IRC channel logs
2017-06-01.log
back to list of logs
<paroneayea>ym: it's in it a state where it runs, but not in a state where you want to use it :) <wingo>nalaginrut: i don't think it is worth it. language is too large; a waste of time <rekado_>I had a really boring dream in which I implemented R on top of Guile. <nalaginrut>rekado_: I hope you did blas too, that's really important <wingo>nalaginrut: honestly i think finishing lua would be a more finishable task :) <rekado_>nalaginrut: it was a dream, not a nightmare. <nalaginrut>wingo: well yes, I just hope the methodology could do more things ;-) <paroneayea>did documentation disappear with geiser support and guile 2.2? <amz3>OrangeShark: I tried distcheck in guile-git but it failed <OrangeShark>amz3: hmm some sort of failure in creating files from the tests? <amz3>OrangeShark: tx! Have a good day in the meantime :) <OrangeShark>amz3: I have question about wiredtiger, what is prefix compression and how is it good for time series? <amz3>prefix compression is like it will group all rows that starts with the same prefix <amz3>so you have 111 112 113 114 etc... collocated <OrangeShark>oh, since time series will be dates, it can easily group them? <amz3>and compressed, I guess it means it doesn't repeat the prefix <amz3>it also support classic compression with bzip2 and snappy, but it must be recompiled <amz3>I am not sure how to activate prefix compression <amz3>it must be a configure flag <amz3>I read prefix compression is desactivated by default <OrangeShark>thanks, was just curious after I read your latest email release for your bindings <amz3>I read it's compression is useful to save space <wingo>hehe i have that problem sometimes :) <wingo>ACTION trying to figure out how much goops class redefinition is worth <sneek>civodul, you have 1 message. <sneek>civodul, bavier` says: re mumps/metis, the error looks familiar so I'll take a look at it <wingo>paroneayea: how do you feel about class redefinition <paroneayea>wingo: I feel it's very useful for live hacking, as a user <paroneayea>wingo: I don't understand how it works fully under the hood, and it seems like the mechanisms are complex <paroneayea>wingo: one example where I've found it useful is in 8sync the mapping of action of a message (a symbol) to a specific message handler is done on a class level <paroneayea>I can change how actors work while the program is running <paroneayea>and it was nice because I didn't have to restart the server as I was walking around and changing thigns <wingo>in 3.0 i would like to simplify structs. there they incur some overhead in space and time because they support the goops feature of class redefinition <wingo>i.e. a struct with one field has three words: one for the vtable, one for the single slot, and one pointing to the slots <ijp>death to fat structs! <wingo>to get to slot 0 you have to do a double-dereference: first to get the slots, then to get the slot <ijp>or is this what the face of modern spam looks like <wingo>i think this is just the face of modern spam :/ <ijp>I'm partly impressed, partly very confused <paroneayea>wingo: I personally think that low-level structs shouldn't need to be redeinition'able <wingo>paroneayea: yeah i guess that's where i'd like to go. i wonder though... <paroneayea>wingo: can goops slot rederinition still work even if low-level records don't follow that? <wingo>like there's some GC implications that if we had a better GC, maybe we could support redefinition without the indirection <wingo>that would require a moving gc though <wingo>paroneayea: yes it can but a goops object will have to contain that indirection itself <wingo>basically all goops objects would become two-word structs; boxes that can be mutated <paroneayea>wingo: I expect GOOPS to be a bit more expensive than say srfi-9 records <paroneayea>wingo: would it really be much more expensive leaving the indirection in goops than it is having it in the lower layer? <wingo>paroneayea: i think it would be more expensive but i don't know by how much. <wingo>but anyway i think we can continue to support it and not too expensive, so it sounds like a win <wingo>in some ways i would like the expressiveness of goops without the overhead of the ability to redefine classes, but i think we can't abandon that capabilit <paroneayea>wingo: which other expressiveness... generic methods and etc? <paroneayea>wingo: though iirc you were hoping to get generic methods to work with srfi-9 records too right? <wingo>the expressiveness of goops relative to just records, i mean <wingo>so yeah, generic methods and such <wingo>good question about class-of on srfi-9 records tho <paroneayea>wingo: if we could get records to support generic methods, we might be able to have goops basically be the place to go when you do want inheritance and/or redefinition live hacking foo, but have records be a lot more expressive than they are now <paroneayea>over time I've come to feel like generics have simplified a lot of my code, though this may be influence from that Sussman talk on live hacking