diff options
| author | Ilya Biryukov <ibiryukov@google.com> | 2019-03-15 14:00:49 +0000 |
|---|---|---|
| committer | Ilya Biryukov <ibiryukov@google.com> | 2019-03-15 14:00:49 +0000 |
| commit | 373bee85c2b31332d49db329cdbfb5f7c7342a04 (patch) | |
| tree | 2a67c227a420cee6ebd813fc1b797b0e657a1940 /llvm/lib/IR/ModuleSummaryIndex.cpp | |
| parent | 339daae806b208cad060ce93f63e40f24a311812 (diff) | |
| download | llvm-373bee85c2b31332d49db329cdbfb5f7c7342a04.zip llvm-373bee85c2b31332d49db329cdbfb5f7c7342a04.tar.gz llvm-373bee85c2b31332d49db329cdbfb5f7c7342a04.tar.bz2 | |
[clangd] Tune the fuzzy-matching algorithm
Summary:
To reduce the gap between prefix and initialism matches.
The motivation is producing better scoring in one particular example,
but the change does not seem to cause large regressions in other cases.
The examples is matching 'up' against 'unique_ptr' and 'upper_bound'.
Before the change, we had:
- "[u]nique_[p]tr" with a score of 0.3,
- "[up]per_bound" with a score of 1.0.
A 3x difference meant that symbol quality signals were almost always ignored
and 'upper_bound' was always ranked higher.
However, intuitively, the match scores should be very close for the two.
After the change we have the following scores:
- "[u]nique_[p]tr" with a score of 0.75,
- "[up]per_bound" with a score of 1.0.
Reviewers: ioeric
Reviewed By: ioeric
Subscribers: MaskRay, jkorous, arphaman, kadircet, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D59300
llvm-svn: 356261
Diffstat (limited to 'llvm/lib/IR/ModuleSummaryIndex.cpp')
0 files changed, 0 insertions, 0 deletions