Compilation issues

[mehrdadn]

Isn't Array::RangeEquals purely virtual and unimplemented in PrimitiveArray? How is PrimitiveArray being instantiated?

[pcmoritz]

Hey Mehrdad,

this was changed very recently: apache/arrow@cd1d770

For now you can just use an older commit of arrow. At the moment, much of the code is just a quick hack to measure the performance of arrow. It was also written at a time when arrow was much less complete than it is now (and we knew that, so didn't try hard to make the current numbuf library nice); at some point during the summer, we will write a better version of the library that can be open-sourced, but it is not super high priority right now.

-- Philipp.

[mehrdadn]

Oh! It totally didn't occur to me to check if arrow had changed underneath. Got it, thanks!

[mehrdadn]

Another question: it seems like logging.h is taken from arrow, but both of them are included (indirectly) by worker.cc. Is this intentional? Did you want to get rid of one of them? I'm not even sure how it's compiling right now to be honest, my compiler is complaining about the redefinition of DCHECK_LE but not about the redefinition of FatalLog...

[pcmoritz]

Hey,

the compiler is not complaining about FatalLog, because it is in the numbuf namespace. Not the DCHECK macros however, they should probably be prefixed by NUMBUF_ (or removed). Let's change it together when we meet tomorrow!

-- Philipp.

[mehrdadn]

Whoops, momentarily forgot the namespace change :) sounds good!

[mehrdadn]

Hey, could you fix this Windows compatibility issue? In src/pynumbuf/numbuf.h, change

    ftruncate(fileno(file), size);

to

#ifdef _WIN32
    _chsize_s(fileno(file), size);
#else
    ftruncate(fileno(file), size);
#endif

[pcmoritz]

working on this in ray-project/ray-legacy#123