Howard Chu wrote in b0311a63-3757-5b69-676a-bd8293e34c1e@symas.com: |steffen@sdaoden.eu wrote: |> Full_Name: Steffen Nurpmeso |> Version: 0.9.23 |> OS: Linux |> URL: https://ftp.sdaoden.eu/lmdb-bogo.tar.xz |> Submission from: (NULL) (109.40.130.211) |> |> |> A follow-up to private communication with Quanah Gibson-Mount. |> You can find more in the tarball linked below |> |> |Your README states that the behavior in LMDB 0.9.23 is different than \ |in 0.9.22 but that |is incorrect. No changes to set_mapsize() behavior were made between \ |0.9.22 and 0.9.23.

Yeah!! Mind you, i have even looked into the git history once AlpineLinux switched to 0.9.23 and i came to the very same conclusion!

But one thing is plain: on AlpineLinux, with musl libc, from July 2018 to January 2019, i ran bogofilter with LMDB 0.9.22 exactly as it is still in bogofilter upstream and on the one branch in the ball i provided under the URL above. Doing monthly dump + rebuild of my spam database to vacuumize the thing. Without any problems. Then after my weekly system upgrade, which (also?) came with LMDB 0.9.23 i suddenly saw crashes, on a Saturday night! You can read the two mails [1,2] if you wanna.

[1] https://www.bogofilter.org/pipermail/bogofilter-dev/2019q1/003515.html [2] https://www.bogofilter.org/pipermail/bogofilter-dev/2019q1/003516.html

|Closing this ITS.

Well ok, if it is so designed that set_mapsize() is not meant to survive a transaction abort, then it may be so. But then it should possibly be documented that after a transaction abort all environmental settings need to be reestablished, as they vanish after an abortion. I.e., that they only become persistent after a successful commit.

--steffen | |Der Kragenbaer, The moon bear, |der holt sich munter he cheerfully and one by one |einen nach dem anderen runter wa.ks himself off |(By Robert Gernhardt)