[PATCH] CI: Don't execute static analysis on tests.

[Lukas Slebodnik]

ehlo,

The simple patch is attached.
The side effect of this patch is reduction time spent in the scan-make-tests
from 9 minutes to 4 minutes.

LS

[Lukas Slebodnik]

On (09/10/14 10:01), Lukas Slebodnik wrote:
> ehlo,
>
> The simple patch is attached.
> The side effect of this patch is reduction time spent in the scan-make-tests
> from 9 minutes to 4 minutes.
>
> LS
> From c0beec11cb406b68fed8a3367795bafdd108aefb Mon Sep 17 00:00:00 2001
> From: Lukas Slebodnik 
> Date: Thu, 9 Oct 2014 09:50:50 +0200
> Subject: [PATCH] CI: Don't execute static analysis on tests.
>
> clang does not understand semantic of asserts from libraries.
>
> Example of false positevive.
> 121     data->dirp = opendir(TESTS_PATH);
>    // Value assigned to field 'dirp'
> 122     if (data->dirp == NULL) {
>    //Assuming pointer value is null
>    //Taking true branch
> 123     int err = errno;
> 124     fprintf(stderr, "Could not open directory:'%s' [%s]\n",
> 125     TESTS_PATH, strerror(err));
> 126     }
> 127     assert_non_null(data->dirp);
> 129     data->dir_fd = dirfd(data->dirp);
>    //Null pointer passed as an argument to a 'nonnull' parameter
> Execution of tests would fail on line 127.
>
> contrib/ci/run | 6 ++----
> 1 file changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/contrib/ci/run b/contrib/ci/run
> index 00cf95094e62a16dfa60b5365ed4373f01bfdf51..cd5897c66bfca15f9ed87298affc147bc4500a58 100755
> --- a/contrib/ci/run
> +++ b/contrib/ci/run
> @@ -255,13 +255,11 @@ function build_coverage()
>                                            "${CONFIGURE_ARG_LIST[@]}" \
>                                            --with-test-dir="$test_dir"
> # Build everything, including tests
> # Not building "tests" due to https://fedorahosted.org/sssd/ticket/2350
> stage scan-make-tests   scan_build_single \
> stage scan-make         scan_build_single \
>                            "$scan_report_dir" \
>                            -plist-html \
>                            --html-title="sssd - scan-build report" \
>                            make-check-wrap -j $CPU_NUM check -- true
>                            make -j $CPU_NUM all

^^^
                                                I can call make with V=0
                                                if we decide to do it.

LS

[Nikolai Kondrashov]

On 10/09/2014 01:05 PM, Lukas Slebodnik wrote:
> On (09/10/14 10:01), Lukas Slebodnik wrote:
> >                            make -j $CPU_NUM all
>                                                    ^^^
>                                              I can call make with V=0
>                                              if we decide to do it.

I think it would be irrelevant to this patch. Let's add it to whatever
solution we implement for enforcing Clang check.

Nick

[Nikolai Kondrashov]

On 10/09/2014 11:01 AM, Lukas Slebodnik wrote:
> ehlo,
>
> The simple patch is attached.
> The side effect of this patch is reduction time spent in the scan-make-tests
> from 9 minutes to 4 minutes.

ACK

It looks fine, but I haven't tested it.
It would be nice to have a link to CI test results to go with the patch.

Can we work towards fixing the test frameworks so Clang understands
what's going on?

Nick

[Nikolai Kondrashov]

On 10/10/2014 12:06 PM, Nikolai Kondrashov wrote:
> On 10/09/2014 11:01 AM, Lukas Slebodnik wrote:
> > ehlo,
> >
> > The simple patch is attached.
> > The side effect of this patch is reduction time spent in the scan-make-tests
> > from 9 minutes to 4 minutes.

Actually, there should be no reduction in overall run time, as tests will be
analyzed when they're built in the following "make-check" stage.

This does reduce number of warnings in the report, though.

Nick