fix duplicated search result

Summary:
Bug mentioned [here](https://fb.workplace.com/groups/factsaboutcode/permalink/2412053725810528/)

Symbols are duplicated in search result. I suspect we return descriptions from both `fbsource.fbcode.python` and `fbsource.fbcode.python.incr` (I got that from the repo_hash'es in the result, which match instances of these dbs).

Sledge-hammer fix, keep only one (arbitrary but  deterministic) description per key (symbol_id/repo) in [search result](https://www.internalfb.com/code/fbsource/[e81c7e9fcf17]/fbcode/glean/glass/if/glass.thrift?lines=576).

There may be better ways to fix this.  Query only one db? or return the most recent symbol

```
> buck2 run  fbcode//glean/glass/facebook/cli:glass   -- search  -d --kinds Class
```
Before

```
fbsource/fbcode/neteng/drainer/services/driver/py/core/lib/option.py:13:1-52:45:fbsource/py/fbcode/neteng.drainer.services.driver.py.core.lib.option.EnumAction neteng.drainer.services.driver.py.core.lib.option.EnumAction
fbsource/fbcode/ricardo/src/main.py:1309:1-1364:44:fbsource/py/fbcode/ricardo.src.main.EnumAction ricardo.src.main.EnumAction
fbsource/fbcode/hphp/facebook/tools/automator/__init__.py:81:1-111:31:fbsource/py/fbcode/automator.EnumAction automator.EnumAction
fbsource/fbcode/neteng/drainer/services/driver/py/core/lib/option.py:13:1-52:45:fbsource/py/fbcode/neteng.drainer.services.driver.py.core.lib.option.EnumAction neteng.drainer.services.driver.py.core.lib.option.EnumAction
fbsource/fbcode/ricardo/src/main.py:1309:1-1364:44:fbsource/py/fbcode/ricardo.src.main.EnumAction ricardo.src.main.EnumAction
fbsource/fbcode/hphp/facebook/tools/automator/__init__.py:81:1-111:31:fbsource/py/fbcode/automator.EnumAction automator.EnumAction
```

After

```
fbsource/fbcode/ricardo/src/main.py:1309:1-1364:44:fbsource/py/fbcode/ricardo.src.main.EnumAction ricardo.src.main.EnumAction
fbsource/fbcode/hphp/facebook/tools/automator/__init__.py:81:1-111:31:fbsource/py/fbcode/automator.EnumAction automator.EnumAction
fbsource/fbcode/neteng/drainer/services/driver/py/core/lib/option.py:13:1-52:45:fbsource/py/fbcode/neteng.drainer.services.driver.py.core.lib.option.EnumAction neteng.drainer.services.driver.py.core.lib.option.EnumAction
```

Reviewed By: pepeiborra

Differential Revision: D67762205

fbshipit-source-id: e284f0f7e9bc57a9d9e4632cae18efe0adfd22fe

glean/glass/Glean/Glass/Handler/Symbols.hs

@@ -82,7 +82,7 @@ import Glean.Glass.NameSearch (
     SearchQuery(..), SingleSymbol, FeelingLuckyResult(..),
     QueryExpr(..), RepoSearchResult, SymbolSearchData(..),
     toSearchResult, ToSearchResult(..), AngleSearch(..), srEntity,
-    buildLuckyContainerQuery, buildSearchQuery
+    buildLuckyContainerQuery, buildSearchQuery, dedupSearchResult
   )
 import Glean.Glass.XRefs ( GenXRef(..) )
 import Glean.Glass.Search as Search
@@ -408,13 +408,14 @@ joinSearchResults
 joinSearchResults mlimit terse sorted xs = SymbolSearchResult syms $
     if terse then [] else catMaybes descs
   where
+    uniqXs = dedupSearchResult <$> xs
     (syms,descs) = unzip $ nubOrd $ case (mlimit, sorted) of
       (Nothing, _) -> flattened
       (Just n, False) -> take n flattened
       -- codehub/aka "sorted" mode grouping, ranking and sampling
-      (Just n, True) -> takeFairN n (concatMap sortResults xs)
+      (Just n, True) -> takeFairN n (concatMap sortResults uniqXs)
 
-    flattened = concat xs
+    flattened = concat uniqXs
 
 --
 -- DFS to first singleton result.

glean/glass/Glean/Glass/NameSearch.hs

@@ -17,6 +17,7 @@ module Glean.Glass.NameSearch
   , RepoSearchResult
   , FeelingLuckyResult(..)
   , SingleSymbol
+  , dedupSearchResult
 
   -- * Search
   -- ** Search flags
@@ -56,6 +57,7 @@ import Glean.Glass.Utils (splitOnAny, QueryType )
 import qualified Glean.Schema.CodemarkupTypes.Types as Code
 import qualified Glean.Schema.CodemarkupSearch.Types as CodeSearch
 import qualified Glean.Schema.Code.Types as Code
+import qualified Data.HashMap.Strict as Map
 
 --
 -- Finding entities by name search
@@ -651,6 +653,18 @@ instance ToSearchResult CodeSearch.SearchByScope where
 -- | Type of processed search results from a single scm repo
 type RepoSearchResult = [SingleSymbol]
 
+-- | Ensure all SymbolResults in a (repo-wide) search result are unique.
+--
+--   We can have multiple descriptions when querying dbs
+--   having the same content (for instance, incremental and full
+--   DBs). In that case, descriptions differ only in the repo_hash
+--   field.
+--   dedupSearchResult picks one abitrarily. TODO pick the most
+--   recent revision one, and don't discard a description if it
+--   differs by more than the repo_hash field
+dedupSearchResult :: RepoSearchResult -> RepoSearchResult
+dedupSearchResult results = Map.toList $ Map.fromListWith max results
+
 -- An un-concatenated set of query results to search for unique hits in
 -- within one scm repo, across dbs, across queries, a set of result symbols.
 newtype FeelingLuckyResult = FeelingLuckyResult [[RepoSearchResult]]