On 11/06/18 01:16, L A Walsh wrote: > Pádraig Brady wrote: >> Rather than add a new option, I thought it cleaner >> to extend the existing --preserve-root option to support >> --preserve-root=all, as that can be interpreted to >> preserve the root of all specified file systems. >> > --- > My first comment was about hoping it wasn't going to become a > default
Not on by default. Only applies to directories passed as arguments to rm. > but more importantly: a mount != the root of a volume. > > so 2) One might have many subdirs of a volume remounted in other > locations. If the goal is to protect the root of a volume > and not merely all test & tmp that have sub-volumes mounted > on them, it seems prudent to test that the mount is > indeed the root of a volume. The intent is to protect directory mount points. > 3) Also, I've often used tmp dirs for testing that I > later wipe w/rm-fr, especially if I reuse the volume > for multiple tests. Scratch, test and development > volumes are often easily cleared off with rm. > > 4) are you only protecting mounts on a cmdline? or any mount > crawled by rm underneath cmdline args? Cuz if it is the 2nd, > wouldn't that already be handled by --one-file-system? Yes that's an additional protection one might like to enable. > 5a) I believe I pointed out that --one-file-system was flawed > in that it didn't limit deletions only to 1 file system, since > the option isn't really named --one-file-system-per-arg. > How many people want to delete multiple file systems at one > time by specifying them on the command line to clean out a dir, > like rm foo/* -- where * expands to various mount points > in the dir. > > > RFE1: Is there a reason not to fix that? I.e. check that all args > are on 1 file system if that option is specified? > > RFE2: Could '-x' as a short-op for that? I can handle short > flags and would really rather see --one-file-system be enforced > at the cmdline lvl as its name infers. > > *cheers*... I considered having the patch use --one-file-system=all which would cater for your use case I think with a single option rather than two. Though that would be a bit less flexible as one then couldn't protect only cli args. I'm not fully decided on whether that flexibility is needed. cheers, Pádraig
