| *** tristan <tristan!tristan@121.136.56.237> has quit IRC | 05:19 | |
| *** tristan <tristan!tristan@14.39.170.85> has joined #buildstream | 07:39 | |
| *** ChanServ sets mode: +o tristan | 07:39 | |
| *** tristan <tristan!tristan@14.39.170.85> has quit IRC | 09:41 | |
| nanonyme | Jürg Billeter: does buildbox-casd persist hardlink information? | 20:07 |
|---|---|---|
| juergbi | nanonyme: no, CAS doesn't currently have any provisions to persist hardlink information (outside a non-standard 'NodeProperty') | 22:07 |
| nanonyme | Should it? It seems like a useful thing to have | 22:08 |
| nanonyme | I guess it might not work without assumption of buildbox-fuse anyway though | 22:10 |
| juergbi | nanonyme: not sure it's really important (except for exactly matching the result of building outside BuildStream) | 22:10 |
| juergbi | in the context of build outputs, it's typically just an optimization | 22:11 |
| juergbi | e.g. for hardlinked binaries that change their behavior depending on argv[0] | 22:11 |
| nanonyme | Hmm, right but won't not implementing hardlinks actually break those? | 22:12 |
| juergbi | as it's content-addressable storage, it would be fairly simple to create hardlinks of binaries based on matching digests, if that's desired | 22:12 |
| nanonyme | Unless someone manually breaks hardlinks and replaces them with symlinks during build | 22:12 |
| juergbi | I could see BuildStream needing some support for that, though, but it might not need a REAPI extension | 22:12 |
| juergbi | it won't break but it could lead to excessive package sizes if there is no deduplication when packaging | 22:13 |
| nanonyme | Hmmmm, right, true. The check for name would still work | 22:14 |
| juergbi | i.e. I consider it mainly an optimization issue, although a potentially crucial one | 22:14 |
| juergbi | if improvements are necessary, it's probably best to consider it based on a concrete use case | 22:15 |
| nanonyme | I wonder if it would actually make the non-buildbox-fuse staging faster | 22:17 |
| nanonyme | In cases where we have hardlinks and large files | 22:18 |
| juergbi | I think the only reasonable staging without FUSE is the use of reflinks | 22:31 |
| juergbi | in which case, hardlinking a few binaries wouldn't provide a significant performance benefit | 22:32 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!