Re: [Unicon-group] Proposal of naming conventions in IPL

[Majorinc, Kazimir]

Actually, on the second tought, best one can do with 'shallow' packages is 
to divide total of n names into sqr(n) packages of sqr(n) names, with name 
collision problem reoccurring in each of the packages and between them. 
With deeply nested packages, there would be no collision, but reference of 
the global would require many prefixes separated with "::": (for binary 
tree hierarchy log2(n) of them), and many new package names. Every package 
solution is between those two extremes; either there is too many (possibly 
duplicated) names in packages and library or too many prefixes. Note also 
that good organization of the hierarchy is very ambitious pursuit, even if 
lot is known (see Britannica Propaedia); if future development of the 
library is unpredictable, yet backward compatibility needs to be preserved, 
it cannot succeed. Packages implementation problems (I see only name 
leaking from imported (and inner) packages toward used but undeclared 
locals in users program) can be added on top, but it is less important.

Even with packages, one can apply unique names (and solve mentioned 
problems), but then packages only complicate things. Without intention to 
be disrespectful - obviously serious work is done on packages 
implementation -- I think it would be wise not to include them in the first 
release distribution, it will procrastinate satisfactory solution and 
require further maintenance. It is not irreparable harm, any case, 
impossibility of the name collision and free mixing of any two parts of 
library (even without link/import, theoretically, linker can find unique 
names on his own) suggest unique name as more promising approach.



----
Kazimir Majorinc, Zagreb, Croatia