Keepin constants, configuration values, etc. in Python - dedicated module or what?

[cl]

I am developing some code which runs on a (remote from me most of the
time) Beaglebone Black single board computer. It reads various items
of data (voltages, currents, temperatures, etc.) using both a 1-wire
bus system and the Beaglebone's ADC inputs. The values are stored
at hourly intervals into a database which is regularly rsync'ed
across to my home system where I use the values for monitoring etc.

Associated with each value are various parameters, i.e. for each ADC
input of the Beaglebone there is the ADC input number, a short name
for the value, a conversion factor and a description. Similarly for
the 1-wire inputs there is the 1-wire filesystem filename, a short
name for the data and a longer description.

I am puzzling where and how to keep these configuration values. My
current design has them in dedicated tables in the database but this
is rather clumsy in many ways as there's an overhead reading them
every time the program needs them and changing them isn't particularly
convenient at the Beaglebone doesn't have a GUI so it has to be done
using SQL from the command line.

Does it make sense to add them to the modules which handle the reading
of the inputs? I already have modules defining classes called Adc and
OneWire, is it a reasonable approach to add the configuration to these
as, probably, dictionaries? Modifying it would be pretty simple -
just edit the python source (one of the easiest things for me to do on
the Beaglebone as my sole access is via ssh/command line).

Thus I'd have something like (apologies for any syntax errors):-

cfg = { "LeisureVolts": ["AIN0", 0.061256, "Leisure Battery Voltage"],
"StarterVolts": ["AIN1", 0.060943, "Starter Battery Voltage"],
"LeisureAmps1": ["AIN2", 0.423122, "Leisure Battery Current"}

(It might be better to makes those lists dictionaries, but it shows
the idea)

Are there any better ways of doing this? E.g. some sort of standard
configuration file format that Python knows about? I would actually
quite like to keep the configuration data separate from the code as it
would simplify using the data at the 'home' end of things as I'd just
need to copy the configuration file across. This was why the database
approach appealed at first as all I need to do is copy the database
and everything is in there.

I'm not really expecting quick/glib answers, just some ideas and
comments on the various ways of doing this and their advantages and
disadvantages.

[Dave Angel]

Post by cl
I am developing some code which runs on a (remote from me most of the
time) Beaglebone Black single board computer. It reads various items
of data (voltages, currents, temperatures, etc.) using both a 1-wire
bus system and the Beaglebone's ADC inputs. The values are stored
at hourly intervals into a database which is regularly rsync'ed
across to my home system where I use the values for monitoring etc.
Associated with each value are various parameters, i.e. for each ADC
input of the Beaglebone there is the ADC input number, a short name
for the value, a conversion factor and a description. Similarly for
the 1-wire inputs there is the 1-wire filesystem filename, a short
name for the data and a longer description.
I am puzzling where and how to keep these configuration values. My
current design has them in dedicated tables in the database but this
is rather clumsy in many ways as there's an overhead reading them
every time the program needs them and changing them isn't particularly
convenient at the Beaglebone doesn't have a GUI so it has to be done
using SQL from the command line.

It is very useful for the database to be self contained and self
describing. Among other things it means that you can easily take
a snapshot at any time, and that snapshot is not tied to any
specific version of the code. As for changing with sql, you can
always add a configuration utility when using sql becomes a
nuisance, or introduces errors.

The key thing is to manage change. You need to decide which
things are conceivably going to change and how you'll manage both
the change itself and the inevitable split between prechange data
and post. One kind of change might be fixing the spelling of
battery. No biggie, just let new accesses get the new one.
Another kind might be the addition of a new instance of an adc
converter. Not a problem, as long as you don't reuse an old
name. And presumably you never remove an old name from the
config.

More troublesome is adding a new kind of data. You have to decide
whether it's worth generalizing the code to anticipate the
change, or just admit that the code will grow at that point and
that old code won't deal with the new data.

Post by cl
Does it make sense to add them to the modules which handle the reading
of the inputs? I already have modules defining classes called Adc and
OneWire, is it a reasonable approach to add the configuration to these
as, probably, dictionaries? Modifying it would be pretty simple -
just edit the python source (one of the easiest things for me to do on
the Beaglebone as my sole access is via ssh/command line).

Nope, read it from the db. You still might be loading it to a cfg
variable, however.

Post by cl
Thus I'd have something like (apologies for any syntax errors):-
cfg = { "LeisureVolts": ["AIN0", 0.061256, "Leisure Battery Voltage"],
"StarterVolts": ["AIN1", 0.060943, "Starter Battery Voltage"],
"LeisureAmps1": ["AIN2", 0.423122, "Leisure Battery Current"}
(It might be better to makes those lists dictionaries, but it shows
the idea)

Actually it's probably better to use named tuples. All those lists
are apparently of identical length with identical meaning of a
given element offset. But if named tuple isn't flexible,enough,
you probably need a new class.

Post by cl
Are there any better ways of doing this? E.g. some sort of standard
configuration file format that Python knows about? I would actually
quite like to keep the configuration data separate from the code as it
would simplify using the data at the 'home' end of things as I'd just
need to copy the configuration file across. This was why the database
approach appealed at first as all I need to do is copy the database
and everything is in there.
I'm not really expecting quick/glib answers, just some ideas and
comments on the various ways of doing this and their advantages and
disadvantages.

Main next question is whether you can restart each system when you
add to the configuration.

[Rustom Mody]

Post by cl
I would actually
quite like to keep the configuration data separate from the code as it
would simplify using the data at the 'home' end of things as I'd just
need to copy the configuration file across. This was why the database
approach appealed at first as all I need to do is copy the database
and everything is in there.

Of course

Post by cl
Are there any better ways of doing this? E.g. some sort of standard
configuration file format that Python knows about?

Umm this is getting to be a FAQ...
Maybe it should go up somewhere?

Yes there are dozens:
- ini
- csv
- json
- yml
- xml
- pickle
- And any DBMS of your choice

I guess Ive forgotten as many as Ive listed!!

[cl]

Post by Rustom Mody

Post by cl
I would actually
quite like to keep the configuration data separate from the code as it
would simplify using the data at the 'home' end of things as I'd just
need to copy the configuration file across. This was why the database
approach appealed at first as all I need to do is copy the database
and everything is in there.

Of course

Post by cl
Are there any better ways of doing this? E.g. some sort of standard
configuration file format that Python knows about?

Umm this is getting to be a FAQ...
Maybe it should go up somewhere?
- ini
- csv
- json
- yml
- xml
- pickle
- And any DBMS of your choice
I guess Ive forgotten as many as Ive listed!!

Yes, I know, I've found most of those. I'm really asking for help in
choosing which to use. I think I can reject some quite quickly:-

ini - doesn't work so well with lists/dictionaries (though possible)
csv - rather difficult to edit
json - one of the most likely possibilities, but prefer yml
yml - front runner if I go for configuration files
xml - horrible, nasty to edit, etc. I don't like XML! :-)
pickle - not user editable as I understand it
dbms - current solution in sqlite3 as described

What I'm really asking for is how to choose between:-

DBMS - present solution, keeps config with data easily but more
code and less easy to change
yml - easy to edit config, keeps data separate from code but needs
YAML installed and separate files to manage
python - just keep config in the modules/classes, not easy to use
at 'both ends' (home and remote), otherwise quite simple

My requirements are:-
Easy to change, i.e.human readable format which can be edited
simply as I have to do this via a terminal/ssh.
Easy to use at local end as well as remote end, the remote end is
where it lives and is the 'driver' as it were but I need to know
the configuration at the local end as well.

[jkn]

might this be of interest (though old)?

https://wiki.python.org/moin/ConfigParserShootout

Cheers
Jon N

[Rustom Mody]

Post by cl

Post by Rustom Mody

Post by cl
I would actually
quite like to keep the configuration data separate from the code as it
would simplify using the data at the 'home' end of things as I'd just
need to copy the configuration file across. This was why the database
approach appealed at first as all I need to do is copy the database
and everything is in there.

Of course

Post by cl
Are there any better ways of doing this? E.g. some sort of standard
configuration file format that Python knows about?

Umm this is getting to be a FAQ...
Maybe it should go up somewhere?
- ini
- csv
- json
- yml
- xml
- pickle
- And any DBMS of your choice
I guess Ive forgotten as many as Ive listed!!

Yes, I know, I've found most of those. I'm really asking for help in
choosing which to use. I think I can reject some quite quickly:-
xml - horrible, nasty to edit, etc. I don't like XML! :-)

Heh! Youve proved yourself a pythonista!

Post by cl
ini - doesn't work so well with lists/dictionaries (though possible)
csv - rather difficult to edit

Have you tried with comma=tab?

Post by cl
yml - front runner if I go for configuration files

Yeah my favorite as well

Post by cl
json - one of the most likely possibilities, but prefer yml

Seems to be most popular nowadays -- maybe related to being almost yaml
and in the standard lib

Post by cl
pickle - not user editable as I understand it

Well not in any reasonably pleasant way!

Post by cl
What I'm really asking for is how to choose between:-

<snipped>

Post by cl
python - just keep config in the modules/classes, not easy to use
at 'both ends' (home and remote), otherwise quite simple

Can work at a trivial level.

As soon as things get a bit larger data and code mixed up is a recipe for mess up.

[Chris Angelico]

Post by Rustom Mody

Post by cl
python - just keep config in the modules/classes, not easy to use
at 'both ends' (home and remote), otherwise quite simple

Can work at a trivial level.
As soon as things get a bit larger data and code mixed up is a recipe for mess up.

True, but it's certainly possible to break out the config data into an
importable module that conceptually just provides constants.
Technically it's code, yes, but it'll normally be code that looks like
your standard "name = value" config file:

# docs for first option
# more docs
# examples
# etcetera
first_option =123

# docs for second option
second_option = 234


Is that Python code, or is it a sectionless INI file, or what? There's
no difference. And you get expressions for free - simple stuff like
"7*24*60*60" to represent the number of seconds in a week (for people
who aren't intimately familiar with 604800), or calculations relative
to previous data, or whatever. Sometimes it's helpful to have just a
little code in your data.

ChrisA

[Rustom Mody]

Post by Chris Angelico

Post by Rustom Mody

Post by cl
python - just keep config in the modules/classes, not easy to use
at 'both ends' (home and remote), otherwise quite simple

Can work at a trivial level.
As soon as things get a bit larger data and code mixed up is a recipe for mess up.

True, but it's certainly possible to break out the config data into an
importable module that conceptually just provides constants.
Technically it's code, yes, but it'll normally be code that looks like
# docs for first option
# more docs
# examples
# etcetera
first_option =123
# docs for second option
second_option = 234
Is that Python code, or is it a sectionless INI file, or what?

Yeah I was going to say that this is possible

Post by Chris Angelico
There's no difference.

But there is! Its code that looks like data.

Post by Chris Angelico
And you get expressions for free - simple stuff like
"7*24*60*60" to represent the number of seconds in a week (for people
who aren't intimately familiar with 604800), or calculations relative
to previous data, or whatever. Sometimes it's helpful to have just a
little code in your data.

Not free at all. Power of code means cost of code
See http://www.w3.org/2001/tag/doc/leastPower.html

I'd reiterate though what I first said: In this case its probably
ok if the code (=data) does not cross trivial limits

[Chris Angelico]

Post by Rustom Mody

Post by Chris Angelico
And you get expressions for free - simple stuff like
"7*24*60*60" to represent the number of seconds in a week (for people
who aren't intimately familiar with 604800), or calculations relative
to previous data, or whatever. Sometimes it's helpful to have just a
little code in your data.

Not free at all. Power of code means cost of code
See http://www.w3.org/2001/tag/doc/leastPower.html

It's free once you've already decided (for other reasons) to make your
config file use Python syntax, at which point you've already paid the
cost of code. But you're quite right, allowing code does have costs.
Most notably, you have to quote your strings; although the direct
benefits you get (line continuation, unambiguous handling of
leading/trailing spaces, etc) may outweigh that on their own.

Post by Rustom Mody
I'd reiterate though what I first said: In this case its probably
ok if the code (=data) does not cross trivial limits

I'd agree, where "trivial limits" is defined by each individual item.
Going with straight Python code is fine for huge projects with long
config files, as long as each config entry is itself simple. You even
get a form of #include: "from otherfile import *".

ChrisA

[Rustom Mody]

Post by Chris Angelico

Post by Rustom Mody

Post by Chris Angelico
And you get expressions for free - simple stuff like
"7*24*60*60" to represent the number of seconds in a week (for people
who aren't intimately familiar with 604800), or calculations relative
to previous data, or whatever. Sometimes it's helpful to have just a
little code in your data.

Not free at all. Power of code means cost of code
See http://www.w3.org/2001/tag/doc/leastPower.html

It's free once you've already decided (for other reasons) to make your
config file use Python syntax, at which point you've already paid the
cost of code. But you're quite right, allowing code does have costs.
Most notably, you have to quote your strings; although the direct
benefits you get (line continuation, unambiguous handling of
leading/trailing spaces, etc) may outweigh that on their own.

Post by Rustom Mody
I'd reiterate though what I first said: In this case its probably
ok if the code (=data) does not cross trivial limits

I'd agree, where "trivial limits" is defined by each individual item.
Going with straight Python code is fine for huge projects with long
config files, as long as each config entry is itself simple. You even
get a form of #include: "from otherfile import *".

Well in programming we sometimes use strict/formal methods and sometimes more
informal.

In this case using configparser or pyyaml or whatever means that the config
file's syntax is rigorously fixed by that library

On the other hand if you impose a convention: Constants file-module has
NOTHING but constants, thats a non-formal convention and thats ok.

If however you mix it up with other (real) code, you'll get a bloody mess.
This kind of stuff "7*24*60*60" is borderline and in my experience
its a slippery slope that ends up being more trouble than its worth.

Experience being emacs where because in lisp code and data are
the same, all kinds of design messes are perpetrated

[Chris Angelico]

Post by Rustom Mody
On the other hand if you impose a convention: Constants file-module has
NOTHING but constants, thats a non-formal convention and thats ok.
If however you mix it up with other (real) code, you'll get a bloody mess.
This kind of stuff "7*24*60*60" is borderline and in my experience
its a slippery slope that ends up being more trouble than its worth.

Agreed, and which side of the border it's on is a matter of opinion.
Suppose you were writing an INI-style parser - would you include this
feature? I probably wouldn't include arithmetic, as it's usually more
effort than it's worth, but it isn't a bad feature inherently. I'd
definitely not include backreferencing (FOO = "bar", followed by
FOO_LEN = len(FOO) or something), or the ability to chain assignments
(FOO = BAR = 0), so those would be advised against in the convention
docs. On the other hand, a quoted string requirement ("foo\nbar")
makes a huge amount of sense in certain contexts, and I can imagine
implementing quite a lot of a programming language's string literal
syntax.

ChrisA

[cl]

Post by Rustom Mody

Post by Chris Angelico
# docs for first option
# more docs
# examples
# etcetera
first_option =123
# docs for second option
second_option = 234
Is that Python code, or is it a sectionless INI file, or what?

Yeah I was going to say that this is possible

Post by Chris Angelico
There's no difference.

But there is! Its code that looks like data.

The main trouble with this approach is that I need some way to have
the python/config file available at the 'home' end of this as well as
at the 'remote' end. I guess I could write a copy of the file into
the database but then I have the editing issue again, changing it
becomes messy. If it's not in the database then how do I 'tie' it to
the data?

[Chris Angelico]

Post by cl
The main trouble with this approach is that I need some way to have
the python/config file available at the 'home' end of this as well as
at the 'remote' end. I guess I could write a copy of the file into
the database but then I have the editing issue again, changing it
becomes messy. If it's not in the database then how do I 'tie' it to
the data?

That's a design question. Maybe it's better for you to do your config
in the database. I usually find that these config files include
database credentials (server, port, user name, password), so they have
to be (a) outside the database, (b) outside source control, and (c)
separately configurable for test and production systems, even if they
run on the exact same hardware. So for those situations, it makes more
sense to have them in a script or INI file, rather than the database.
Your situation may well be completely different.

ChrisA

[Steven D'Aprano]

Post by Chris Angelico
I'd agree, where "trivial limits" is defined by each individual item.
Going with straight Python code is fine for huge projects with long
config files, as long as each config entry is itself simple. You even
get a form of #include: "from otherfile import *".

I would argue the opposite.

If I have a large Python project, with big config files, then the added
complexity of moving the config data into separate files using a
non-executable format (say, JSON or INI) is minimal, and keeping the data
and code separate will pay off. But for small Python projects, with only a
few data values, keeping them separate is overkill.

So, to the Original Poster, I think that depending on the size of your
program and the amount of config data you have to deal with, there's
nothing wrong with including it directly in the module.

[Chris Angelico]

On Thu, Oct 2, 2014 at 12:49 AM, Steven D'Aprano

Post by Steven D'Aprano

Post by Chris Angelico
I'd agree, where "trivial limits" is defined by each individual item.
Going with straight Python code is fine for huge projects with long
config files, as long as each config entry is itself simple. You even
get a form of #include: "from otherfile import *".

I would argue the opposite.

Welcome to design debates, where there are as many valid and
justifiable views as there are participants :)

Though I wasn't precluding small config files from being code - just
saying that I don't think size of project is a factor at all.

ChrisA

[Rustom Mody]

Post by Rustom Mody

Post by cl

Post by Rustom Mody

Post by cl
I would actually
quite like to keep the configuration data separate from the code as it
would simplify using the data at the 'home' end of things as I'd just
need to copy the configuration file across. This was why the database
approach appealed at first as all I need to do is copy the database
and everything is in there.

Of course

Post by cl
Are there any better ways of doing this? E.g. some sort of standard
configuration file format that Python knows about?

Umm this is getting to be a FAQ...
Maybe it should go up somewhere?
- ini
- csv
- json
- yml
- xml
- pickle
- And any DBMS of your choice
I guess Ive forgotten as many as Ive listed!!

Yes, I know, I've found most of those. I'm really asking for help in
choosing which to use. I think I can reject some quite quickly:-
xml - horrible, nasty to edit, etc. I don't like XML! :-)

Heh! Youve proved yourself a pythonista!

Post by cl
ini - doesn't work so well with lists/dictionaries (though possible)
csv - rather difficult to edit

Have you tried with comma=tab?

Post by cl
yml - front runner if I go for configuration files

Yeah my favorite as well

Post by cl
json - one of the most likely possibilities, but prefer yml

Seems to be most popular nowadays -- maybe related to being almost yaml
and in the standard lib

Post by cl
pickle - not user editable as I understand it

Well not in any reasonably pleasant way!

Post by cl
What I'm really asking for is how to choose between:-
python - just keep config in the modules/classes, not easy to use
at 'both ends' (home and remote), otherwise quite simple

Can work at a trivial level.
As soon as things get a bit larger data and code mixed up is a recipe for mess up.

Just came across this
https://github.com/henriquebastos/python-decouple/

[cl]

Post by Dave Angel

Post by cl
I am puzzling where and how to keep these configuration values. My
current design has them in dedicated tables in the database but this
is rather clumsy in many ways as there's an overhead reading them
every time the program needs them and changing them isn't particularly
convenient at the Beaglebone doesn't have a GUI so it has to be done
using SQL from the command line.

It is very useful for the database to be self contained and self
describing. Among other things it means that you can easily take
a snapshot at any time, and that snapshot is not tied to any
specific version of the code. As for changing with sql, you can
always add a configuration utility when using sql becomes a
nuisance, or introduces errors.

I guess I can write a script to do the change but it's more code to
write and maintain as opposed to simply editing a configuration file
with tools I know well.

Post by Dave Angel
The key thing is to manage change. You need to decide which
things are conceivably going to change and how you'll manage both
the change itself and the inevitable split between prechange data
and post. One kind of change might be fixing the spelling of
battery. No biggie, just let new accesses get the new one.
Another kind might be the addition of a new instance of an adc
converter. Not a problem, as long as you don't reuse an old
name. And presumably you never remove an old name from the
config.

The only things really likely to change (and may change regularly) are
the conversion factors, drifting voltage references etc. will
inevitably require these to be recalibrated every so often. Other
than that it's just typos or me deciding to change the name of
something.

Post by Dave Angel
More troublesome is adding a new kind of data. You have to decide
whether it's worth generalizing the code to anticipate the
change, or just admit that the code will grow at that point and
that old code won't deal with the new data.

There's a separate module for each input type (just ADC and 1-wire at
the moment), I don't anticipate any more though I suppose there might
be digital inputs one day. So a configuration [file] for each type
seems to make sense, generalising it would be more trouble than it's
worth I think.

Post by Dave Angel

Post by cl
Does it make sense to add them to the modules which handle the reading
of the inputs? I already have modules defining classes called Adc and
OneWire, is it a reasonable approach to add the configuration to these
as, probably, dictionaries? Modifying it would be pretty simple -
just edit the python source (one of the easiest things for me to do on
the Beaglebone as my sole access is via ssh/command line).

Nope, read it from the db. You still might be loading it to a cfg
variable, however.

Post by cl
Thus I'd have something like (apologies for any syntax errors):-
cfg = { "LeisureVolts": ["AIN0", 0.061256, "Leisure Battery Voltage"],
"StarterVolts": ["AIN1", 0.060943, "Starter Battery Voltage"],
"LeisureAmps1": ["AIN2", 0.423122, "Leisure Battery Current"}
(It might be better to makes those lists dictionaries, but it shows
the idea)

Actually it's probably better to use named tuples. All those lists
are apparently of identical length with identical meaning of a
given element offset. But if named tuple isn't flexible,enough,
you probably need a new class.

Post by cl
Are there any better ways of doing this? E.g. some sort of standard
configuration file format that Python knows about? I would actually
quite like to keep the configuration data separate from the code as it
would simplify using the data at the 'home' end of things as I'd just
need to copy the configuration file across. This was why the database
approach appealed at first as all I need to do is copy the database
and everything is in there.
I'm not really expecting quick/glib answers, just some ideas and
comments on the various ways of doing this and their advantages and
disadvantages.

Main next question is whether you can restart each system when you
add to the configuration.

Yes, restarting isn't a problem, the Beaglebone is dedicated to this
task and doesn't control anything so restarts are not an issue.

[Dave Angel]

Post by cl

Post by Dave Angel
....
name. And presumably you never remove an old name from the
config.

The only things really likely to change (and may change regularly) are
the conversion factors, drifting voltage references etc. will
inevitably require these to be recalibrated every so often. Other
than that it's just typos or me deciding to change the name of
something.

Right there you have an important design decision. You then presumably
have to have a way that each particular data point can record what
configuration it's associated with. If you just change the calibration
data externally, then all the old records are invalidated.

Post by cl

Post by Dave Angel
More troublesome is adding a new kind of data. You have to decide
whether it's worth generalizing the code to anticipate the
change, or just admit that the code will grow at that point and
that old code won't deal with the new data.

There's a separate module for each input type (just ADC and 1-wire at
the moment), I don't anticipate any more though I suppose there might
be digital inputs one day. So a configuration [file] for each type
seems to make sense, generalising it would be more trouble than it's
worth I think.

[Neil D. Cerutti]

Post by cl
Thus I'd have something like (apologies for any syntax errors):-
cfg = { "LeisureVolts": ["AIN0", 0.061256, "Leisure Battery Voltage"],
"StarterVolts": ["AIN1", 0.060943, "Starter Battery Voltage"],
"LeisureAmps1": ["AIN2", 0.423122, "Leisure Battery Current"}
(It might be better to makes those lists dictionaries, but it shows
the idea)

Using configparser.ConfigParser to read an ini-format seems like a good
idea. I use it for my own numerous fixed format data file definitions,
and it's been convenient and even extensible.

[LeisureVolts]
Description=Leisure Battery Voltage
Code=AIN0
Value=0.061256

[StarterVolts]
Description=Starter Battery Voltage
Code=AIN1
Value=0.060943

[LeisureAmps1]
Description=Leisure Battery Current
Code=AIN2
Value=0.423122

I've set it up so I can understand abbreviations for the field names,
for when I want to be lazy.

Some of my values are dictionaries, which looks like this in my files
(an alternate spelling of one of the above entries):

LeisureVolts=desc:Leisure Battery Voltage
code:AIN2
value:0.061256

It's simple to hook into ConfigParser to get whatever meaning you'd
like, and whatever verification you'd find necessary.

[cl]

Post by Neil D. Cerutti

Post by cl
Thus I'd have something like (apologies for any syntax errors):-
cfg = { "LeisureVolts": ["AIN0", 0.061256, "Leisure Battery Voltage"],
"StarterVolts": ["AIN1", 0.060943, "Starter Battery Voltage"],
"LeisureAmps1": ["AIN2", 0.423122, "Leisure Battery Current"}
(It might be better to makes those lists dictionaries, but it shows
the idea)

Using configparser.ConfigParser to read an ini-format seems like a good
idea. I use it for my own numerous fixed format data file definitions,
and it's been convenient and even extensible.
[LeisureVolts]
Description=Leisure Battery Voltage
Code=AIN0
Value=0.061256
[StarterVolts]
Description=Starter Battery Voltage
Code=AIN1
Value=0.060943
[LeisureAmps1]
Description=Leisure Battery Current
Code=AIN2
Value=0.423122

That's OK except that it doesn't associate the different sections in
any way, I need another level which includes these to indicate that
these are values read from the ADC. There's another collection of
config data for values read from the 1-wire bus.

I guess I could have two config files, one for ADC and one for 1-wire.

In fact I think that's the best approach so far. Ini format files are
human readable (and more to the point editable), I can store them in
the same directory as the database so they will get carried around
with the data and the configparser module is built-in.