Re: RFS: cdk 1.4.9-1

[Niels Thykier <niels@thykier.net>]

On 2012-05-23 13:37, Egon Willighagen wrote:
> NIels,
> 
> On Wed, May 23, 2012 at 10:04 AM, Niels Thykier <niels@thykier.net> wrote:
>> On 2012-05-19 14:56, Onkar Shinde wrote:
>> It seems that some public methods are changing signature and also that
>> some public constants are disappearing in this release (compared to our
>> current version).
> 
> Which ones are those? The 1.4 series should be API stable... I love to
> hear how you tested that!
>

Hi,

I "just" read the diff between our current version (1.2.10) and the RFS
version 1.4.9.  Actually, thanks you just made me realize that it FTBFS
when built twice in a row!

Anyhow, it is quite possible that I saw false positives as I just read
the diff once without considering new interfaces/super classes and
whether the methods were "defined internal" (despite being public).
Anyhow, to mention examples:

PartialFilledStructureMerger:
 -public IAtomContainer generate() throws CDKException
 +public IAtomContainer generate(IAtomContainerSet atomContainers)
         throws CDKException

SmilesParser:
 -private IAtom assembleAtom(String s) throws InvalidSmilesException
 +private IAtom assembleAtom(String s, IAtom lastNode, boolean
          bondExists) throws InvalidSmilesException

CDKSourceCodeWriter*:
-public void writeAtomContainer(IAtomContainer molecule)
         throws Exception {
+private void writeAtomContainer(IAtomContainer molecule)
         throws Exception {


IChemObjectBuilder lost a lof interface methods and it does not extend
any other interface that could have gained those methods.

IBond:
-public int getStereo();
+public IBond.Stereo getStereo();

These are examples of what I found; they may (or may not) be an issue.
I also noticed now that some of the files disappearing appears to
auto-generated JJTree or so?  Anyway, with the clean rule not working my
diff is currently a bit tainted.


*) Probably intended to be private from the beginning, but changing the
visiblity from public -> private is a possible source of breakage.

>> Have you (build-)tested the reverse dependencies of cdk?
> 
> I do not think there are any reverse dependencies... ? I wish there
> were, and I hope Bioclipse will be one of those; but at this moment I
> cannot think of any... I cannot find right now where to find that rev
> dep info...
> 
> Egon
> 

For reverse dependencies (not including build-depends) I notice at least
3 (potentional) rdeps in Debian:

$ apt-cache rdepends libcdk-java
libcdk-java
Reverse Depends:
  debichem-cheminformatics
  science-chemistry
  python-cinfony

Actually, the first two turns out to be meta-packages, but
python-cinfony appears to be a genuine reverse dependency.

~Niels