Re: [Cdk-devel] A suggestion for setting a flag if a preprocessing step has been attempted

[Rajarshi G. <raj...@gm...>]

On Tue, Nov 8, 2011 at 3:28 AM, Egon Willighagen
<ego...@gm...> wrote:

> This has been discussed a couple of times already,

Heh, indeed :)

> Keeping track of modifications
>
> Option 1: the cheap way
>
> Have a get/setIsModified(bool), which gets updated by all methods that
> change the structure, like addAtom(), removeBond(), etc.

Yes, the easy way; but will be tricky to ensure that we have caught
all possible places where a change may occur

> Option 2: the immutable pattern

This is nice, but it seems it has two issues:

* if we go for immutable objects, pretty much all core classes need to
become immutable. Otherwise, from a library user point of view, it
will be a pain to keep track of what is immutable and what is not. So
it's pretty much all immutable or not

* as far as I can tell it does not address the issue of whether an
algorithm (say aromaticity) needs to be run or not. Sure, once a
molecule is constructed it does not change. Removing a bond, would
lead to a new molecule being constructed. But it would still need some
flag to indicate that aromaticity perception needs to be rerun

> The Cache:
>
> The second design decision that needs to be made is where to put the
> cache. Some discussion here is appreciated too. One option is to just
> use the IChemObject.properties... A second option is to use an
> external, possible global cache.

How would a global cache help here? Shouldn't the cache be local to a
molecule (to address the issue of whether a molecule has become
'dirty' or not)? Or are you talking about a pool of core class
instances?

-- 
Rajarshi Guha
NIH Chemical Genomics Center