[Ovmsdev] ESP-IDF v4 / v5 - baby steps

[Ludovic LANGE]

Hi list,

I am interested in upgrading our codebase to a latest ESP-IDF version 
(preferentially v5 which has just happened end of 2022 - Happy New Year 
! :-)) - mostly to be able to use some components for which a backport 
seems difficult (for example : 
https://github.com/openvehicles/Open-Vehicle-Monitoring-System-3/issues/752 
), and as a matter of balance between stability vs latest bugfixes, 
security fixes, new features etc...

In the past it seems that some endeavours already set the basis to make 
this happen 
(https://github.com/openvehicles/Open-Vehicle-Monitoring-System-3/issues/263, 
https://github.com/openvehicles/Open-Vehicle-Monitoring-System-3/issues/360 
).

I'm now trying to revive this effort, and after some work managed to have:

  * OVMS codebase building under v4.x using Makefile
  * OVMS codebase building under v4.x using cmake / idf.py (not 100%
    finished, multiple vehicle_* components are missing)

(/*WARNING*: some important features of OVMS have been deactivated for 
the moment, see end of post)/

The produced binaries seem to "work on my machine board", as they end 
with a "OVMS#" prompt and a few (very) basic tests look OK (to be 
honest, unfortunately no real tests have been done except to check that 
the binary does not crash during boot and a few seconds of run, checking 
that WiFi / web Interface is OK, sdcard is OK - but very basic)


You'll find the changes here : 
https://github.com/llange/Open-Vehicle-Monitoring-System-3/tree/experimental-esp-idf-v4-rework-network
(In the README.md, there is a short notice at the top with a few useful 
informations)


I'm now reaching out here mostly for :

  * gathering feedback both around the effort (is it worth pursuing this
    goal ?) and the implementation (do my CMakeLists.txt look weird ?
    yes they do.)
  * having volunteers using those builds on real-world workload (which I
    cannot do at the moment) and finding bugs
  * guidance and fixes regarding the CMakeLists.txt files


My next steps could be (of course, subject to feedback) :

 1. Have 100% of the cmake implementation finished (all the missing
    vehicle_* components) -> easy (only missing time)
 2. Work on the warnings (currently disabled in the builds) and try to
    fix as much as possible
 3. Iron out all the bugs, improve the dev documentation -> need help on
    finding them with real-world use
 4. 1st goal is to propose this result as a merge request (either as a
    whole, or split depending on feedback - ideally I would like to have
    a first (big) MR accepted quickly so that people can play with it,
    and deliver incremental updates as expected)
 5. then work on the v5 and do the same.


(Please note that my cmake skills are inexistent - so be careful when 
opening those files to not hurt yourself.)

# The guidelines I tried to follow:

  * Stay compatible with our 3.3.x branch which should still compile
    without any issue nor loss of functionality
  * Keep Makefiles running for 3.3.x (and 4.x if possible)
  * When possible, isolate differences (between 3.x, 4.x, 5.x) by
    testing the versions of ESP-IDF.
  * Focus on 5.x (if there are impossibilities, let's make it easier to
    build v5.x than v4.x)
  * https://docs.espressif.com/projects/esp-idf/en/release-v4.4/esp32/api-reference/network/tcpip_adapter_migration.html
  * https://docs.espressif.com/projects/esp-idf/en/release-v4.4/esp32/api-guides/build-system.html#migrating-from-esp-idf-gnu-make-system
  * https://docs.espressif.com/projects/esp-idf/en/latest/esp32/migration-guides/release-5.x/5.0/index.html
  * https://docs.espressif.com/projects/esp-idf/en/latest/esp32/migration-guides/release-5.x/5.1/index.html


# Missing features:

In the course of this work I had to make some decisions to speed up the 
work  - that will need to be acted upon:

  * I disabled the crash handler (`xt_set_error_handler_callback` and
    `esp_task_wdt_get_trigger_tasknames`) for the moment, we need to
    decide whether we "fork" ESP-IDF again to port it ; or if the new
    APIs are enough to (partially ?) reimplement it
  * There is a crash in `OvmsConsole::Poll` which I did not analyse
    (yet) and which I worked around by declaring a variable static.
  * I decided to transform our local copies of `wolfssh` and `wolfssl`
    in submodules (and move them one level below in terms of
    directories) - mainly to be able to have a CMakeLists.txt different
    from the upstream one.
    In the process, I "lost" one of our patches :
    https://github.com/openvehicles/Open-Vehicle-Monitoring-System-3/commit/51444539047daef7bd2accb23ef40d1bc14fdb20
    and we need to decide how to handle this.
  * A lot of dependencies are now explicitly (hard-)coded in the
    CMakeLists.txt - which may, or may not be a good thing. Let's
    discuss it.
  * I had to change a set of defines into a header generation (in
    ovms_webserver) because I was unable to implement those in a
    satisfying manner in cmake. If you manage to do it, it's a win !
  * Build warnings were disabled to keep the output clean and focus on
    the build issues - but now they need to be re-activated.
  * etc...
  * Even if there is experimental support for cmake / idf.py in ESP-IDF
    3.3.x, I did not make any effort to be compatible with it:
      o Our own fork does not seem to include this build system
      o It was experimental, the component API changed and I did not
        want to have edge case for something not used at the moment.


Let me know your comments and questions.

Best regards,
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openvehicles.com/pipermail/ovmsdev/attachments/20230123/a9e2c433/attachment.htm>