Re: [PATCH v3 2/4] tests/qtest/migration: Add infrastructure to skip tests on older QEMUs

[Fabiano Rosas <farosas@suse.de>]

Peter Xu <peterx@redhat.com> writes:

> On Fri, Jan 05, 2024 at 03:04:47PM -0300, Fabiano Rosas wrote:
>> We can run the migration tests with two different QEMU binaries to
>> test migration compatibility between QEMU versions. This means we'll
>> be running the tests with an older QEMU in either source or
>> destination.
>> 
>> We need to avoid trying to test functionality that is unknown to the
>> older QEMU. This could mean new features, bug fixes, error message
>> changes, QEMU command line changes, migration API changes, etc.
>> 
>> Add a 'since' argument to the tests that inform when the functionality
>> that is being test has been added to QEMU so we can skip the test on
>> older versions.
>> 
>> Also add a version comparison function so we can adapt test code
>> depending on the QEMU binary version being used.
>> 
>> Signed-off-by: Fabiano Rosas <farosas@suse.de>
>> ---
>>  tests/qtest/migration-helpers.h |  1 +
>>  tests/qtest/migration-helpers.c | 48 +++++++++++++++++++++++++++++++++
>>  tests/qtest/migration-test.c    | 29 ++++++++++++++++++++
>>  3 files changed, 78 insertions(+)
>> 
>> diff --git a/tests/qtest/migration-helpers.h b/tests/qtest/migration-helpers.h
>> index e31dc85cc7..7b4f8e851e 100644
>> --- a/tests/qtest/migration-helpers.h
>> +++ b/tests/qtest/migration-helpers.h
>> @@ -47,4 +47,5 @@ char *find_common_machine_version(const char *mtype, const char *var1,
>>                                    const char *var2);
>>  char *resolve_machine_version(const char *alias, const char *var1,
>>                                const char *var2);
>> +int migration_vercmp(QTestState *who, const char *tgt_version);
>>  #endif /* MIGRATION_HELPERS_H */
>> diff --git a/tests/qtest/migration-helpers.c b/tests/qtest/migration-helpers.c
>> index 24fb7b3525..bc4fd93e8c 100644
>> --- a/tests/qtest/migration-helpers.c
>> +++ b/tests/qtest/migration-helpers.c
>> @@ -292,3 +292,51 @@ char *resolve_machine_version(const char *alias, const char *var1,
>>  
>>      return find_common_machine_version(machine_name, var1, var2);
>>  }
>> +
>> +int migration_vercmp(QTestState *who, const char *tgt_version)
>> +{
>> +    g_autofree char *version = g_strdup(tgt_version);
>> +    int major = 0, minor = 0, micro = 0;
>> +    int tgt_major = 0, tgt_minor = 0, tgt_micro = 0;
>> +    const char *delim = ".";
>> +    char *tok;
>> +
>> +    qtest_query_version(who, &major, &minor, &micro);
>> +
>> +    tok = strtok(version, delim);
>> +    assert(tok);
>> +    tgt_major = atoi(tok);
>> +    assert(tgt_major);
>
> I'd consider dropping this.  But I don't think "since: 0.8" is valid. :)
> More like a nitpick.
>
>> +
>> +    if (major > tgt_major) {
>> +        return -1;
>
> This means the QEMU version is newer, the function will return negative.
> Is this what we want?  It seems it's inverted.

The return "points" to which once is the more recent:

QEMU version | since: version
-1           0         1

> In all cases, document this function with retval would be helpful too.
>

Ok.

>> +    }
>> +    if (major < tgt_major) {
>> +        return 1;
>> +    }
>
> Instead of all these, I'm wondering whether we can allow "since" to be an
> array of integers, like [8, 2, 0].  Would that be much easier?

I don't see why push the complexity towards the person writing the
tests. The string is much more natural to specify.

>> +
>> +    tok = strtok(NULL, delim);
>> +    assert(tok);
>> +    tgt_minor = atoi(tok);
>> +
>> +    if (minor > tgt_minor) {
>> +        return -1;
>> +    }
>> +    if (minor < tgt_minor) {
>> +        return 1;
>> +    }
>> +
>> +    tok = strtok(NULL, delim);
>> +    if (tok) {
>> +        tgt_micro = atoi(tok);
>> +    }
>> +
>> +    if (micro > tgt_micro) {
>> +        return -1;
>> +    }
>> +    if (micro < tgt_micro) {
>> +        return 1;
>> +    }
>> +
>> +    return 0;
>> +}
>> diff --git a/tests/qtest/migration-test.c b/tests/qtest/migration-test.c
>> index d520c587f7..001470238b 100644
>> --- a/tests/qtest/migration-test.c
>> +++ b/tests/qtest/migration-test.c
>> @@ -637,6 +637,12 @@ typedef struct {
>>      bool use_dirty_ring;
>>      const char *opts_source;
>>      const char *opts_target;
>> +    /*
>> +     * If a test checks against new functionality that might not be
>> +     * present in older QEMUs this needs to be set so we can skip
>> +     * running it when doing compatibility testing.
>> +     */
>> +    const char *since;
>>  } MigrateStart;
>>  
>>  /*
>> @@ -850,6 +856,17 @@ static int test_migrate_start(QTestState **from, QTestState **to,
>>          qtest_qmp_set_event_callback(*from,
>>                                       migrate_watch_for_stop,
>>                                       &got_src_stop);
>> +
>> +        if (args->since && migration_vercmp(*from, args->since) < 0) {
>> +            g_autofree char *msg = NULL;
>> +
>> +            msg = g_strdup_printf("Test requires at least QEMU version %s",
>> +                                  args->since);
>> +            g_test_skip(msg);
>> +            qtest_quit(*from);
>> +
>> +            return -1;
>> +        }
>>      }
>>  
>>      cmd_target = g_strdup_printf("-accel kvm%s -accel tcg "
>> @@ -872,6 +889,18 @@ static int test_migrate_start(QTestState **from, QTestState **to,
>>                                   migrate_watch_for_resume,
>>                                   &got_dst_resume);
>>  
>> +    if (args->since && migration_vercmp(*to, args->since) < 0) {
>> +        g_autofree char *msg = NULL;
>> +
>> +        msg = g_strdup_printf("Test requires at least QEMU version %s",
>> +                              args->since);
>> +        g_test_skip(msg);
>> +        qtest_quit(*to);
>> +        qtest_quit(*from);
>> +
>> +        return -1;
>
> Nitpick: you can do both check here, then avoid duplicating some code, and
> always free from+to.
>

Hm.. I think I had a good reason for doing it this way. I don't remember
now. I'll double check and apply your suggestion if it works.

>> +    }
>> +
>>      /*
>>       * Remove shmem file immediately to avoid memory leak in test failed case.
>>       * It's valid because QEMU has already opened this file
>> -- 
>> 2.35.3
>>