validation: change UpdateTip for multiple chainstates
ClosedPublic
Actions

Authored by PiRK on Oct 17 2022, 15:03.

Details

Reviewers

Fabien

Group Reviewers

Restricted Project

Commits

Summary

Only perform certain behavior (namely that related to servicing
the getblocktemplate RPC call) for the active chainstate when
calling UpdateTip.

Co-authored-by: Jon Atack <jon@atack.com>

Notes:

the main difference from the source material is that we do not have the code handling BIP9 deployments in UpdateTip. As a result, there is no need for the warning_messages arguments in UpdateTipLog
the const qualifier for the CblockIndex *pindexNew argument in UpdateTip was missed in D1972

Test Plan

ninja all check-all

Repository

rABC Bitcoin ABC

Lint

Lint Passed

Unit

No Test Coverage

Build Status

Buildable 20551
Build 40772: Build Diff	lint-circular-dependencies · build-without-wallet · build-diff · build-debug · build-clang-tidy · build-clang
Build 40771: arc lint + arc unit

PiRK created this revision.Oct 17 2022, 15:03

PiRK requested review of this revision.Oct 17 2022, 15:03

Fabien added inline comments.

src/validation.cpp
2138
2147	Does that make sense to always log (with no category for filtering) when this is not the active chain ? What's the result in the log ?
src/validation.h
1175	This was missed in D1972

PiRK edited the summary of this revision. (Show Details)Oct 18 2022, 08:04

~~this->~~CoinsTip();

src/validation.cpp
2147	I think it makes sense. The philosophy here is that the background chainstate is the important one, the one that goes through the complete IBD and verifies everything. And the assumed valid chain in foreground is just temporary, until the IBD caches up with the current tip. It is going to be verbose for the whole duration of the IBD, and the background chain will print most of the messages, so it will probably make the new blocks added by the active chain every 10 minutes a bit hard to spot, but that's not a problem with a negative grep : `tail -f debug.log \| grep -v [background validation] "` I vote for sticking with the upstream choice on this one. In the current master branch of Core this is still logged nconditionally.