the indexer emits glean facts directly

Summary:
In this diff, the swift indexer not only emits SCIP index, but also some glean facts for the new swift.angle schema.

For now the facts emitted are for predicate SymbolLanguage. We'll add more in future diffs.
This is based on the info obtained from demangling the USR.

Note: for now, these new facts are ignored by the indexing workflow

Rabbit hole:

Demangling the USRs was using std::regex which has really bad perf. The indexing time for my test went from 5 seconds to 5 minutes. I rewrote the code to do some basic parsing instead. Regression is now only 13%.

While I was at it, I added support for clang structs and enum USRs.

I also added a unit test for demangling.

Differential Revision: D79262434

fbshipit-source-id: e3048134756478b663513d85b9df4dd80809a6e9

Author: Catherine Gasnier

glean/lang/swift/Glean/Indexer/Swift.hs

@@ -15,7 +15,9 @@ import Glean.Indexer
 import Glean.Indexer.External
 import Glean.Indexer.SCIP ( derive )
 import qualified Glean.SCIP.Driver as SCIP
-import System.FilePath (takeDirectory)
+import System.FilePath ( (</>), takeDirectory)
+import Glean.LocalOrRemote ( serializeInventory )
+import qualified Data.ByteString as BS
 
 data Swift = Swift
   { scipGen :: FilePath
@@ -40,6 +42,9 @@ indexer = Indexer {
   indexerDescription = "Index Swift code",
   indexerOptParser = options,
   indexerRun = \Swift{..} backend repo IndexerParams{..} -> do
+    let tmpDir        = indexerOutput
+        inventoryFile = tmpDir </> "inventory.data"
+    serializeInventory backend repo >>= BS.writeFile inventoryFile
     val <- SCIP.runIndexer SCIP.ScipIndexerParams {
         scipBinary = scipGen,
         scipArgs = \outFile ->
@@ -48,6 +53,7 @@ indexer = Indexer {
             , "--target", target
             , "--swift-only"
             , "--output-type", "scip"
+            , "--inventory", inventoryFile
             , "--build-indexer" ],
         scipRoot = indexerRoot,
         scipWritesLocal = False,