[v2,3/3] test: support trace-autotest when enable trace

  There are a bug[1] when exit application while enable tracing, this
bug has not been discovered for a long time, to quickly detect such
bugs, this patch was introduced.

This patch adds a testcase with trace enabling, it also depends on
patch[2] because it has a long file-prefix.

[1] eal: fix segment fault when exit trace
[2] eal: fix trace init fail with long file-prefix

Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
---
 app/test/meson.build | 11 +++++++++++
 1 file changed, 11 insertions(+)
  

On Tue, Jun 14, 2022 at 8:06 AM Chengwen Feng <fengchengwen@huawei.com> wrote:
>
> There are a bug[1] when exit application while enable tracing, this
> bug has not been discovered for a long time, to quickly detect such
> bugs, this patch was introduced.
>
> This patch adds a testcase with trace enabling, it also depends on
> patch[2] because it has a long file-prefix.
>
> [1] eal: fix segment fault when exit trace
> [2] eal: fix trace init fail with long file-prefix

This commitlog feels more like a cover letter.

Please describe what the impact of the patch is, like mention that the
trace_autotest unit test is being called twice, once with traces
disabled, and once with traces enabled.
And that the traces file is being written in a directory part of the
build directory.


>
> Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
> ---
>  app/test/meson.build | 11 +++++++++++
>  1 file changed, 11 insertions(+)
>
> diff --git a/app/test/meson.build b/app/test/meson.build
> index 7fe261cae8..eb37aa632a 100644
> --- a/app/test/meson.build
> +++ b/app/test/meson.build
> @@ -509,6 +509,17 @@ foreach arg : fast_tests
>                  is_parallel : false,
>                  suite : 'fast-tests')
>      endif
> +
> +    if run_test and arg[0] == 'trace_autotest' and (not is_windows)

Calling this test on Windows should not be an issue.
Is it not the case?


> +        test_args += ['--trace=.*']
> +        test_args += ['--trace-dir=@0@'.format(meson.current_build_dir())]
> +        test(arg[0], dpdk_test,

58/68 DPDK:fast-tests / trace_autotest                   OK              0.16s
59/68 DPDK:fast-tests / trace_autotest                   OK              0.16s

By using the same test name, it's hard to tell what the difference is
between those two lines.
And practically, as a developer reproducing/troubleshooting a test
failure, you can't call only one of the test case.

I am not sure what could be done to enhance this.., how about using a
dedicated test name?
Like below snippet, replacing the whole proposed patch:

@@ -508,6 +508,16 @@ foreach arg : fast_tests
                 timeout : timeout_seconds_fast,
                 is_parallel : false,
                 suite : 'fast-tests')
+        if arg[0] == 'trace_autotest'
+            test_args += ['--trace=.*']
+            test_args += ['--trace-dir=@0@'.format(meson.current_build_dir())]
+            test(arg[0] + '_with_traces', dpdk_test,
+                    env : ['DPDK_TEST=' + arg[0]],
+                    args : test_args,
+                    timeout : timeout_seconds_fast,
+                    is_parallel : false,
+                    suite : 'fast-tests')
+        endif
     endif
 endforeach
  

Hi David, thanks for your review.

On 2022/6/15 22:00, David Marchand wrote:
> On Tue, Jun 14, 2022 at 8:06 AM Chengwen Feng <fengchengwen@huawei.com> wrote:
>>
>> There are a bug[1] when exit application while enable tracing, this
>> bug has not been discovered for a long time, to quickly detect such
>> bugs, this patch was introduced.
>>
>> This patch adds a testcase with trace enabling, it also depends on
>> patch[2] because it has a long file-prefix.
>>
>> [1] eal: fix segment fault when exit trace
>> [2] eal: fix trace init fail with long file-prefix
> 
> This commitlog feels more like a cover letter.
> 
> Please describe what the impact of the patch is, like mention that the
> trace_autotest unit test is being called twice, once with traces
> disabled, and once with traces enabled.
> And that the traces file is being written in a directory part of the
> build directory.
> 
> 
>>
>> Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
>> ---
>>  app/test/meson.build | 11 +++++++++++
>>  1 file changed, 11 insertions(+)
>>
>> diff --git a/app/test/meson.build b/app/test/meson.build
>> index 7fe261cae8..eb37aa632a 100644
>> --- a/app/test/meson.build
>> +++ b/app/test/meson.build
>> @@ -509,6 +509,17 @@ foreach arg : fast_tests
>>                  is_parallel : false,
>>                  suite : 'fast-tests')
>>      endif
>> +
>> +    if run_test and arg[0] == 'trace_autotest' and (not is_windows)
> 
> Calling this test on Windows should not be an issue.
> Is it not the case?

I notice that eal_parse_common_option still keep windows macro for trace parameter parse:

#ifndef RTE_EXEC_ENV_WINDOWS
	case OPT_TRACE_NUM: {
		...
	case OPT_TRACE_MODE_NUM: {
		if (eal_trace_mode_args_save(optarg) < 0) {
			RTE_LOG(ERR, EAL, "invalid parameters for --"
				OPT_TRACE_MODE "\n");
			return -1;
		}
		break;
	}
#endif /* !RTE_EXEC_ENV_WINDOWS */

Maybe trace already support in windows, but I havn't the windows enviorment to verify it.

So I think we should keep '(not is_windows)' currently.

> 
> 
>> +        test_args += ['--trace=.*']
>> +        test_args += ['--trace-dir=@0@'.format(meson.current_build_dir())]
>> +        test(arg[0], dpdk_test,
> 
> 58/68 DPDK:fast-tests / trace_autotest                   OK              0.16s
> 59/68 DPDK:fast-tests / trace_autotest                   OK              0.16s
> 
> By using the same test name, it's hard to tell what the difference is
> between those two lines.
> And practically, as a developer reproducing/troubleshooting a test
> failure, you can't call only one of the test case.
> 
> I am not sure what could be done to enhance this.., how about using a
> dedicated test name?
> Like below snippet, replacing the whole proposed patch:

Good idea, already fix in v3

> 
> @@ -508,6 +508,16 @@ foreach arg : fast_tests
>                  timeout : timeout_seconds_fast,
>                  is_parallel : false,
>                  suite : 'fast-tests')
> +        if arg[0] == 'trace_autotest'
> +            test_args += ['--trace=.*']
> +            test_args += ['--trace-dir=@0@'.format(meson.current_build_dir())]
> +            test(arg[0] + '_with_traces', dpdk_test,
> +                    env : ['DPDK_TEST=' + arg[0]],
> +                    args : test_args,
> +                    timeout : timeout_seconds_fast,
> +                    is_parallel : false,
> +                    suite : 'fast-tests')
> +        endif
>      endif
>  endforeach
> 
>
  

On Fri, Jun 17, 2022 at 4:43 AM fengchengwen <fengchengwen@huawei.com> wrote:
>
> Hi David, thanks for your review.
>
> On 2022/6/15 22:00, David Marchand wrote:
> > On Tue, Jun 14, 2022 at 8:06 AM Chengwen Feng <fengchengwen@huawei.com> wrote:
> >>
> >> There are a bug[1] when exit application while enable tracing, this
> >> bug has not been discovered for a long time, to quickly detect such
> >> bugs, this patch was introduced.
> >>
> >> This patch adds a testcase with trace enabling, it also depends on
> >> patch[2] because it has a long file-prefix.
> >>
> >> [1] eal: fix segment fault when exit trace
> >> [2] eal: fix trace init fail with long file-prefix
> >
> > This commitlog feels more like a cover letter.
> >
> > Please describe what the impact of the patch is, like mention that the
> > trace_autotest unit test is being called twice, once with traces
> > disabled, and once with traces enabled.
> > And that the traces file is being written in a directory part of the
> > build directory.
> >
> >
> >>
> >> Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
> >> ---
> >>  app/test/meson.build | 11 +++++++++++
> >>  1 file changed, 11 insertions(+)
> >>
> >> diff --git a/app/test/meson.build b/app/test/meson.build
> >> index 7fe261cae8..eb37aa632a 100644
> >> --- a/app/test/meson.build
> >> +++ b/app/test/meson.build
> >> @@ -509,6 +509,17 @@ foreach arg : fast_tests
> >>                  is_parallel : false,
> >>                  suite : 'fast-tests')
> >>      endif
> >> +
> >> +    if run_test and arg[0] == 'trace_autotest' and (not is_windows)
> >
> > Calling this test on Windows should not be an issue.
> > Is it not the case?
>
> I notice that eal_parse_common_option still keep windows macro for trace parameter parse:
>
> #ifndef RTE_EXEC_ENV_WINDOWS
>         case OPT_TRACE_NUM: {
>                 ...
>         case OPT_TRACE_MODE_NUM: {
>                 if (eal_trace_mode_args_save(optarg) < 0) {
>                         RTE_LOG(ERR, EAL, "invalid parameters for --"
>                                 OPT_TRACE_MODE "\n");
>                         return -1;
>                 }
>                 break;
>         }
> #endif /* !RTE_EXEC_ENV_WINDOWS */
>
> Maybe trace already support in windows, but I havn't the windows enviorment to verify it.
>
> So I think we should keep '(not is_windows)' currently.

Indeed.