Please, next time attach intermediate diff (i.e. diff between two versions) or
inline each hunk as answer to comment. It is quite complicated to review
full patch (especially when it comes to patches containing hundreds of lines)
each time.

https://travis-ci.org/tarantool/tarantool/jobs/436232731

box/sql.test.lua fails on Travis. Please, fix it.

Now it is good on travis:
https://travis-ci.org/tarantool/tarantool/builds/439704198

Diff between two last versions:


diff --git a/src/box/sql/analyze.c b/src/box/sql/analyze.c
index 95c516a..a886d8a 100644
--- a/src/box/sql/analyze.c
+++ b/src/box/sql/analyze.c
@@ -797,13 +797,13 @@ vdbe_emit_analyze_space(struct Parse *parse, struct space *space)
     assert(space->index_count != 0);
     struct Vdbe *v = sqlite3GetVdbe(parse);
     assert(v != NULL);
-    const char *tab_name = space_name(space);
+    MAYBE_UNUSED const char *tab_name = space_name(space);
     sqlite3VdbeAddOp4(v, OP_IteratorOpen, tab_cursor, 0, 0, (void *) space,
               P4_SPACEPTR);
     sqlite3VdbeAddOp2(v, OP_Integer, space->def->id, space_id_reg);
     for (uint32_t j = 0; j < space->index_count; ++j) {
         struct index *idx = space->index[j];
-        const char *idx_name;
+        MAYBE_UNUSED const char *idx_name;
         /*
          * Primary indexes feature automatically generated
          * names. Thus, for the sake of clarity, use

Actually, this diff doesn’t look like fix of that failed test.
I guess it is simply flaky, so this time you get lucky and it is passed.
Did you checked that test-trace from Travis fails on fresh 2.0 as well?
Did you manage to understand the reason of failure? Otherwise there is
no guarantee that you patch is innocent in this situation.
Without any investigation I can give my approval on this patch.