#general
- b
bored-art-40741
03/31/2016, 6:34 PMIt drops the lock - r
rough-minister-58256
03/31/2016, 6:34 PMso its fundamentally broken in that sense? - b
bored-art-40741
03/31/2016, 6:34 PMI'm not sure what you mean by fundamentally broken. The notion that you can store a message of some kind in your flock file? - b
bored-art-40741
03/31/2016, 6:34 PMI think the answer there is yes - b
bored-art-40741
03/31/2016, 6:35 PMUnless there's further trickiness with appending + only locking a certain byte range I haven't thought of - r
rough-minister-58256
03/31/2016, 6:35 PMby fundamentally broken i mean the flock is no longer intact after writing the pid - b
bored-art-40741
03/31/2016, 6:35 PMAh yes, correct - r
rough-minister-58256
03/31/2016, 6:35 PMah ok - b
bored-art-40741
03/31/2016, 6:35 PM(Though at that point you'd might as well just write a message about who you are rather than writing a pid) - r
rough-minister-58256
03/31/2016, 6:35 PMor not if it breaks things? - b
bored-art-40741
03/31/2016, 6:35 PMSorry? - r
rough-minister-58256
03/31/2016, 6:36 PMif writing a pid breaks you, what good does writing a message do on top of that? - b
bored-art-40741
03/31/2016, 6:36 PMThat's my point - b
bored-art-40741
03/31/2016, 6:36 PMIf we could write a pid, we'd might as well just write a message instead. But we can't do either - r
rough-minister-58256
03/31/2016, 6:36 PMah nm, you’re talking in terms of the .message file - b
bored-art-40741
03/31/2016, 6:36 PMBut yeah, we can do it if we also agree that the lock also locks access to its .message file, by convention - b
bored-art-40741
03/31/2016, 6:37 PMSo you only monkey with the .message file if you're holding the lock - b
bored-art-40741
03/31/2016, 6:37 PMIt sounds like we're okay with that - b
bored-art-40741
03/31/2016, 6:37 PMIf so, question for the gallery: what's the best way to test this thing other than by hand? It fundamentally requires multiple processes and some weird integration trickiness if we're going to test it - w
witty-crayon-22786
03/31/2016, 6:38 PM@bored-art-40741: multiprocessing makes that pretty easy - b
bored-art-40741
03/31/2016, 6:38 PMPerhaps, I'll try it. I was worried about subprocesses inheriting file handles and messing it up - a
ancient-stone-50795
03/31/2016, 6:39 PMI think putting it in a separate file is fine. Currently the ivy lock file is in ~/.ivy2/pants No reason for any of that to be in git. If there is a global lock somewhere else can't we use a pattern in .gitignore? - w
witty-crayon-22786
03/31/2016, 6:39 PMi do wonder what happens when you fork while holding a flock. but i think the test would need to avoid acquiring until after the fork anyway - b
bored-art-40741
03/31/2016, 6:39 PMWe also use this for the pants "global" lock. And yes, we can just gitignore it - b
bored-art-40741
03/31/2016, 6:40 PMAh that's a good point @witty-crayon-22786, the parent doesn't need to hold the lock, just coordinate the children - b
bored-art-40741
03/31/2016, 6:40 PMOkay I think I can test it thoroughly then - w
witty-crayon-22786
03/31/2016, 6:40 PMcombined with `multiprocessing`'s own locks, shouldn't be too bad. - w
witty-crayon-22786
03/31/2016, 6:41 PMthanks for tackling that. - b
bored-art-40741
03/31/2016, 6:41 PMSo right now I'm zeroing in on a fully green jenkins run - a
ancient-stone-50795
03/31/2016, 6:41 PMYes, thank you for working on this👌 1