Re: [PATCH] fstests: generic/362: remove the old file to reflect new mount options

On Thu, May 28, 2026 at 07:12:24AM +0930, Qu Wenruo wrote:
> 
> 
> 在 2026/5/27 22:33, Christoph Hellwig 写道:
> > On Wed, May 27, 2026 at 04:20:38PM +0930, Qu Wenruo wrote:
> > > > I'd rather add a copy of the test that runs on the scratch device
> > > > that can control the environment.  Or maybe just change the test
> > > > to use the scratch device?
> > > 
> > > I'm completely happy with using scratch device instead of test dev.
> > > 
> > > If no special reason to use test dev, the next update will go scratch dev
> > > instead.
> > 
> > I went back and re-read this.  What is the special mount option you
> > change?   If you want to always exercise this as nocow, maybe add a new
> > btrfs test that is a copy and uses the scratch device?  Or am I
> > misunderstanding something?
> > 
> 
> My point is, the current script prevent the test case to respect certain
> mount options, thus reduce the coverage.
> 
> E.g. in a multi-section fstests setup, default and nodatasum mount options
> are defined in different sections, and default mount option will be run
> first.
> 
> Then the next nodatasum section will still utilize the old inode created
> with regular default options, thus reduce the coverage.
> 
> 
> Sure, a dedicated btrfs copy will help, but that will introduce duplication.
> Not sure what is the proper way to handle such situation.

If there is a specific btrfs bug (commit) that can be 100% reproduced by
slightly modifying this case (e.g. use a specific mkfs or mount option),
I think it’s worth splitting it into a separate test case for better
test coverage and known issue tracking.

Thanks,
Zorro

> 
> 
> For now I'll follow Filipe's review to use _cleanup() to delete those
> involves test cases.
> 
> Thanks,
> Qu
>