Get the bundled JAR size under control #911
Comments
|
The overheads:
|
|
JENA (due to its size) and xerces (due to the potential issues of having various XML libs on the path) seem like primary candidates for removal if possible. What is the zipped SDF in the cdk-builder? |
|
Removing Xerces does not remove functionality. So, removing that from the bundle makes sense. I also understand the argument for removing the |
|
w.r.t to iordf I wonder if we just write our own owl reader/writer. I presume the specs of these are super extensive though so might be an impossible task. Otherwise we could make this an optional dependency and exclude it from the bundle. i.e. if you want this functionality you probably already have JENA on your class path so are OK to pay that cost. It is frustrating how much it's bloated, the older JENA was quite compact. @uli-f The Zipped SDF is 3D templates, improving the 3D generation capabilities would perhaps mean it could be removed. |
|
As a follow on to #915 is there a relatively straightforward way to generate the WAR file as above? |
|
Set packaging type of cdk-bundle to “war” instead of “pom” then add an empty web.xml file in the correct place (it tells you where). But I’ve given the sizes there, what rcdk needs is another bundle like pom of it’s own that only includes what it needs. |
|
Although I think as rajarshi pointed out, it basically is just all functionality made available so really “everything” is needed but choices can be made to cut the less essential things. |
|
@johnmay, how are we doing? |
Note the RInChI inclusion will make this larger so it's a losing battle :-). |
|
41 still down from the peak of 45 but still larger |
|
Up to 42 again... |
The newer Apache JENA (cdk-iordf) version pulls in a lot of dependencies which ultimately means our "cdk-bundle" size is nearly double. A neat trick to work out where bloat is coming from is to create a WAR file.
Currently we are now sitting at 45MB (up from 26MB for CDK 2.7.1). I would hate for this to spiral further and up in the RDKit's 700MB conda install size territory :-).
Here is the break down:
If I remove cdk-iordf from the bundle it drops back down to 26MB. Which is still big IMO but doubling size between releases is not ideal.
The text was updated successfully, but these errors were encountered: