I diskussionen om den seneste række rettelser, der blev foreslået til inklusion i kernen 6.11-rc5 af Bcachefs' forfatter, Linus Torvalds, nævnte han, at han begynder at fortryde, at han har accepteret Bcachefs-filsystemet i kernen. Torvalds' utilfredshed skyldes, at Kent Overstreet, forfatteren af Bcachefs, sender for omfattende rettelser i de finale kandidater til udgivelser, som i realiteten ikke kun retter fejl, men også fortsætter udviklingen af funktionaliteten, selvom funktionelle ændringer kun tillades i den indledende fase af udviklingen af en ny gren, og der i rc5-fasen kun accepteres enkelte små rettelser, der ikke bør overstige 100 linjer kode.
Den indsendte patch til Bcachefs, som tilføjede 1309 linjer og fjernede 671 linjer i 39 filer, omfattede udover fejlfinding også to betydelige ændringer: støtte til en ny datastruktur til at håndtere listen over frie elementer i cachen og muligheden for at transformere hash-tabellerne til cachen. Den nye struktur blev tilføjet for at eliminere unødvendige cachelås og undgå situationer, der kan føre til lås-konflikter under flertrådede belastninger.
Dette er ikke første gang, at store rettelser til Bcachefs sendes efter ændringsvinduet, og det er ikke første gang, at mere omfattende funktionsudvidelser fremmes snarere end simple fejlfindinger, der brings frem via pull-anmodninger i de finale udviklingsfaser af en ny gren. Linus Torvalds svarede på den indsendte pull-anmodning:
«Stop. The last pull request was also substantial. This pull request is too large as well, encompassing things unrelated to Bcachefs, and it isn't even remotely a regression. At some point, 'fixing something' just turns into development, and this is that moment. No one in their right mind uses Bcachefs and expects stability from it, so every user is an area for experimentation. The patches for Bcachefs have become a sort of 'large development during the release preparation cycle, not before it,' to the point that I'm starting to regret adopting Bcachefs. If Bcachefs cannot function properly according to the main (upstream) kernel development schedule, perhaps it shouldn't be in the main kernel. This is now beyond ridiculous.»
Kent Overstreet responded that Bcachefs is definitely more reliable than Btrfs and he is working to make it more robust and reliable than XFS and EXT4, thanks to the implementation of an end to end data integrity mechanism. Kent also emphasized that the proposed changes are directly related to improving reliability, and if a bug in other core subsystems might just lead to crashes and reboots, in file systems the cost of a bug is very high and can lead to hidden data corruption that might only be noticed weeks later.
Torvalds countered that there are still no major Linux distributions using Bcachefs, and pointed out that the rules for releasing Linux kernels should be followed, as there is a risk of introducing new bugs due to such large patches containing more than 1000 lines of code. Currently, the discussions regarding fixes to Bcachefs have not been accepted into the Linux kernel, and the Bcachefs developer has not yet prepared an updated version of the pull request that eliminates the addition of new functionality and focuses solely on bug fixes.
Kilde: opennet.ru
