@ChShersh Maybe some of this would be easier with re-use of existing tech? For example, write packages for your language in an existing package manager. Extend an existing build tool to support your language. Formatters might be able to support some styles reusably from a parse tree.
@wuoulf@condaforge I have a little half-baked kakoune conda package that I usually rebuild whenever I setup a computer that I'm going to be using for a while. `make` usually works out of the box as long as your C++ compiler is recent enough.
@incomputable This might be what I need to start adding types to more quantities which shouldn't be mixed up. Making the safe thing easy is probably better than the benefit from explicit to_underlying.
@mycoliza I think the magic here is all of the bugs which the db authors have already found and addressed such as when popular filesystems don't meet their own guarantees. The D for durability is really hard and you shouldn't invent your own lightly if you need it to work.
@MKBHD I do app updates infrequently unless I need to. In a world of move fast and break things, my apps are less likely to work after update than by not updating. At least if I wait until my current version breaks, I know your new update won't make things worse.
@lemire This changes when you get roommates. With poorly controlled conditions (eg. desktop OS), memory usage needs ROI. Some Chrome memory has good ROI, but diminishing returns ended before it starts stealing from the rest of my applications.
@jntrnr @awesomekling I'd recommend occasionally trying dprintf to dynamically add print statements in gdb to skip the re-compile part of the cycle or to filter output.
I think the pollution of optimizations aren't talked about enough. Most important is thread pollution from multithreading excessively or outside the bottleneck. TLB ๐ฉ from random read access. L1 pollution from lookup tables. Branch predictor pollution. Monomorphism inst cache ๐ฉ
@emollick I don't see how we can get away from an emphasis on critical thinking and detecting errors including hallucinations in every class. It is already out there in media and it will be a part of many workplaces.
y-cruncher v0.8.1 is now ready. Speedups on of #AVX512 on #Zen4 grows from 2% to 20+%.
Download here: https://t.co/U26kSl0Jqv
https://t.co/hWvKRBJ0s1
If you're into bignums, The Hybrid NTT algorithm that was the backbone for years is now detailed here: https://t.co/F5SNXvNdAl
@bcantrill@oxidecomputer@ahl Just thinking about how it would work if people could subscribe to moderation of content separately from the content itself.