There's a non-trivial performance regression in the Connect postprocess time. Part of the problem is that the vector is not preallocated & the elements are copy-constructed. This commit should fix those two concerns, and address the performance regression significantly.
s/o to @morcos for noticing that this was a slowdown.
3018 | @@ -3019,14 +3019,17 @@ static void NotifyHeaderTip() { 3019 | bool ActivateBestChain(CValidationState &state, const CChainParams& chainparams, const CBlock *pblock) { 3020 | CBlockIndex *pindexMostWork = NULL; 3021 | CBlockIndex *pindexNewTip = NULL; 3022 | + std::vector<std::tuple<CTransaction,CBlockIndex*,int> > txChanged;
In c++11 the ugly '> >' (with a space in between) is no longer needed. Might as well fix it when you're changing it.
I mean at this point thats just like...coding style, right?
Ok, maybe I incorrectly assume that I'm not alone with finding this ugly and distracting :)
I think we should just wait for the eventual clang-formattening; unless you really need that byte back.
I don't think a mass clang-format will happen. It's just too invasive everywhere at the same time. It couls be done piecewise like the -Woverwrite changes, but that's a simple thing and it takes a long time already too.
My view is that whenever code is rewritten, we should try to bring it up to par with coding style. But if people disagree that '> >' is something to avoid, I will shut up about it :)
Forgot to include with initial PR, it's something like 12% faster in connect postprocess.
push -f'ed a version where txData type template extra whitespace is removed as per @sipa's request.
Travis might fail because I force pushed twice and there's a known race condition on that.
utACK ec81881
utACK ec81881b86b9680fcdcc42fd3ba31f04b8d09714