Open PallHaraldsson opened 1 year ago
It's common if you time using three times in a row, that second time is also slow, I think it's an indication of something (packages that load quick tend to have second (and third) fast). Here I think it's unusual that there are more allocations the second time. I've not noticed such before, not sure why or if some hint. It may not be unusual, only me noticing first now.
This is when loading the package causes the code loading code itself to get invalidated.
I don't really understand the rest of the post.
A.
For Kristoffer at #704 (unclear which Julia and HTTP versions used):
vs. for me on 1.5.0, and Julia 1.8.1 (I don't think I had MbedTLS nor that it would matter):
It's common if you time
using
three times in a row, that second time is also slow, I think it's an indication of something (packages that load quick tend to have second (and third) fast). Here I think it's unusual that there are more allocations the second time. I've not noticed such before, not sure why or if some hint. It may not be unusual, only me noticing first now.1.4.0 had slightly lower allocations, so also a regression, which brings me to an unrelated point:
B. Strange installation experience.
why did I not get 1.5.0 straight away, which was installable? To be fair, the installation of 1.4.0 was smooth (didn't crash my computer), unlike for 1.5.0.
For 1.5.0, from syslog, julia add triggered OOM, felt unresponsive for like 15 min. recovered, even firefox, even though sub-process of killed:
To my surprise this DID install (i.e. no other packages blocking going to new[est] version; the problems not related to your package, and "fatal" not really fatal), also way to many upgrades, thus precompilation, e.g. Graphviz_jll must be unrelated to HTTP:
Note I killed with CTRL-C (which usually works, just losing the blinking cursor, as here; a known problem), likely would have finished otherwise (or OOM, and julia Pkg sub-process failed, still recovered). I'm not blaming your package (some other might), and too much memory on install a known problem (fixed in newer Julia allegedly, but then seems the fix maybe didn't work).