdc++
versions are shipped as part of the toolchain).
* Remove duplicated SDK files
The SDK will link in the same compilation unit in the LWIP builds.
Remove them to shrink the repo size by ~28MB.
When multiple inputs were active, the frequency was being scaled two
times resulting in incorrect sampling speed. Correct to only scale
the calculation and not the stored value (which is used in `begin`).
Fixes#2754
Pico-SDK changed the output extension of their compilation from "x.c.obj"
to "x.c.o" meaning the removed Newlib and STDIO SDK wrappers were still
being linked in certain situations. Update make-libpico Cmakefile to
remove the new names.
Fixes debug `printf` output.
The memcpy-version of the ABM::write updated the destination and count
but neglected to move the source forward. If a block write crossed a
frame boundary then the 2nd frame would repeat the data from the first
half of the buffer.
Fix by incrementing the source pointer by the same amount as was written.
Calculating the DMA clock divider for PWMAudio can take a very long time
because there are 65K floating point divisions required. But most audio
will be one of a dozen sample rates, so it is possible to precalculate
the rates on the host and use a small lookup table to avoid any math at
all. Removes occasional scratching in PWMAudio when the BackgroundAudio
library changes sample rates.
Optimize AudioRequestBuffer writing when large blocks are available (i.e.
I2S writes of full MP3 or AAC frames in BackgroundAudio). Update I2S to use
the new call. Reduces 1152 calls to arb::write() to a single call/return
and optimized memcpy in that case.
The audio output objects all have the same general necessary configuration
calls. Abstract them out to a generic AudioOutputBase interface class that
they will inherit from. Simplifies letting applications use different
output channels.
Should be backwards compatible with existing code.
Semihosting is handy for debugging, so allow the core to use `SerialSemi` as the
::printf port. Add menu item to the IDE to allow selection.
Add RISC-V implementation of semihost call
* Add MDNS.addServiceTxt() to SimpleMDNS
Fixes#2678
A simple mapping allows for basic service text addition even when using
SimpleMDNS and ArduinoOTA.
* Protection against duplicate services addition
Adds a menu item to enable onboard profiling. This requires significant
RAM and really only makes sense on devices with PSRAM to store the state.
When the menu item is selected, allocates RAM and tracks function calls and
periodically samples the PC to generate a histogram of application usage.
The onboard gmon.out file can be written over Semihosting or
some other way to transfer to a PC for analysis.
Adds a profiling example with command lines.
Enable ARM-only semihosting mode. This mode allows applications on the
Pico to write to the OpenOCD console and read and write files on the
host system (i.e. debugging dump information, etc.)
It is not very fast because of the way it uses breakpoints on the Pico
to communicate, but it is useful in cases when you want to get a single
file off of the Pico while debugging.
Note that this **requires** a connected OpenOCD and GDB or else the
semihosting will cause a system panic.
