You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Jul 31, 2019. It is now read-only.
RE: My recent PR we know that marshal is dangerous but fast. However doing some rudimentary testing I think using struct + simple byte concat is faster on pre-Python3.5 (apparently someone decided to overhaul the marshal serialization I suppose). Here's a rudimentary test script I used to run some very basic timings. Using this approach means we have to deal with unicode vs. bytes ourselves, but since clog converts lines into UTF-8 anyways I don't think that will be much of an issue. How about stream names? It doesn't look like the ScribeLogger converts that to UTF-8, so we might need to reconstitute the unicode format by storing the encoding somehow.
Anyways, struct + raw bytes would be safer than marshal and would work in this case since we can make some strong assumptions about the arguments. I'll look into changing it but I think we should get the current PR in first just to fix the marshal.loads problem.
The text was updated successfully, but these errors were encountered:
RE: My recent PR we know that marshal is dangerous but fast. However doing some rudimentary testing I think using struct + simple byte concat is faster on pre-Python3.5 (apparently someone decided to overhaul the
marshal
serialization I suppose). Here's a rudimentary test script I used to run some very basic timings. Using this approach means we have to deal with unicode vs. bytes ourselves, but since clog converts lines into UTF-8 anyways I don't think that will be much of an issue. How about stream names? It doesn't look like the ScribeLogger converts that to UTF-8, so we might need to reconstitute the unicode format by storing the encoding somehow.Anyways, struct + raw bytes would be safer than
marshal
and would work in this case since we can make some strong assumptions about the arguments. I'll look into changing it but I think we should get the current PR in first just to fix themarshal.loads
problem.The text was updated successfully, but these errors were encountered: