dejagnu parse: drop unique ddl

While unique() assertions in the table ddl are good for data hygiene,
they also cost considerable storage.  Since the pipeline manages
analysis engine lifecycles carefully and non-concurrently, we don't
really need them to assert desirable uniqueness properties in the
tables.  Instead, some ".eqp on" inspired indexes can do the job
of accelerating queries, subject to easier further tuning.

diff --git a/analysis/i-dejagnu-parser b/analysis/i-dejagnu-parser
index 41ec0231fc419341c27e1ecc2421205770cad61d..58b1e048400f338dc27ffe4006c4cc943df37ae4 100755 (executable)
--- a/analysis/i-dejagnu-parser
+++ b/analysis/i-dejagnu-parser
@@ -36,8 +36,16 @@ def parse_dejagnu_log_sum(db, commit, trid, sumfile, logfile):
     logging.debug ("processing sumfile %s logfile %s tool %s",
                    sumfile.path, logfile.path if logfile else None, tool)
 
+    # Drop preexisting rows in dejagnu_testsuite and dejagnu_testcase,
+    # so we can create new rows without risk of duplication, even without
+    # unique indexes.  We save space by enforcing this constraint manually.
+    # This will drop foreign-key derived tables too, so other i-* engines
+    # need to run, which normally is automatic.
+    db.execute("delete from dejagnu_testsuite where tr=? and sumfile=?", (trid, sumfile.path))
+    db.commit()
+
     # create dejagnu testsuite record
-    db.execute("insert or ignore into dejagnu_testsuite (tr, logfile, sumfile) values (?, ?, ?);",
+    db.execute("insert into dejagnu_testsuite (tr, logfile, sumfile) values (?, ?, ?);",
                (trid, logfile.path if logfile else None, sumfile.path))
     db.commit()
     # and fetch the testsuite record id right away
@@ -163,7 +171,7 @@ def parse_dejagnu_log_sum(db, commit, trid, sumfile, logfile):
                 logline_number += 1
 
             # time to add the dejagnu_testcase record
-            db.execute("insert or ignore into dejagnu_testcase (testsuite, expfile, subtest, result, logcursor, logcursor2, sumcursor)"+
+            db.execute("insert into dejagnu_testcase (testsuite, expfile, subtest, result, logcursor, logcursor2, sumcursor)"+
                        "values (?,?,?,?,?,?,?);",
                        (tstid, expfileid, subtestid, resultid,
                         logline_number_last if loglinefound else None,
@@ -213,8 +221,9 @@ CREATE TABLE if not exists dejagnu_testsuite (
        tr integer not null,     /* testrun ref */
        logfile string,          /* e.g. testsuite/systemtap.log */
        sumfile string not null, /* e.g. testsuite/systemtap.sum */
-       unique (tr, sumfile),
+       /* unique (tr, sumfile), */ /* one sumfile per testsuite */
        foreign key (tr) references testrun(id) on delete cascade);
+CREATE INDEX if not exists dejagnu_ts_tr on dejagnu_testsuite(tr);
 CREATE TABLE if not exists dejagnu_string (   /* string interning table */
        id integer primary key,
        name string unique);
@@ -226,10 +235,11 @@ CREATE TABLE if not exists dejagnu_testcase (
        logcursor integer, /* line# in logfile */
        logcursor2 integer, /* line# in logfile */
        sumcursor integer, /* line# in logfile */
-       unique (testsuite, sumcursor), /* one test case record per sum file line; dupe subtests sometimes appear */
+       /* unique (testsuite, sumcursor), */ /* one test case record per sum file line; dupe subtests sometimes appear */
        foreign key (expfile) references dejagnu_string(id) on delete cascade,
        foreign key (subtest) references dejagnu_string(id) on delete cascade,
        foreign key (testsuite) references dejagnu_testsuite(id) on delete cascade);
+CREATE INDEX if not exists dejagnu_testcase_te on dejagnu_testcase (testsuite, expfile);
 CREATE VIEW if not exists dejagnu_testcase_v (testsuite, expfile, subtest, result, logcursor, logcursor2, sumcursor) as
        select t.testsuite, es.name, ss.name, rs.name, t.logcursor, t.logcursor2, t.sumcursor
        from dejagnu_testcase t, dejagnu_string es, dejagnu_string ss, dejagnu_string rs