Log file Viewer

[00:42:26] <lifeless> dstufft: ^ breaking out little bits

[02:29:36] <dstufft> lifeless: sorry I havne't gotten back to your PR yet tonight, trying to get some data for a friend for tomorrow morning

[02:32:45] <lifeless> dstufft: de nada

[02:32:58] <lifeless> dstufft: I'm pottering through the consequences of the initial review

[02:33:49] <dstufft> lifeless: for the record, I don't feel very bad if deletes and "wierd" things like that break, though if we can get a solution where they don't that's better. I do care about the leakage between runs with different indexes and such

[02:35:04] <dstufft> I don't feel very bad about duplicating into the wheel cache dir, but I would prefer it not do that just for disk space savings (plus the http cache will handle more edge cases better)

[02:35:37] <dstufft> I think those were the major questions I saw in the bit i read in email

[02:36:09] <lifeless> so the duplicate bit

[02:36:20] <lifeless> I want to ignore, it can be covered by 2567 if/when thats done

[02:36:31] <lifeless> the different indice thing, ok - I will address

[02:38:01] <dstufft> not using the --find-links logic and instead manually replacing links in a cache keyed by the url would solve that problem I think, the different indexes thing and it would also fix the deleted file thing at the same time. Not sure how hard it would be to do that though off the top of my head

[02:38:42] <dstufft> it could also solve the duplicate bit if you just made it conditional on whether the link that PackageFinder discovered was already a wheel or not

[02:39:40] <lifeless> not really, because the logic that triggers that is the same that is used for pop wheel

[02:39:50] <lifeless> the lookaside bit is the read-back aspect

[02:41:12] <dstufft> ok

[02:43:29] <dstufft> lifeless: fwiw I know a good number of people who have been waiting for this feature and who are excited to see it being worked on

[02:45:47] <lifeless> ok so here's a problem

[02:46:29] <lifeless> there's a 1:M relation between sdist and wheels

[02:46:40] <lifeless> for non universal wheels

[02:47:15] <dstufft> hrm

[02:47:28] <lifeless> so any oracle that predicts the wheel for an sdist needs to know all the variables that can influence

[02:47:31] <lifeless> that

[02:49:01] <dstufft> unless looking up an item from the cache just returns all the wheels, and use the Wheel().supported() thing

[02:49:05] <dstufft> to filter the list

[02:49:31] <dstufft> https://github.com/pypa/pip/blob/0c9af4cad94aca9b4367bd4b3b4cc855d27fdb07/pip/wheel.py#L536-L540

[02:49:47] <dstufft> that class just needs the wheel filename

[02:50:27] <lifeless> so, it would be a little wasteful on inodes

[02:50:36] <lifeless> but cache/wheels/$sdistname/wheelfiles

[02:50:38] <lifeless> would work

[02:51:48] <dstufft> cool

[02:52:17] <lifeless> alternatively we could do cache/wheels/$sdistname_$wheelname

[02:52:34] <lifeless> i'm inclined to do the former for now

[18:28:54] <lifeless> o/

[19:14:39] <lifeless> dstufft: so today - https://github.com/pypa/pip/pull/2661

[19:36:37] <lifeless> dstufft: my sketch (untested as yet) of the cache - http://paste.ubuntu.com/10775889/