13:36 <_whitenotifier-f> [prjxray] fox6666 opened issue #1490: PR bitstream size - https://git.io/Jke6S

15:34 <whitequark> mithro: i figured out why the conda installer doesn't work on debian

15:35 <whitequark> when your install script runs pip, it really does run pip from the conda environment, and that uses the conda python

15:36 <whitequark> however... conda python, for some incomprehensible reason, loads a third of the packages from the conda env, a third from ~/.local, and a third from /usr

15:36 <whitequark> in other words, a classic conda experience

16:03 <whitequark> here's why it's broken: https://stackoverflow.com/a/51640558/254415

16:03 <tpb> Title: Global pip referenced within a conda environment - Stack Overflow (at stackoverflow.com)

16:18 <whitequark> https://github.com/QuickLogic-Corp/quicklogic-fpga-toolchain/issues/5#issuecomment-723464697

16:53 <mithro> https://github.com/SymbiFlow/symbiflow-examples/tree/master/eos-s3

16:58 <mithro> @whitequark: I like you wish there was a better option than Conda, but so far have not found one.

17:00 <mithro> We are slowly working towards supporting both docker and static compiled binaries

17:01 <mithro> umarcor is doing some excellent work in that direction

17:02 <mithro> Pip and yowasp style stuff is also interesting

17:04 <whitequark> mithro: I'm not actually opposed to using Symbiflow's conda with isolated environments

17:04 <whitequark> conda in this case is an implementation detail that I (in theory) do not ever have to touch or care about

17:05 <whitequark> unfortunately, in practice, the way symbiflow invokes conda is deeply flawed, and I'm having to fix that

17:07 <whitequark> why does it download a 12 MB file from sharepoint and feed it into tar?!

17:07 <mithro> @whitequark: any idea how we could "poison" the system stuff so CI explodes if it is accidentally used?

17:08 <whitequark> run it on a fresh debian install *grin*

17:08 <mithro> @whitequark: that is the VPR architecture description - it should be more sensibly generated by CI soon

17:09 <whitequark> let me do a more serious check though

17:09 <mithro> I wonder why we didn't add debian to https://github.com/SymbiFlow/symbiflow-examples/blob/master/.travis.yml

17:09 <mithro> Brb just getting up

17:09 <whitequark> i'm not completely certain what the conditions for blowup are

17:11 <whitequark> mithro: with the conda env activated, run: python -c "import sys; assert all('/usr' not in path for path in sys.path)"

17:26 <mithro> @whitequark BTW the way the install "should work" is -- (step 1) download+install conda, (step 2) create environment using environment.yml file, (step 3) start doing stuff

17:27 <mithro> There is currently a (step 2b) which is "manually download the vpr configuration data" which will hopefully go away soon

17:28 <mithro> @whitequark Where we are and where we want to be are never quite the same :-/

17:29 <whitequark> it's not really that hard to write a working install script, which i have demonstrated by spending a hour today to do so

17:30 <whitequark> all i had to do was to read the conda docs

17:30 <whitequark> anyway, as long as you fix it, it's fine

20:34 <_whitenotifier-f> [make-env] mithro opened issue #19: Convert to using GitHub actions for CI (rather than Travis CI). - https://git.io/Jkvmy

20:36 <mithro> @whitequark - https://github.com/SymbiFlow/make-env/pull/18

20:39 <whitequark> great!

20:59 <mithro> whitequark: I think a bunch of the issues we see in conda world stem from there being is a pretty strong mismatch between what the people who are paying the conda devs to do work (mostly enterprises like universities) and the majority of users (scientists/programmers) really want.

21:03 <whitequark> I used to have a very negative opinion of conda; I no longer really think that for a few different reasons

21:03 <whitequark> I think Symbiflow shipping a conda-based toolchain that does not expose the end user to conda issues is basically fine. not perfect, but fine

21:06 <whitequark> what I really hate though is having to work around conda issues myself, having spent several years doing that as an employee

21:07 <whitequark> anyway, once you address all of the issues I filed, I think the Conda distribution of the toolchain should be usable for most people as long as they never actually touch the conda environment inside

23:31 <mithro> whitequark: I'm mostly distracted with finishing off getting stuff published for the skywater-pdk stuff, hopefully acomodi or someone else will be able to finish off that stuff

23:41 <whitequark> sure, that works