Criticism of previous versions - DEPRECATED
In order to do things better, the first step is to acknowledge that some aspects were not as good as expected, and so are candidates for improving.
I guess the main problem of µOS++ SE was the complex structure, making hard to find something without an accurate indexer or a full search.
Other critics:
- no integration with Eclipse; the entire build system, based on xCDL was highly functional, but being written in Python, had no chance of being integrated into Eclipse (anyway, this was a rapid prototyping phase, to prove the concept)
- the complete separation from standard C/C++ libraries requires changes in the sources
- the C++ only, without any possibility to call anything from C, restricted usability
- the lack of standard interfaces for network and file systems
Valued experiences, to be kept:
- the idea behind the configuration mechanism implemented by xCDL
- the separate name spaces for the standard C++ libraries allowed a good testability of library functions
Lessons for the new version:
- allow to write C applications, even if linked with C++ libraries
- stay as compatible as possible to C11 and C++11 (threads and other special features)
- eventually implement a CMSIS-RTOS compatibility layer
- use a standard startup/exit code
- use standard testing infrastructure (Google tests, running on POSIX synthetic platforms and over semihosting on embedded ARM platforms)
- use ARM CMSIS for C projects
- use the POSIX read()/write() API for network and storage
- more usage models (copy files to local project, create libraries, reference to several central repo of libraries), with better Eclipse integration
- by default the standard libraries should be in the standard namespace, but for testing it should be possible to generate separate namespaces
