[tarantool-patches] Re: [PATCH 2/2] sql: take sql field in index_opts_cmp

[Kirill Yukhin]

Hello Nikita,

Thanks for review! My answers inlined, updated patch in the bottom of
the messaga, branch force-pushed.

On 21 авг 13:26, n.pettik wrote:

> > diff --git a/src/box/index_def.h b/src/box/index_def.h
> > index 48a7820..8960e3d 100644
> > --- a/src/box/index_def.h
> > +++ b/src/box/index_def.h
> > @@ -212,6 +212,14 @@ index_opts_cmp(const struct index_opts *o1, const struct index_opts *o2)
> > 		return o1->run_size_ratio < o2->run_size_ratio ? -1 : 1;
> > 	if (o1->bloom_fpr != o2->bloom_fpr)
> > 		return o1->bloom_fpr < o2->bloom_fpr ? -1 : 1;
> > +
> 
> Nitpicking: don’t put extra new line: it breaks whole function look.

Removed.

> > +	if ((o1->sql == NULL) != (o2->sql == NULL))
> > +		return 1;
> > +	if ((o1->sql != NULL) && (o2->sql != NULL)) {
> 
> Nitpicking: after previous check o1->sql and o2->sql are both NULL or not.
> So I guess only one check is enough: if (o1->sql != NULL).

Good catch. Fixed.
 
> > +		int rc = strcmp(o1->sql, o2->sql);
> > +		if (rc != 0)
> > +			return rc;
> 
> Why not simply return strcmp()?

Done.

> > +	}
> > 	return 0;
> > }
> > 
> > diff --git a/test/sql/gh-3613-idx-alter-update-2.test.lua b/test/sql/gh-3613-idx-alter-update-2.test.lua
> > new file mode 100644
> > index 0000000..e2beb7a
> > --- /dev/null
> > +++ b/test/sql/gh-3613-idx-alter-update-2.test.lua
> 
> Can you merge this test file with previous one?
> I strongly dislike the fact that such simple tests are put in
> separate files. Even though we already have sql-tap/alter.test.lua.

No. There's no reason for economy in number of files. There're tons
of pros when each and every regression test lives in separate file, or
even files. I'll update SOP stating that.

--
Regards, Kirill Yukhin

commit a2c7fb02f330819d31591273dbf9ec9e60a5f5df
Author: Kirill Yukhin <kyukhin at tarantool.org>
Date:   Mon Aug 20 19:38:06 2018 +0300

    sql: take sql field in index_opts_cmp
    
    After gh-3613 is imlemented a hidden bug was uncovered:
    during, e.g. recovery opts.sql field didn't take into account
    marking corresponding xlog entry useless. Fix that by comparing
    mentioned field of new and old entries.
    
    Follow up of #3613
---
 src/box/index_def.h                          |  4 ++++
 test/sql/gh-3613-idx-alter-update-2.result   | 28 ++++++++++++++++++++++++++++
 test/sql/gh-3613-idx-alter-update-2.test.lua | 16 ++++++++++++++++
 3 files changed, 48 insertions(+)

diff --git a/src/box/index_def.h b/src/box/index_def.h
index 48a7820..273b8cb 100644
--- a/src/box/index_def.h
+++ b/src/box/index_def.h
@@ -212,6 +212,10 @@ index_opts_cmp(const struct index_opts *o1, const struct index_opts *o2)
 		return o1->run_size_ratio < o2->run_size_ratio ? -1 : 1;
 	if (o1->bloom_fpr != o2->bloom_fpr)
 		return o1->bloom_fpr < o2->bloom_fpr ? -1 : 1;
+	if ((o1->sql == NULL) != (o2->sql == NULL))
+		return 1;
+	if (o1->sql != NULL)
+		return strcmp(o1->sql, o2->sql);
 	return 0;
 }
 
diff --git a/test/sql/gh-3613-idx-alter-update-2.result b/test/sql/gh-3613-idx-alter-update-2.result
new file mode 100644
index 0000000..234336c
--- /dev/null
+++ b/test/sql/gh-3613-idx-alter-update-2.result
@@ -0,0 +1,28 @@
+test_run = require('test_run').new()
+---
+...
+engine = test_run:get_cfg('engine')
+---
+...
+box.sql.execute('pragma sql_default_engine=\''..engine..'\'')
+---
+...
+box.sql.execute('CREATE TABLE t (s1 INT PRIMARY KEY)')
+---
+...
+box.sql.execute('CREATE INDEX i ON t (s1)')
+---
+...
+box.sql.execute('ALTER TABLE t RENAME TO j3')
+---
+...
+-- After gh-3613 fix, bug in cmp_def was discovered.
+-- Comparison didn't take .opts.sql into account.
+test_run:cmd('restart server default')
+box.sql.execute('DROP INDEX i ON j3')
+---
+...
+-- Cleanup
+box.sql.execute('DROP TABLE j3')
+---
+...
diff --git a/test/sql/gh-3613-idx-alter-update-2.test.lua b/test/sql/gh-3613-idx-alter-update-2.test.lua
new file mode 100644
index 0000000..e2beb7a
--- /dev/null
+++ b/test/sql/gh-3613-idx-alter-update-2.test.lua
@@ -0,0 +1,16 @@
+test_run = require('test_run').new()
+engine = test_run:get_cfg('engine')
+box.sql.execute('pragma sql_default_engine=\''..engine..'\'')
+
+box.sql.execute('CREATE TABLE t (s1 INT PRIMARY KEY)')
+box.sql.execute('CREATE INDEX i ON t (s1)')
+box.sql.execute('ALTER TABLE t RENAME TO j3')
+
+-- After gh-3613 fix, bug in cmp_def was discovered.
+-- Comparison didn't take .opts.sql into account.
+test_run:cmd('restart server default')
+
+box.sql.execute('DROP INDEX i ON j3')
+
+-- Cleanup
+box.sql.execute('DROP TABLE j3')