On Mon, Apr 03, 2023 at 09:30:22AM -0700, Tyler Retzlaff wrote: > Improve portability of telemetry code to allow it to be compiled by msvc > unconditionally. > > Remove use of VLA and instead dynamically allocate. MSVC will never > implement VLAs due to misuse / security concerns. > > Remove use of ranged based initializer (a gcc extension) instead just > explicitly initialize individual array elements. > > Tyler Retzlaff (2): > telemetry: use malloc instead of variable length array > telemetry: use portable syntax to initialize array > Is this worth doing, given that DPDK telemetry uses a unix domain socket for it's connectivity, which would not be available on windows anyway? I don't particularly like these patches as: * The removal of the VLAs means we will potentially be doing a *lot* of malloc and free calls inside the telemetry code. It may not be a data path or particularly performance critical, but I know for things like CPU busyness, users may want to call telemetry functions hundreds (or potentially thousands) of times a second. It also makes the code slightly harder to read, and introduces the possibility of us having memory leaks. * The second patch just makes the code uglier. True, it's non-standard, but it really does make the code a whole lot more readable and managable. If we need to make this standards-conforming, then I think we need to drop the "const", and do runtime init of this array with loops for the ranges.
All that said, if we do have a path to get telemetry working on windows, I think we can work together to get a suitable patchset in for it. /Bruce
