18:05:01 <Lemmih> stepcut: Around?
18:16:03 <Lemmih> stepcut: Dude, where are you?
18:31:23 <donri> O brother stepcut, where art thou
21:31:28 <dcoutts_> stepcut: re binary and error messages, not not much better but at errors returned rather than calling error
21:31:49 <dcoutts_> stepcut: to do a better job one really needs an encoding with more redundancy
21:31:55 <dcoutts_> I've got some ideas there
21:32:09 <dcoutts_> and I'd like to do that for a future binary package, but it's not changed yet
21:54:55 <donri> dcoutts_: is it possible to do that without a performance hit?
21:55:18 <dcoutts_> donri: it's unclear
21:55:33 <dcoutts_> on the one hand we could probably be more compact than we are now
21:55:52 <dcoutts_> which has advantages for disk and memory bandwidth
21:56:00 <dcoutts_> on the other, the encoding would be more complex
21:56:03 <donri> also not sure how to parse "re binary and error messages, not not much better but at errors returned rather than calling error"
21:56:15 <dcoutts_> donri: oh as data rather than exceptions
21:56:21 <dcoutts_> sorry, could have been clearer
21:57:09 <donri> i think you missed a few words, maybe. "at least errors are returned" perhaps?
22:01:08 <dcoutts_> yep
22:01:14 <donri> i'm a bit curious if the problems related to the error handling in cereal are now also relevant to binary... i don't fully understand those problems though
22:02:10 <Lemmih> I've patched SafeCopy to add pretty "stack-traces" to error messages.
22:02:20 <donri> iirc something like, it resulted in code in cereal doing things like, strictly read something twice, to get the length before, or such
22:02:52 <donri> *vague is my middle name*
22:03:38 <Lemmih> Dag V. Odenhall
22:03:57 <donri> :)
22:04:29 <donri> stepcut said something to that effect, that the Either errors in cereal made it difficult to be "properly lazy" (or strict?) or something
22:05:22 <Lemmih> I think that's a feature, not a bug.
22:06:07 <Lemmih> The parsing library should be strict. Any laziness or staggered parsing should be layered on top of it.
22:06:38 <donri> doing inefficient things unnecessarily isn't really a feature :) but yea i don't actually have any idea what i'm talking about here
22:06:52 <donri> stepcut: y u no active
22:34:26 <dcoutts_> donri: the latest binary is actually very like cereal
22:34:44 <dcoutts_> particularly in the error handling and incremental input
22:34:51 <dcoutts_> since they're now both continuation based