Wondering if you saw:

Effective June 1, 2021: Phabricator is no longer actively maintained.

Repeatable read is required if statement based replication is used - if row replication is used - repeatable read is no longer a requirement for replication consistency...

On my personal resource limited RasberryPI alike server (4x1.2ghz ARM A53, 2GB ram, php 7.4.12) running Phabricator - this changes saves ~15ms of runtime (180ms vs 167ms after) when running:

./bin/herald test --object T19 --type HeraldManiphestTaskAdapter

Rebase

This change is probably no longer relevant with recent large refactoring of arcanist

@epriestley - empty inline comment with suggestion renders rather peculiar transaction which is somewhat confusing and odd... Any plans on tackling it? Like See context greyed out...

In D21225#271823, @epriestley wrote:

I'm not aware of any current customer use cases for this behavior. I'm open to restoring it if customers are actually using it, but would like to make sure the API is fully supporting whatever the need is.

What is now programmatical way to "submit" unsubmitted inline comments? Some users are using it to post comments from bots, now if we use differential.createinline api method all bot comments are visible as unsubmitted and looks like there is no way to set it in stone...

This is breaking commit breaks drydock: PHP Fatal error: Cannot override final method Future::getResult() in /core/data/drydock/workingcopy-75/repo/phabricator/src/applications/harbormaster/future/HarbormasterExecFuture.php on line 50

• Fix pattern

In jira 8.6.1 settings are now in:

• Administration → Applications → Application links

Reproduction script for some working checked out repository on master and repository permanentRefs is master:

head -c 20 /dev/random | base64 > giberish
git add -A
git commit -m 'Add giberish'
HEAD=$(git rev-list --reverse HEAD~..HEAD)
title=$(git log -n1 --pretty=format:%s HEAD)
title_q=$(printf '%s' "$title" | jq --raw-input --slurp .)
rawdiff=$(git diff-tree --root --cc --no-commit-id HEAD|jq --raw-input --slurp .)
diffid=$(echo "{\"diff\":${rawdiff}}" | arc call-conduit differential.createrawdiff | jq -e -r .response.id )
revid=$(echo "{\"diffid\": $diffid, \"fields\":{\"title\": $title_q}}" | arc call-conduit differential.createrevision | jq -r .response.revisionid)
git push origin HEAD:refs/heads/non-permanent-refs/D$revid
arc diff -m 'Message' --update $revid --head $HEAD "${HEAD}^"
sleep 60
# make sure pushed branch is parsed and then push changes, need to make sure metadata Differential Revision is not added so we cannot use `arc land` here
git push origin master

I will try to provide a script to reproduce flow one of our tools does while creating this regresion some time later today/tonight.

In D20880#264515, @epriestley wrote:

Oh, sorry, I think I misunderstood what you're describing in your test plan -- you're saying that you're currently observing the revision not closing, and believe this fixes it. I'll see if I can reproduce this.

Thanks!

This now renders properly; see T13429 for followup.

Forgot to mention that ordering set is relevance and also same issue happens if you search in maniphest, eg.: https://secure.phabricator.com/maniphest/query/Gk8I9e13KiTw/?after=12975

Uh, sorry, didn't mean to step on toes, take your time :) Just wanted to share/upstream our findings which improve overall performance for large phab installations

Anything? This improves herald rule performance...

Uh, we had couple of cases and this caused extra 500s in logs...

• Throw error if unknown ssh option is passed

• Allow only SendEnv=GIT_PROTOCOL ssh option

Old arcanist

Guys any progress with it... It is really hitting pull workers hard which spin in loop doing nothing but spaming logs about repository not pullable anymore...

Any progress?

Thanks @chad for a tip!

Fixing lint issues