How to interface with sparrow from external code?

[PaulDL-RS]

Dear Jeroen,

As mentionned before, I am writing a Python wrapper around sparrow (https://github.com/PaulDL-RS/spyrrow).

As I didn't fully understand the instance conversion, I chose to use jagua-rs::JsonInstance as the input interface and SPSolution as the output interface.
Do you have additional suggestion for a better interface like SPInstance ?

Additionally, for JsonInstance, should I add checks for 0.0 and -0.0 rotation angles ?
Or are equivalent angles handled gracefully anyway further down the line ?

Yours,

Paul Durand-Lupinski

[JeroenGar]

Hi,

Both SPInstance and SPSolution are internal representations for strip packing (SP) problems in jagua-rs.
Multiple modifications to the original input shapes are performed to aid the optimization process.

I made a rough sketch of the flow here:

JsonInstance is the right interface to 'feed' a instance into jagua-rs/sparrow.
For now, the only way to 'export' a solution from sparrow is via SVG, which for your use-case is not ideal.

In the reference heuristic (lbf) made for jagua-rs, I also implemented export via JsonOutput (in addition to SVG).

I think this would be the way forward.
Feel free to contribute, or otherwise, I'll try to get it done in the coming weeks.

Alternatively you could write an entirely custom input parser and exporter.

Best,
Jeroen

[JeroenGar]

As of #71, final solutions are now also exported as JSON files in addition to SVG.
I also spent some time cleaning up the io module in jagua-rs to improve understanding of how instances are imported into the library and solutions are exported out of it.

Now you should be able to use JsonInstance and JsonSolution (or something similar to this) to interface with sparrow.

What remains is the configuration of the algorithm's parameters.
Currently, these are all hardcoded in config.rs (easiest option during early development).
But ideally, at least a subset of these parameters should be configurable at runtime.

[PaulDL-RS]

Thank you for your input, I will investigate how to use JsonSolution as the Output Interface.

Regarding the parameters, would a parameter struct with pseudo defaut values help with configuration ?

Best,
Paul

[JeroenGar]

Yes, a config struct with defaults is certainly the way to go. However I will have to think about how to implement this in an elegant way as these parameters are used throughout the codebase and global mutable variables are against idiomatic Rust.

Also, not all parameters have to be configurable by the end-user.
I intentionally designed the algorithm not to require instance-specific tuning.
Probably only the parameters that need to be exposed are the ones that need to change depending on the permitted computation time.

I'll get to it, but it might take some time and iterations to get it done properly.
Should not be a blocking issue for spyrrow in the short term anyway.

[JeroenGar]

All parameters of the algorithm have now been split into config.rs, for parameters which are end user/instance specific, and consts.rs, for all parameters which remain fixed.

I'll close this issue and convert it into a discussion