On Tue, Jul 23, 2024 at 10:17:24AM -0400, Theodore Ts'o wrote: > On Tue, Jul 23, 2024 at 03:39:04PM +0200, Christoph Hellwig wrote: > > > > At least in my case it's not really by overriding. E.g. if I force > > an allocation group (or block group in extN terms) to a specific size > > and then want a log that is larger than that, changing the AG size > > is generally a bad idea, and a clear warning to the user is simply the > > better interface. > > Is it just "a bad idea", or "it won't work"? I can imagine that > sometimes we want to have tests that do things that are generally a > bad idea, but it's the best way to force a particular corner case to > happen without having to run the test gazillions of times? > > I do remember some cases where when we were using a 1k block size, the > test wouldn't actually work because the file system needed to be > bigger or the metadata overhead ended up causing an ENOSPC too early, > or something weird like that. So that was a case were the merging > would _work_, and in fact was testing a combination that we actually > wanted to test --- but we had to adjust the test subtly so it would > work both on a 4k block size and a 1k block size. I don't remember > which test it was, or we hacked it, but I'm fairly certain it's > something we've done before. It's messy. > > > Merging the options is what we're currently doing, and it works ok > > most of the time. The question is what to do when it doesn't. > > No matter what, it seems like we'll have to look at each of these > tests and treat them on a per-case basis. We could have options which > allows the test to specify that it shouldn't be merging; but then we'd > still have to decide what we need to do. And what do we do if we > don't want to merge for ext4 and xfs, but it would be useful for btrfs > (for example) to merge the options. It's probably also going to > depend on which test scenarios that various file system developers' > test setups choose to use.... The big question I have is: for at least the standard -g all runs, does this decrease the number of tests selected? AFAICT all it does is converts mkfs option parsing _fail into _notrun, but a 35k shell script patch is a lot to take in. If it doesn't change the number of tests selected to run then I think I'm ok with this. --D
