Skip to contentSkip to conversation

fstests: Don't use gawk's strtonum breaking existing fstests

DISCUSSION Nikolay Borisov 2 participants 4 messages Dec 11, 2019 – Dec 12, 2019 Btrfs file system

lore
Share & tools
lore
#1 lore raw
Quoted 7 lines · Nikolay Borisov

Hello,

Following upstream commit:
https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/commit/?id=37520a314bd472ed720ed0611c6b69e418be9b61

breaks btrfs/095 and btrfs/098 tests.

The problem is that the old code was using gawk's strtonum and returning a base 10 number from an octal input. Whereas the existing code gets an octal number which is again parsed as an octal when using printf. So the path in question needs to either by reverted or extended so that the necessary conversion from octal to base 10 is performed _before_ calling printf.

#2 lore raw
Quoted 11 lines · Nikolay Borisov
Quoted 7 lines · Nikolay Borisov

Hello,

Following upstream commit:
https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/commit/?id=37520a314bd472ed720ed0611c6b69e418be9b61

breaks btrfs/095 and btrfs/098 tests.

Quoted 7 lines

The problem is that the old code was using gawk's strtonum and returning a base 10 number from an octal input. Whereas the existing code gets an octal number which is again parsed as an octal when using printf. So the path in question needs to either by reverted or extended so that the necessary conversion from octal to base 10 is performed _before_ calling printf.

Well, octal values really makes no sense. We should go either hex, or human readable decimal.

Octal should only be left for certain historical use cases, like user privileges. For content dump, octal is never a good use case to show offset.

Since current filter_od can't handle -Ax yet, what about just use hash instead of the problem prone od?
And put the od output into seqres.full for later debug?

Thanks,
Qu

#3 lore raw
Quoted 25 lines · Qu Wenruo
Quoted 11 lines · Nikolay Borisov
Quoted 7 lines · Nikolay Borisov

Hello,

Following upstream commit:
https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/commit/?id=37520a314bd472ed720ed0611c6b69e418be9b61

breaks btrfs/095 and btrfs/098 tests.

Quoted 7 lines

The problem is that the old code was using gawk's strtonum and returning a base 10 number from an octal input. Whereas the existing code gets an octal number which is again parsed as an octal when using printf. So the path in question needs to either by reverted or extended so that the necessary conversion from octal to base 10 is performed _before_ calling printf.

Well, octal values really makes no sense. We should go either hex, or human readable decimal.

Quoted 6 lines

Octal should only be left for certain historical use cases, like user privileges. For content dump, octal is never a good use case to show offset.

Since current filter_od can't handle -Ax yet, what about just use hash instead of the problem prone od?
And put the od output into seqres.full for later debug?

I disagree, using the output of od one can see the actual content of extents and can quickly diagnose the failure.

Quoted 4 lines

Thanks,
Qu