I'm now getting the same failing gerrit behaviour as rzezeski reported recently

error: remote unpack failed: error Missing tree 1d7a8b7ebbc1933fb023ae22621e4bdc34fd2654

maybe something to do with a newer git version, I'll try the --no-thin

yes, that worked

is git maybe defaulting to thin?

w/ the newer version

I can't see anything in the release notes. My client recently upgraded from 2.37.3 -> 2.37.5

andyf: I found in the past a git fetch of the remote was fine.

And cleared it up.

Yeah I frequently do "git fetch --all --prune" before I do anything else

FWIW we run "git gc" against the repository each day

To keep it from growing forever

Maybe that's having an impact

any cases where we've done history-changing operations (like git commit --amend or rebase) on published branchese?

(would be interesting to know what the missing tree actually was - it might be present in the client repository)

not for years.

usually when we've missed a tree like this it's because there's disagreement about where the master branch points

git ls-tree 1d7a8b7ebbc1933fb023ae22621e4bdc34fd2654 in my checkout shows some version of what looks to be ./usr/src/uts/intel/io/vmm

but that's not the most recent version of that tree

but it's the version of usr/src/uts/intel/io/vmm in f2357d972e19ef505ad57ff2f0ee340fa2214015 ("15236 viona ring page holds should heed pause/resume")

(you can see this with git ls-tree f2357d972e19ef505ad57ff2f0ee340fa2214015:usr/src/uts/intel/io

)

that's a pretty recent commit (December 2022 author date..)

committed January 4th 2023

tree was was most recent version of that subtree until 62ac5a8a4bcebad1d1c5639636a038087707a5e0 ("15251 panic in vlapic_get_ccr after live migration / 15312 bhyve has overzealous vlapic ID check") committed on January 10th 2023.

(in case you're wondering how I found the commits: git log --oneline -- usr/src/uts/intel/io/vmm | while read commit rest

do

echo $commit; git ls-tree ${commit}:usr/src/uts/intel/io | (grep 1d7a8b7ebbc; exit 0)

done

(the (grep ... ; exit 0) is to avoid having my shell warn me about failing greps)

the old one-two, grep || true

lots of ways to spell that.

So there are a handful of dangling commits

gc failure dropping something that was actually referenced? eek.

I don't think so

I think it's an artefact of commits submitted for review being actual commits under magic refs

we've definitely definitely seen this before transiently

(that is: I believe the five dangling commits are changes that are not integrated, and is presumably an artefact of an incomplete submission)

The other issue I dunno about

But I expect I have never seen it because I fetch aggressively

It could also be because I'm a slacker that doesn't do that much work

:D

same, on both counts

my git stuff keeps itself current for me

I think the time I saw something like this I had a 'master' that gerrit didn't, or vice versa.

i'm curious as to where "error: remote unpack failed: error Missing tree 1d7a8b7ebbc1933fb023ae22621e4bdc34fd2654" came from. Something logged by gerrit or by the git client doing the push?

(or, well, git-as-run-by-gerrit)

[illumos-gate] 15345 Fix function prototype in in.routed/if.c -- Marco van Wieringen <marco.van.wieringen⊙pen>

Thanks, I'll try doing a pull from the gerrit upstream, I don't do that routinely. This was a new commit in a branch just taken from master, so I was surprised that it didn't work

I think a pull from gerrit gives the client a better idea of what gerrit does and doesn't have already..

(pull before push, that is..)

Midvale Repository for the Gifted

:)

jclulow: hey, but while you're in there, great time for that upgrade... :)

oh goodness my fridge is running brb

heh.. always fun when you're doing a test, pretty sure it should be harmless and the network link to that location dies seconds after you run your command :)

did everything get untangled now then?

jclulow: i'd imagine it can't get too far

most gifted thing I've seen Gerrit do is attempt to create reviews for every single existing commit in the repo when it was confused about where I was pushing to.

I think it might have some safeties against users making that error now

(touch wood)

I have twice now almost pushed a large feature branch by mistake, because habit of which workspace I'm in

luckily, I also habitually -nv

:)

jclulow: this happened >10 years ago and my typo might well have been one of the incidents motivating the creation of the safeties.

did anyone want more time to look at #15260 or #15361? (fenix?)

BUG 15260: ld crashes if group members are stripped by -s (In Progress)

↳ illumos.org/issues/15260 | code.illumos.org/c/illumos-gate/+/2601

and the other, bot...

illumos 15361 (fenix) ?

BUG 15361: ld doesn't fill out PT_DYNAMIC sufficiently for binutils 2.40 (In Progress)

↳ illumos.org/issues/15361 | code.illumos.org/c/illumos-gate/+/2620