💬 andrewtoth commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664508903)
I guess `head` could be though of as a sentinel, which always points to the head of the linked list? I suppose I misnamed that then.

💬 ajtowns commented on pull request "Early logging improvements":
(https://github.com/bitcoin/bitcoin/pull/30386#issuecomment-2206858748)
Easiest way to test is to add some log lines into init.cpp `InitLogging()`, eg:

```c++
for (int i = 0; i < 10000; ++i) {
LogInfo("log spam %d...\n", i);
}
LogWarning("awoogah\n");
LogError("internal bug not detected\nhello, world!\nwas that fun?\n");
```

Running `bitcoind -regtest -logtimemicros` should give a decent indication of the differences.

If you want to modify bitcoin-chainstate to see what the logs would look like, remove the `DisableLogging` call
...

💬 paplorinc commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664539266)
the exact equivalent, yes, but you might not want to test the dirtiness part here - up to you, I don't mind either way.

💬 paplorinc commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664542828)
hmmm, why did this become `=` instead of `|=`?

✅ ajtowns closed a pull request: "logging: Simplify edge cases in logging configuration"
(https://github.com/bitcoin/bitcoin/pull/30384)

💬 ajtowns commented on pull request "Logging cleanup":
(https://github.com/bitcoin/bitcoin/pull/29798#issuecomment-2206882030)
Consider also updating the help text for `-debug` and `-debugexclude`:

```diff
--- a/src/init/common.cpp
+++ b/src/init/common.cpp
@@ -27,9 +27,9 @@ void AddLoggingArgs(ArgsManager& argsman)
{
argsman.AddArg("-debuglogfile=<file>", strprintf("Specify location of debug log file (default: %s). Relative paths will be prefixed by a net-specific datadir location. Pass -nodebuglogfile to disable writing the log to a file.", DEFAULT_DEBUGLOGFILE), ArgsManager::ALLOW_ANY, OptionsCategory::
...

💬 andrewtoth commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664555939)
Whoops bad rebase, thanks for catching!

💬 andrewtoth commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664558253)
But since these are unit tests we are kind of testing the internals, so it would be good to catch if any other flags sneak in. With `IsFresh` it just checks that the FRESH flag is set, but any other flag could be set as well.

💬 andrewtoth commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664558577)
Added `noexcept` everywhere.

💬 ajtowns commented on pull request "Logging cleanup":
(https://github.com/bitcoin/bitcoin/pull/29798#issuecomment-2206898641)
ACK df2422c5f91e5b1b42d6b718cc2527dc413712fc ; looks like it behaves correctly to me

💬 stickies-v commented on pull request "Testnet4 including PoW difficulty adjustment fix":
(https://github.com/bitcoin/bitcoin/pull/29775#discussion_r1664572723)
Is there a reason this can't be simplified to:

```suggestion
if (nHeight % consensusParams.DifficultyAdjustmentInterval() == 0) {
```

💬 paplorinc commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664545575)
what happens if this ends up throwing, it's not a trivial function (e.g. `m_prev` could become nullable in the future and cause havoc)?

🤔 paplorinc reviewed a pull request: "Don't empty dbcache on prune flushes: >30% faster IBD"
(https://github.com/bitcoin/bitcoin/pull/28280#pullrequestreview-2157030164)
I ran the benchmarks without prune (on the previous push), with just a lazy `-reindex` or `-reindex-chainstate`, but couldn't measure any difference - do I need more than 200k blocks? Or is prune critical here? Or is lazy reindexing not enough?

```bash
sudo ./src/bitcoind -stopatheight=200000

sudo hyperfine \
--warmup 1 --runs 5 \
--show-output \
--parameter-list commit master,670084adc53e3d661ea5b4a19743659609a42d5e \
--prepare 'git checkout {commit} && git reset --hard && make -j10
...

💬 paplorinc commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664560161)
maybe it could make sense to document these rare conditions with something like `} else if ([[unlikely]] fOk) {`

💬 paplorinc commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664547028)
does the naming inconsistency apply here as well - given that we ignore the first element with instant `.Next()` as a first step?

💬 andrewtoth commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664597181)
Yeah I suppose we can rename it `m_flagged_sentinel` or something? Do you have a suggestion that would have made this clearer if that was the name originally?

💬 andrewtoth commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#issuecomment-2206960214)
> I ran the benchmarks without prune (on the previous push), with just a lazy -reindex or -reindex-chainstate, but couldn't measure any difference - do I need more than 200k blocks? Or is prune critical here? Or is lazy reindexing not enough?

It shouldn't have much effect on non-pruning nodes, so it's good that there's no regression.
You will need pruning enabled to see the performance benefits. The higher the dbcache and lower the prune settings the better. The higher the block height the b
...

💬 paplorinc commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#issuecomment-2206961708)
> You will need pruning enabled to see the performance benefits

That requires a bit more setup to be able to do it offline

💬 paplorinc commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664601643)
we have "sentinels" on both ends now, right? A dummy at the beginning and a `nullptr` on the end. This is what I found confusing in the test, why not store the actual head and have a `nullptr` or `sentinel` at the end only? I haven't investigated this, maybe I'm completely off again...

💬 andrewtoth commented on pull request "Don't empty dbcache on prune flushes: >30% faster IBD":
(https://github.com/bitcoin/bitcoin/pull/28280#discussion_r1664604753)
Neat! Could also add `[[unlikely]]` here https://github.com/bitcoin/bitcoin/pull/28280/files#diff-f0ed73d62dae6ca28ebd3045e5fc0d5d02eaaacadb4c2a292985a3fbd7e1c77cR189