Home - Waterfall Grid T-Grid Console Builders Recent Builds Buildslaves Changesources - JSON API - About

Console View

Tags: dev stable
Legend:   Passed Failed Warnings Failed Again Running Exception Offline No data

dev stable
Olly Betts
Avoid dividing by 0 in DPHWeight
This could happen with a sharded database when one shard is empty.

The bug caused dphweight3 to fail when run under UBSan with either of
the new test harness backends multi_glass_remoteprog_glass or
Olly Betts
Put AC_CONFIG_FILES in sorted order
Olly Betts
Use BoolOrContext instead of OrContext more
In any unweighted context BoolOrContext is a better choice as it will
build a single BoolOrPostList object optimised for the unweighted case
instead of a tree of general-purpose OrPostList objects.
Olly Betts
Fix ODR violation
Two different source files linked into apitest were each defining a
different `struct test`.  Wrap each in an anonymous namespace to
localise it to the file it is defined and used in.
Olly Betts
Optimise OP_AND_MAYBE better
Combine its LHS with other connected and-like subqueries, and gather
up and hoist the optional subqueries and apply them together above the
combined and-like subqueries and any hoisted positional filters.
Olly Betts
Eliminate use of VLA when normalising feature vecs
If we process each feature in turn we only need to deal with one
maximum value at a time.
Olly Betts
Fix match stats when collapsing with remote shards
When searching with collapsing over multiple shards and at least some
are remote, then the bounds and estimate for the number of matching
documents weren't being handled correctly in some cases.

The fix for this requires us to request check_at_least matches from
each shard in this case.

Bugs uncovered by Tanmay Sachan's upcoming change to enable running
generated testcases for sharded databases with remote or mixed
remote+local shards, which will serve as a regression test for this.
Olly Betts
Eliminate extra parentheses around return values
Olly Betts
Update to reflect current GCC >= 4.8 requirement
Olly Betts
Change $log{} to return an error message
This is upwardly incompatible in cases where $log fails, but means
it's now possible to report logging errors, and one can wrap the
`$log{...}` in `$if{` and `}` to ignore errors which also works
with older versions.
Olly Betts
Avoid warnings with clang
Olly Betts
Say "since Omega 1.5.0" not "since Xapian 1.5.0"
Olly Betts
Sync changes from core configure.ac to letor
Olly Betts
Optimise OP_AND_NOT better
Combine its LHS with other connected and-like subqueries, and gather
up and hoist the negated subqueries and apply them together above the
combined and-like subqueries, just below any positional filters.
Make generated testcases run on remote multi db backends
Olly Betts
Return revision 0 for a Database with no shards
This seems better than throwing InvalidOperationError.
Olly Betts
Treat all BoolWeight queries as scaled by 0
We can optimise better if we know the query is unweighted when
converting it to a PostList tree.
Olly Betts
Optimise away MatchAll when part of and-like query
`foo (NOT bar)` is now turned into the same PostList tree as
`foo NOT bar`, whereas previously the former included an extra all-docs

To enable this to work cleanly the QueryParser now scales MatchAll
weights by 0 in a pure NOT, which seems better in general anyway.
Olly Betts
Handle the RHS of OP_AND_NOT as BoolOrContext
It's unweighted so doesn't need to use OrContext, and BoolOrContext
is a better choice as it will build a single BoolOrPostList object
optimised for the unweighted case instead of a tree of general-purpose
OrPostList objects.
Olly Betts
Use std::find() to search for an Action
Olly Betts
Remove unnecessary "virtual"
HtmlParser::parse() is never overridden.
Olly Betts
Sync common parts of INSTALL between omega & bindings
Olly Betts
Suppress -Wduplicated-branches warning in API headers
Olly Betts
Stop specifying -Winit-self for GCC
It's implied by -Wall -W since GCC 4.8, which is the oldest we
Olly Betts
Improve rounding of MSet::get_matches_estimated()
The conditional check for rounding up was wrong, which meant we would
almost always round down.
Olly Betts
Always catch error objects by reference
All these cases are in fact empty objects that are only identified
by type, but by avoiding doing this we can consider enabling
GCC's -Wcatch-value=2 where available.
Olly Betts
Implement scriptindex 'gap' action
This allows leaving a gap in the term positions between fields to
stop phrases and positional operators from matches across fields.
Olly Betts
Don't check __GNUC__ in visibility.h
The configure check gating defining XAPIAN_ENABLE_VISIBILITY checks
that the visibility attributes work so there's no need to disallow
this for a hypothetical compiler which supports them but doesn't define
Olly Betts
Fix indexing of positions in CJK corner case
If indexing CJK ngrams, a stemmer is in use, and the stemming mode
isn't STEM_SOME then we'd add bigrams with positional information
(whereas only unigrams are meant to have positional information).

We don't provide any stemming algorithms for CJK languages, though
someone might enable this situation if indexing a mixture of CJK and
text in a language there's a stemmer for.  Or they might have a
custom stemmer.

The typo which caused this is also present in 1.4.x, but there it's
harmless because the CJK n-gram code is different.
Olly Betts
quest: Report bounds and estimate of number of matches
Olly Betts
Remove outdated comment about retrying -Wconversion
It was toned down and added to -Wall in GCC 4.6 so already enabled
by all GCC versions we support anyway.
Olly Betts
Eliminate some C-style casts
Olly Betts
The autoconf manual marks it as "obsolescent", and it seems clear
that nobody's relying on it as we're missing the "'AC_LIBOBJ'
replacement for 'memcmp'" which it would try to require if needed.
Olly Betts
Mark some internal classes as final
Compiling with GCC and -Wsuggest-final-types suggests this will allow
some method calls to be devirtualised.
Olly Betts
Fix documentation typo
Olly Betts
Fix bug introduced by previous commit
pls needs to still be valid while we build the positional filters.
Olly Betts
Fix corner-case bug with optimising away MatchAll
If an AND-like query contains an optimisable MatchAll and another
subquery which matches nothing and produces a NULL postlist then
we would combine to give a MatchAll query instead of a NULL postlist.
Olly Betts
Point to actual docs for MALLOC_PERTURB_
Olly Betts
Comment tweak
It's libstdc++ that's the key thing here.  You can use libstdc++ with
clang as well as GCC.
Olly Betts
Improve report when revision not available
Specially handle the cases of a DB with multiple shards and a
backend which doesn't support get_revision().