What would it take to remove boost from XRT? about xrt HOT 7 CLOSED

I understand your motivation.
But from the bigger picture point of view, this part just™ works™, which is an undeniable quality. :-)
Boost is widely used world-wide and not the worse thing to use or to package.
At the end it might just add to the NIH syndrome which is already a problem.
If you have some spare brain cycles and extra budget, there are other few issues to fix I can see in my priority list: just make it to compile #7806, cleaning up the code (you can dive into https://github.com/Xilinx/XRT/tree/master/src/runtime_src/core/pcie/emulation), improve Vitis and ROCm packaging... :-)
But anyway, just my personal feeling.

from xrt.

@newling This is work-in-progress. We are removing boost dependencies as we type.

I think we are very close to have no link time dependencies left in the xrt shared libraries, but we have a few inline boost library dependencies some of which can be replaced as we move to use more modern c++. However boost property_tree is the more difficult one. What is the problem with the few inline libraries being used?

As for linking with boost, the stand-alone tools part of XRT do bring in boost library dependencies, but that is not really a problem related to IREE's use of XRT.

from xrt.

@keryell Your ubuntu23.10 issue is assigned and being fixed now.

from xrt.

@stsoe it is good to hear that you're close to having no link time dependencies to boost for the xrt libraries. Header only dependencies are much better (I didn't realize property_tree was header only).

from xrt.

@stsoe do you have a rough idea of when XRT will be link time dep free? Weeks/months?

from xrt.

Hi @newling. I don't see any boost link time dependencies in the shared libraries even today. So this was implemented with the removal of boost::filesystem within the last month or so.

from xrt.

Super, thanks Soren.

from xrt.