ilya-biryukov added a comment.
> Seems like this optimization is not worth (yet?). As soon as we get more
> proximity paths (and hence more OR iterator children) that might make sense.
WDYT about storing **all** the elements with the minimal doc-id outside the
heap? I.e. we can pop **all** elem
kbobyrev updated this revision to Diff 165471.
kbobyrev marked 3 inline comments as done.
kbobyrev edited the summary of this revision.
kbobyrev added a comment.
Fixed the bug with incorrect assumption of `Children` being sorted (instead of
being a heap). This caused a huge overhead (~30%). When
ilya-biryukov added inline comments.
Comment at: clang-tools-extra/clangd/index/dex/Iterator.cpp:128
+// Return LHS > RHS.
+auto Compare = [](const std::unique_ptr &LHS,
ilya-biryukov wrote:
> NIT: use triple-slash comments.
> NIT: LHS > RHS seems to be exactly
ilya-biryukov added inline comments.
Comment at: clang-tools-extra/clangd/index/dex/Iterator.cpp:128
+// Return LHS > RHS.
+auto Compare = [](const std::unique_ptr &LHS,
NIT: use triple-slash comments.
NIT: LHS > RHS seems to be exactly what's defined by this f
kbobyrev created this revision.
kbobyrev added reviewers: ioeric, sammccall, ilya-biryukov.
kbobyrev added a project: clang-tools-extra.
Herald added subscribers: kadircet, arphaman, jkorous, MaskRay.
Use min-heap invariant for OR iterator's children. This helps to avoid
iterating through all chi
