Re: ezunits fundamental_units quite slow

Dear Robert,

> I'm pretty happy with ezunits as a whole at this point, and I feel
> like some substantial speed-ups might be pretty easy to implement, so
> I will try to take a look at that in the near future.
Indeed I am also quite happy with ezunits. I recently ported a largish 
computation (~150 pages as PDF export) from Mathematica to Maxima and it 
is faster in Maxima. I can't as yet how much because I am not yet 
finished with the port and also I did some things slightly different. 
But it is definitely not so that ezunits is painfully slow in general. 
The computation has complicated antiderivatives (for some I use FriCAS* 
via a simple export/syscall/import method) which leads to huge terms and 
ezunits works just fine with this. What I can say is that the unit 
system in Mathematica is were verbose on input and very hard to use. The 
interface of ezunits is very nice.
> I suspect a lot of the slow operations have to do with processing the 
> same expressions over and over and over.

I am not sure the slowness of fundamental_units can be fully explained 
this way. As far as I understand after a quick look it uses a 
combination of rules, substitutions, recursion and infinite evaluation 
with a large list of unit conversions. The tricky thing is that units 
can be defined based on other units. What I had in my patched version 
was computing the fundamental units right away on definition of a unit 
and hashing them. This has the effect than one cannot change units later 
any more - or that derived units would not change with them - but this 
is not a sensible use case IMHO.

Let me also have a look and let's discuss optimization opportunities.

> You may be interested to hear I have pushed commit 6da388 which
> implements a new function, all_fundamental_units, which converts all
> explicit dimensional expressions (i.e., stuff like a ` b) to
> fundamental units. This is a function which can be applied explicitly;
> it isn't applied automatically, since in a lot of cases one would want
> to work with non-fundamental units.

In the case I mentioned I convert all inputs to fundamental units and 
convert the output. If I would compute with standard units I would 
anyway end up with stuff worse than J^9/(Pa^8*mm^8*m^16) for J - the 
(automatically) reduced fundamental units are more readable in such 
cases - and this also makes addition work without rules. But for simpler 
cases all_fundamental_units is definitely nice.

The caching works fine for me as a work around. I even dump the cache 
contents in the end and initialize it with the dump, so I have no 
fundamental_units calls any more unless I add new stuff.

Best regards,

Michael

* I will start a separate thread on this topic.