@Ronounours

Here is an example of regression in 1552, there may be others ...

gmic 256,256,1,3 -spline 127,5,0,0,127,210,800,-400,1,255,0,255

Command '-spline' has been recoded as a custom command in G'MIC (instead of a native one, in C++), and maybe you can consider this is a regression, but from my point of view, this is more an improvement
Having more and more custom commands (written in G'MIC language instead of in C++) is a great thing, as in one hand we gain in flexibility, and on the other hand we can better cope with necessary API changes if required in the future, without the need to release a new binary version. In your case, that's only the meaning of the arguments that has been changed, nothing really problematic. I would suggest always making your filters compatible with the latest stable version, as the API is converging to something more and more robust, so this means less changes to do in the future.
In your case :

gmic 256,256,1,3 -spline 127,5,0,0,127,210,800,-400,100,255,0,255

does the trick.

David.

PS : Also, almost all reasonable developers will tell you that breaking some parts of an API is good. This is not only to bother you, but to converge into something better. The code of open source projects can be seen by everyone, so probably, most of the open source developers are concerned by the 'beauty' of the code and the API, rather than keeping the backward compatibility at any price (which often means does some crappy hacks in the source code).

I've done as well the necessary changes to include by default all external filters for the next release.
The current beta I have display more than 170 filters even without update