connecting to server.
Inside af426d0315, the `mqtt_socket_timer`
function was modified so that instead of checking the presense of
allocated `mud->pesp_conn` structure, `mud->connected` field was used
on determining if the timer need to be disarmed.
However, this is not entirely correct. If the TCP socket is actively
connecting and haven't timed out yet, then `mud->connected` is also
`false` and the timer will think the connection is broken and
disarms itself. This has two consequences:
* The connection timeout counter is no longer decremented and checked
* After connection succeeds, keepalive heartbeat is no longer being
sent (#3166). This is particularly noticeable in MQTT over TLS
connections, because those usually takes longer than 1 second
to finish and the timer would had chance to execute before connection
is established
This commit checks the presense of `pesp_conn->proto.tcp` pointer
instead, which was allocated in the same place as the (old) `pesp_conn`
struct, and according to my test indeed fixes the above issue.
- Lots of minor but nasty bugfixes to get all tests to run clean
- core lua and test suite fixes to allow luac -F to run cleanly against test suite
- next tranch to get LFS working
- luac.cross -a options plus fixes from feedback
- UART fixes and lua.c merge
- commit of wip prior to rebaselining against current dev
- more tweaks
* espconn: remove unused espconn code, take 1
This is the easiest part of https://github.com/nodemcu/nodemcu-firmware/issues/3004 .
It removes a bunch of functions that were never called in our tree.
* espconn: De-orbit espconn_gethostbyname
Further work on https://github.com/nodemcu/nodemcu-firmware/issues/3004
While here, remove `mqtt`'s charming DNS-retry logic (which is neither
shared with nor duplicated in other modules) and update its :connect()
return value behavior and documentation.
* espconn: remove scary global pktinfo
A write-only global! How about that.
* net: remove deprecated methods
All the TLS stuff moved over there a long time ago, and
net_createUDPSocket should just do what it says on the tin.
* espconn_secure: remove ESPCONN_SERVER support
We can barely function as a TLS client; being a TLS server seems like a
real stretch. This code was never called from Lua anyway.
* espconn_secure: more code removal
* espconn_secure: simplify ssl options structure
There is nothing "ssl_packet" about this structure. Get rid of the
terrifying "pbuffer" pointer.
Squash two structure types together and eliminate an unused field.
* espconn_secure: refactor mbedtls_msg_info_load
Split out espconn_mbedtls_parse, which we can use as part of our effort
towards addressing https://github.com/nodemcu/nodemcu-firmware/issues/3032
* espconn_secure: introduce TLS cert/key callbacks
The new feature part of https://github.com/nodemcu/nodemcu-firmware/issues/3032
Subsequent work will remove the old mechanism.
* tls: add deprecation warnings
* luacheck: net.ifinfo is a thing now
* tls: remove use of espconn->reverse
* mqtt: stop using espconn->reverse
Instead, just place the espconn structure itself at the top of the user
data. This enlarges the structure somewhat but removes one more layer
of dynamic heap usage and NULL checks.
While here, simplify the code a bit.
* mqtt: remove redundant pointer to connect_info
Everywhere we have the mqtt_state_t we also have the lmqtt_userdata.
* mqtt: doc fixes
* mqtt: note bug
* tls: allow :on(...,nil) to unregister a callback
* mqtt: expose "connfail" callback via :on()
This makes it just like all the other callbacks in the module and is a
revision of behavior called out in
https://github.com/nodemcu/nodemcu-firmware/pull/2967
* mqtt: clarify when puback callback fires
* mqtt: Don't reference stack buffers from the heap
The confusingly-named "mqtt_connection_t" object is just a triple of
- a serialized mqtt message pointer and length
- a buffer pointer (to which the above can be written)
- a message identifier
The last of these must be passed around the mqtt state machine, but the
first two are very local and the buffer is always sourced from the C
stack. Unfortunately, because the entire structure is persisted in the
heap, some callers assume that they can always use the structure without
reinitialization (see mqtt_socket_close), which will trash the C stack.
Sever the pairing between message id and local state, punt the local
state entirely out of the heap, and rename things to be less confusing.
The PR removed the bulk of non-newlib headers from the NodeMCU source base.
app/libc has now been cut down to the bare minimum overrides to shadow the
corresponding functions in the SDK's libc. The old c_xyz.h headerfiles have been
nuked in favour of the standard <xyz.h> headers, with a few exceptions over in
sdk-overrides. Again, shipping a libc.a without headers is a terrible thing to do. We're
still living on a prayer that libc was configured the same was as a default-configured
xtensa gcc toolchain assumes it is. That part I cannot do anything about, unfortunately,
but it's no worse than it has been before.
This enables our source files to compile successfully using the standard header files,
and use the typical malloc()/calloc()/realloc()/free(), the strwhatever()s and
memwhatever()s. These end up, through macro and linker magic, mapped to the
appropriate SDK or ROM functions.
* mqtt:connect() secure parameter should be boolean
Continue to honor the old 0/1 values, but make them undocumented and add
a deprecation warning to the code and docs. Eventually, this should go
away.
* mqtt: rip out deprecated autoreconnect
* mqtt: expose all the callbacks via :on
* MQTT: handle large/chunked/fragmented messages properly
If a message spans multiple TCP packets it must be buffered before
delivered to LUA. Prior code did not do this at all, so this "patch"
really adds proper handling of fragmented MQTT packets.
This could also occur if multiple small messages was sent in a
single TCP packet, and the last message did not completely fit in that
packet.
Introduces a new option to the mqtt.Client constructor:
max_publish_length which defaults to 1024
Introduces a new 'overflow' callback.
Fixes issue #2308 and proper fix for PR #2544.
* mqtt.md: clarified heap allocation
* mqtt: ensure ack is sent for overflowed publish
If QoS is used we should still acknowledge that we received it, or server might retransmit it later.
Any TCP packet with more than 1024 bytes of payload was silently
dropped. With MTU of 1500 the TCP payload can be up to 1460 bytes
(1500 - 20(IP hdr) - 20(TCP hdr))
* pmsleep refactor
* Shortened swtmr disabled message
* Added swtimer debug module option to user_modules.h.
* Added comments to user_config.h.
* Fixed error in documentation for node.sleep()
* remove blank sntp.c that got added in during rebase onto dev(6218b92)
* Added #ifdefs around SWTIMER_REG_CB to prevent inclusion of disabled
code
Disable the error callback of mqtt.client:connect() after a successful connection.
This will prevent this function to be called after a future disconnection.
Instead the "offline" function is called.
* Fix the error callback from not being called sometimes
* Moved the setting of the reconnect status to after the connack is recevied
* Increase the irom0_seg size
* Updated the documentation
* Make it clearer that autoreconnect is deprecated
- Process the CONNACK message received from the broker after Connect
- Provide feedback to Lua via failure callback on client:connect()
- Also provide failure information for other situations not covered by CONNACK
Module creation & registration now made a lot simpler. In essence,
each module file is now self-contained and only needs a
NODEMCU_MODULE(MYNAME, "myname", myname_map, luaopen_myname);
line to both be automatically recognised by the Lua initialization
as well as honor the LUA_USE_MODULES_MYNAME #define.
As per #810 & #796, only LUA_OPTIMIZE_MEMORY=2 & MIN_OPT_LEVEL=2 are
supported when building. This commit effects that limitation.
With this change modules/auxmods.h no longer needs to be updated for
every new module, nor do module writers need to cater for a hypothetical
LUA_OPTIMIZE_MEMORY < 2 scenario.