Community
Participate
Working Groups
Having a generic architecture interface, will make easier to implement initialization and de-initialization of features specifc of each platform
New Gerrit change created: https://git.eclipse.org/r/97837
Gerrit change https://git.eclipse.org/r/97837 was merged to [develop]. Commit: http://git.eclipse.org/c/4diac/org.eclipse.4diac.forte.git/commit/?id=822ea47f9d5012c464588399fa2beb7b23dfe0e7
This is still open. the idea is to have the same interface for all architectures. What has been done in rcX was a test to see if it fits. forte_instance.cpp/h files are the proposed interface, and forte_architecture.cpp is the file that has to be implemented for each arch. A unique main file only could then use these files. I thought that modules could be also defined in CMake and then generate somehow the initialization, since many of them must be initialized first before using it. A special case arose with ROS in posix, where the module needs a parameter.
So as the last try was done in 2017 according to the bugs gerrit comments, and your comments is this even feasible? Or to ask it in another way, is the effort worth the benefit?
I like the benefit and I think the effort is not too much, at least for most of the archs, since no new code is necessary, mostly reorganizing. The only thing missing is to define a proper interface in forte_instance. Right now, depending on the architecture, starting forte is different for each one, and when a new platform is developed, this part of the code must be copied. Having the same interface will help specially when forte is used as library. The modules idea is another thing, and maybe that could go in another bug, if necessary.
Yes, lets move the modules into a new bug, so that we get manageable chunks. So what's then the proposed timeline for this bug?
My idea was to have it for 1.10 release. I'd say at least discuss and define a proper interface (if possible) before the release, and if time and human-resource are avaiable, implement it before it.