[sqlite] sql::Recovery working under USE_SYSTEM_SQLITE.

third_party/sqlite/BUILD.gn

Index: third_party/sqlite/BUILD.gn
diff --git a/third_party/sqlite/BUILD.gn b/third_party/sqlite/BUILD.gn
index 8400837db8c2d34243e75288573eb8b5f5a81ac4..c6fb95d2486846ed3aeea2f8a8fb8d707a48dada 100644
--- a/third_party/sqlite/BUILD.gn
+++ b/third_party/sqlite/BUILD.gn
@@ -223,6 +223,9 @@ if (use_system_sqlite) {
 
   source_set("sqlite") {
     public_configs = [ ":sqlite_config" ]
+    deps = [
+      ":sqlite_recover",
+    ]

[Scott Hess - ex-Googler, 2016/03/02 01:24:44]

Since this is a custom-to-us module, we should be able to safely include it for
_any_ use_system_sqlite case.  sqlite_regexp is more debatable (though I suspect
it would also work fine for any use_system_sqlite case, too).

     if (is_ios) {
       deps = [
         ":sqlite_regexp",
@@ -230,6 +233,15 @@ if (use_system_sqlite) {
     }
   }
 
+  source_set("sqlite_recover") {
+    sources = [
+      # TODO(shess): Move out of the SQLite source tree, perhaps to ext/.
+      "src/src/recover.c",
+      "src/src/recover.h",
+      "src/src/recover_varint.c",
+    ]
+  }

[Scott Hess - ex-Googler, 2016/03/02 01:24:44]

This module could be pulled outside of the use_system_sqlite case entirely, and
the sql/BUILD.gn could depend on sqlite:sqlite_recover for either case.  It
should work either way, since the module uses no globals.  The main reason I
went with this split is simply being conservative.

Of course, another solution would be to pull the entire module into sql/
somewhere.  That would make the in-SQLite testing harder, though it could be
rewritten as a unit test.  I'd like to hold off on that kind of move until I try
upstreaming things.  If SQLite accepts it, then lots of problems are solved.

+
   if (is_ios) {
     source_set("sqlite_regexp") {
       defines = [