Thanks Michael,

I have taken it in, and I'll use std::string in poll once code. I guess it might make sense to convert everything else to std::string.  It wouldn't be a big task.

 I had just taken out that commit and reworked anyway, and now that you say it's not the best way of doing things, I'm even happier I made that choice, as I had intended to possibly make more use of it and won't now.  At the moment I'm using it to get the VIN (snipped) where I'll try a couple of times only, rather than continually poll for it.

Hmm... yea - I guess I could do some dynamically created poll table.  Will definitely consider that.

Further comments inline below:

 

On Sun, 23 Oct 2022 at 16:15, Michael Balzer <dexter@expeedo.de> wrote:

Michael,

what I really dislike about your change is that it adds complexity to the underlying function just to fulfill an -- in my eyes more academic -- type style choice: having to handle multiple buffer pointers just to provide different typing approaches is bad. You could introduce a new construct to provide a single polymorphic pointer, but that again adds complexity.

Agreed, std::string is a bit type-sloppy in depending on char being unsigned 8 bit, but that's all to consider, and can easily be enforced if ever necessary.

On the plus side, std::string has far better library support, better interoperability (see our use cases), and small string optimization, which is especially useful for our platform, as most poll requests and responses easily fit into the initial capacity, normally eliminating the need for heap allocations.

So, besides the type independency, I think std::vector has no advantage here.


There's another thing I'd like to address: it seems you're planning to use single polls extensively?

If so, please don't, unless you're also going to do a full rewrite of the poller towards a job/worker service task architecture (that's a very old todo).

The single poll utility is really meant for sparse/manual polls outside the normal polling scheme. See my comment on the method:

 * PollSingleRequest: perform prioritized synchronous single OBD2/UDS request
 *  Pass a full OBD2/UDS request (mode/type, PID, additional payload).
 *  The request is sent immediately, aborting a running poll list request. The previous
 *  poller state is automatically restored after the request has been performed, but
 *  without any guarantee for repetition or omission of an aborted poll.

This was really meant more for user/script use than for standard vehicle polls, though it's OK to use it, where applicable, with caution.

For all standard polling, install a PID list. You can vary a list by the poll state, and you can switch lists if necessary. PID lists can easily be generated dynamically, see the VWUP code for an example.


The place for general CAN utility methods is the components/can/src/canutils module, general utility methods go into the main/ovms_utils module.

Ahh.. cool. 

Regarding CAN buffer data extraction, there are currently probably as many approaches as there are vehicle modules. Certainly would be nice to have some standard way here, but won't be simple to rework the existing vehicle adaptors.

The template approach is a nice one for a general solution, but keep in mind not every vehicle will use big endian encoding, in some cases not even consistently throughout the installed devices.

I had already implemented little-endian and big-endian versions of the data extract. I will now explicitly mark all of them.

 

To be really general, your byte addressing needs to support more than 256 bytes in the buffer.

Consider that fixed.

 

Your templates also lack bit masking and shifting. Some devices use packed structs with odd sized bit fields. The DBC engine has support for all this, but not generalized. I remember Mark getting a headache from this…

I have a get_bit<>() function and so I would propose it be implemented in a similar way... ie exract the byte first, then get the bits.

Something like: extract_int<1,2>(byte)  --> Extract bits len 2 starting from bit 1. We can easily sign-extend that too.

 

Regards,
Michael

Am 23.10.22 um 02:23 schrieb Michael Geddes:

Hi Michael,

Just wanted to say that if you override my objections on the std::vector<unit8_t> in the single poll code then I'm fine with that, I recognised that it's absolutely your call and that bit has a small impact on my code so far.   I can even add versions of the extraction functions that use std::string if you want (they would just be inline template wrappers around common code anyway so zero compiled cost).  

Also: Is there a better place to put those templated functions?  A version of my sign_extend function is already being proposed being used here: https://github.com/openvehicles/Open-Vehicle-Monitoring-System-3/pull/736 - it would be good if both versions of this function were available in a single place.  (There's   sign_extend<UINT, INT>(int_value, sign_bit)  and  sign_extend<UINT, INT, SIGNBIT>(int_value)  - so they should work fine as overloaded template functions).

I would consider also whether I use std::string in my own code where I'm extracting my data out of complete ISOTP poll results. Up to you. That has slightly more impact, but the code change is pretty easy.

//.ichael

On Sat, 22 Oct 2022 at 07:56, Michael Geddes <frog@bunyip.wheelycreek.net> wrote:

Thanks Michael,

I had already created my commits under the assumption that you would prefer the general changes separate from the vehicle-specific commits.

I have created 3 pull requests so far.

  1) the bug - that's a no-brainer.  

  2) 'finalisation' of temperatures and voltages. (Draft, see below)

  3) vector poll. This is more a point of discussion (see below)

A couple of the next bugs are kind of dependent on the second, so I'll wait until I get that sorted out.

More comments below.

On Fri, 21 Oct 2022 at 23:31, Michael Balzer <dexter@expeedo.de> wrote:

Michael,

Am 19.10.22 um 07:26 schrieb Michael Geddes:
> Hi all,
>
> I have 6 commits that I would _like_ to get pushed up that are changes
> to components/vehicles required for my Ioniq 5 changes to work.
>
> I can push them all up in a single merge request for review, or
> separate them out into bits.  Some of them I am expecting there to be
> opinions on. I'm prepared to simplify some, and one in particular has
> grounds for nixing. I think it's a good change, but I could manage
> without it.

Please keep PRs as small as possible, focusing on one specific
modification / addition at a time. That way, every merge commit can
easily be reverted if necessary, and changes can be followed and
referred to easily. Especially keep framework changes separate from
vehicle specific ones.

It's also OK to open the PRs tagged as drafts / work in progress, that
helps in commenting on certain parts and tracing refinements. Create one
branch for each PR, pushing new commits into the branch will
automatically update any open PR created from it.

Ooh. That's useful. It's been a while since I've used github for collab. Switched the battery sensor P/R over to a draft.

 

 

>
> One is a simple bugfix - a literal one-character change:
> void OvmsVehicle::BmsRestartCellTemperatures()
>    {
>    m_bms_bitset_t.clear();
> -  m_bms_bitset_v.resize(m_bms_readings_t);
> +  m_bms_bitset_t.resize(m_bms_readings_t);
>    m_bms_bitset_ct = 0;
>    }

Obviously a copy & paste bug, nice find. I wonder how that could slip
through without causing crashes, as there are more voltages than
temperatures in most packs.

Please don't hesitate with opening PRs even for single fixes like this,
the sooner we include them, the better.

Yep. Done and understood. 

>
> On is a simple code readability change. Using an enum instead of magic
> status values for m_bms_talerts.  Simple enough, but I'm prepared to
> simplify that to a standard enum or get rid of it.

The magic values are simply alert levels, but go ahead, readability is a
good goal.

I'll include it once the Battery Cell code is finalised. 

>
> There's some battery related ones where I want to be able to not quite
> know the number of cells from the start and let the OBD responses specify.

Dynamic pack layout reconfiguration is no issue, I've done that on the
Twizy (look for comments "update pack layout" or calls to
BmsSetCellArrangement*).

BmsSetCellArrangement does clear all the values... which is probably the main thing I was trying to avoid.  Ie- I get to the end and go 'ooh.. only this number of cells were filled in'.  More comments in the P/R about what I want to achieve and fall-back positions.

>
> There's also a battery one where the info for the cells have different
> timings (because one has other info), so I want to be able to clear a
> range of cells rather than just reset and start filling the info again.

That may be a bad idea, maybe better to implement some synchronization /
drop readings out of sync.

Reaching a consistent state, the BMS will calculate deviations etc..
Crucial on that part is an analysis of the consistency of the cell
readings to determine the reliability of deviations exceeding thresholds.

See my comments in BmsSetCellVoltage() and list thread:
http://lists.openvehicles.com/pipermail/ovmsdev/2021-February/015079.html

I've skimmed the comments but will read properly and digest.   Yeah it might be best if I just clear the values at the start of the more often running one and then once the other ones come in it will finalise once it gets to the end of those.. so they aren't too far apart.

>
> I've modified slightly the dump for the battery cell
> voltage/temperature, allowing for either or both columns to be visible
> if available. It also copes with the above 'not set' scenario.

Sounds good.

I'll push this once the FinaliseCellVoltage pull-request is finalised and accepted.

>
> Finally, the controversial one:  I am using std::vector<uint8_t> as a
> buffer in my ioniq5 code.. and I wanted to be able to use it for the
> polling instead of std::string. I've not replaced the std::string but
> added the ability to have either.
>
> If my aversion to using std::string for binary data is misplaced, then
> I'm ok with the currently small modification to my code that would use
> the std::string version.

The general use case of ISO-TP data is actually dynamic binary strings.
They are meant to and need to be easily aggregated, cut, merged,
partially extracted etc., so they really are strings, not vectors (type
wise).

Hmm.. I'm not sure I entirely agree about needing to aggregate/cut/merge; we know the size of the buffer after the first packet comes in so we can pre-set the buffer internal size and append data as it comes in without casting from uint8_t* to char *.  Even if the size is wrong, we can still very easily append to a vector.

ON the extraction, I have efficient templated functions that extract the data from the std::vector<uint8_t> buffer.. including to a std::string if required, or to big-endian/little-endian signed/unsigned integers of 1..4 bytes with sign-extension now.  These also check bounds so are quite safe - possibly safer than the macros.  The bounds check has already proved itself useful.

I have pushed up a draft  request 'single poll to buffer' as it is the easiest way to describe what I want to do.  The first commit is the meat of the change ... but the second is adding all the utility functions.  They probably don't belong in vehicle.h/cpp, but that's easy enough to change.  I had them in my ioniq 5 specific code - but have shoved them in there just for a point of discussion.

The great thing about the template functions here is that they should expand inline at compile time to quite efficient code.

Regards,
Michael

//.ichael 

_______________________________________________
OvmsDev mailing list
OvmsDev@lists.openvehicles.com
http://lists.openvehicles.com/mailman/listinfo/ovmsdev

-- 
Michael Balzer * Helkenberger Weg 9 * D-58256 Ennepetal
Fon 02333 / 833 5735 * Handy 0176 / 206 989 26

_______________________________________________
OvmsDev mailing list
OvmsDev@lists.openvehicles.com
http://lists.openvehicles.com/mailman/listinfo/ovmsdev