Define preferred notation of units in models

[PTraeder]

Since #348 it is possible to write 1 ms as shorthand for 1*ms.
Additionally, predefined variables for each unit have been registered, so that notations like foo/ms or 1 ms * mV have become possible.
In the case foo/ms the factor 1 is implied for the ms unit.

@jougs and @heplesser expressed that they would like to see literals in front of every unit.
I have some questions regarding this:

A verbose interpretation of this demand could look like:
1 ms / (1 mV * 1 nS) vs. ms/(mv*nS)
foo * 1 nS vs. foo * nS
foo / 1 nS vs. foo / nS
1 nS * foo vs. nS * foo
1 nS / foo vs. nS / foo

I have concerns regarding the legibility of the first case in large expressions.
An alternative could be that we only require one literal per "compound unit", where a compound unit would be a factor in an expression that consists only of units, e.g.:
1 ms/(mV*nS)

Which version, if any, would you prefer?

Are there any other conventions wrt. notation of units that you would like to see?

Also would you like to see some form warning if the conventions we agree upon are breached?

[PTraeder]

@DimitriPlotnikov @Silmathoron

[Silmathoron]

I'm not sure @heplesser said he wanted to see litterals in front of every unit...
I think his point was that, in the examples (for the sake of clarity), the unit should always be attached to the variable it belongs to.

E.g. for PSConInit_E nS/ms = 1.0 nS * e / tau_syn_ex, their logic would be to keep it that way, since it is composed by:

• a normalisation factor 1 nS
• a dimensionless value e
• a time which comes with its own unit value.

[PTraeder]

I honestly do not understand the pattern you are describing.

[jougs]

What @heplesser and I suggested (and what @Silmathoron propably meant) is that we should always have one single normalization factor (usually 1 as an integer, not real) in compound initialization expressions, instead of none at all.

In other words, we rather want 1 ns * e / tau_syn_ex instead of just ns * e / tau_syn_ex as it currently is used in the changes introduced in #348 . In the same spirit, we also want compound initializing expressions to be 1 ms / (mv * nS) instead of 1 ms / (1 mV * 1 nS) or ms / (mV * nS). This is in accordance with common practice in physics and neuroscience.

I think this is all possible with the implementation proposed in #348. The code does just not reflect this convention or at least this is just not used consistently (cf. old and new line 89 of models/aeif_cond_alpha.nestmlin the linked PR).

@PTraeder: does this clear up the confusion?

[PTraeder]

Yes. Thank you, I can work with that.

[heplesser]

I just want to confirm that @jougs understood me correctly.

[Silmathoron]

Is there anything left to do with this issue or should we close it?

[jougs]

@Silmathoron: I don't think so. Thanks for the heads up.