FWIW though I did inspect the "pex" file before, and it was only four bytes long
We wrote a little program called jarcat that we use to concatenate zip (.jar, and .pex) files together which is must faster than unzipping and rezipping them. Maybe it's not happy with the .zip for some reason.
Oh well that seems wrong...
well, now I have a different problem!
failing to import the numpy C extensions
You probably need to mark numpy as not being zipsafe. There's a flag on the pip_libarary()/python_wheel() rules.
but I'm running with a fresh clone of the codelabs
the problem is that python libraries sometimes expect to find resources as files and it can't if they're xipped up
Bah it's probably trying to compile numpy from sources because your platform doesn't have a .whl available or something
that might be it - I guess I can try replacing it with a
python_wheel rule instead of pip_library...
I guess that'll add a bunch of extra complexity though
Yeah but those expect you to host them yourself... pip is a nightmare to interface with. It's so inconsistent across platforms. Sorry about that.
I might try and go through the codelabs on manjaro. I think I have a vm for it somewhere. A little chocca a the moment though.
Yeah, no worries! I'll keep hacking around a bit and come back here if I come up with anything useful.
Thanks for your help!
No worries, the long term plan is to fork pip and bend it to my iron will but I've got v16 to get out the door first.
3 replies
@tatskaari:matrix.org ^ sorry wouldn't let me send in thread
The logs are from Please, but the direct install of numpy with pip produces an error with the same root cause when importing numpy from a python shell.
(Symbol cblas_sgemm not found)
Possibly numpy expects libatlas to be installed / linked?
(Symbol cblas_sgemm not found)
Possibly numpy expects libatlas to be installed / linked?
It doesn't support cross compiling so you have to have CI agents for each arch you want to support (unless you home brew your installs)
Jonathan Poole: It might be worth getting our local python_wheel hack cleaned up - because you don't have to host the wheel externally - you can just build the wheel locally and feed the OUT directly to python_wheel
2 replies
e.g
I have please in a docker image on windows. It's able to build but the next step is failing. Docker logs end at [4/5] and the next command seems to be 0:plz run: plz-out/bin/module/aws/file.sh base dev and the error message is: CRITICAL: Error running command plz-out/<path>file.sh: no such file or directory. What can be wrong here?
5 replies
@Tatskaari: at least with WSL1 you would bind mount /mnt/c to /c and put your $HOME in the /c/Users/<user> - and then the host access / file system stuff work mostly work properly between something like Desktop Docker and your WSL sessions.
Although, the WSL fs-layer is trash, so you are pretty much better off running Please with Linux inside a VM.
And if you need an IDE, VS Code's ssh-remote works quite well
Native windows support it rising up my priority list though. A lot of the C++ guys I've been speaking to say Bazel's msvc support is janky and their dep on Java is a dealbreaker
If we solve those for them, they'd be more likely to take Please over CMake
Is native windows support something that would benefit anybody here that's already using Please?
I don't remember having this in the past.
thanks
that's the last thing before I start writing the doc :)
mostly dockerless docker
b
;)
thanks