Closed bmatsuo closed 10 years ago
Performance appears to be approximately the same as master. The choice of io.Copy (edit: io.CopyN) is probably not the most performant, it will create its own buffers, but its use is not on hot code paths so I didn't optimize it.
Sorry. The description in the pull requests was slightly incorrect. I made no substantial changes to the existing tests. So you can feel somewhat safer knowing that what was being tested before is still being tested in the same way.
nice work @bmatsuo, thanks.
Will look it over more closely...
last thing I'm thinking through is this EOF
business...
@mreiferson what is giving you pause about the EOF changes? The reasoning or the implementation (which basically amounts to the "noeof" and "noeof64" functions I added)?
OK, I'm good.... nice work :+1:
Want to squash?
Squashed.
Thanks for putting up with me @mreiferson. I'm just super excited about snappy right now. I want this package to be rock solid and rip through streams lightspeed. You know? :smile:
hah, np, thanks for fixing up my shitty code :100:
Fixes #5 along with other bugs related to chunk size and handling of EOF.
simplify reader tests.(edit: I ended up leaving the existing tests mostly the same)