Re: vms:test variables

[John E. Malmberg]

On 4/7/2014 4:37 PM, h.becker wrote:
> On 04/06/2014 07:34 PM, John E. Malmberg wrote:
> > Ok, so now we have the a baseline test suite ported, what is it going to
> > take to get this stuff committed so that we can start working on the
> > bugs that have been exposed?
>
> I don't know who will/should answer such a question. I still have no
> luck nor fun with these tests:
>
> $ @TEST_MAKE
> ------------------------------------------------------------------------------
>                        Running tests for GNU make on VMS
>                                 GNU Make 4.0.90
> ------------------------------------------------------------------------------
>
> Clearing work...
> Finding tests...
>
> features.dir/archives ...................................
         ^^^^^^

This is a day one bug in the older Perl on VMS versions that was fixed 
in about the Perl 5.10 timeframe, but the fix was not made the default 
behavior until Perl 5.18, as it is tied to supporting EFS character 
filenames.

Perl 5.18 and later suppresses the incorrect ".dir" being appended to 
paths by default.  The incorrect behavior can be optionally turned on 
for existing PERL/VMS scripts that expect the buggy behavior.  I did not 
test the script with older versions of Perl, or with the older behavior 
enabled.

FEATURES/ARCHIVES ....................................... Error running 
make (expected 0; got 512): make -f 
/lcl_root/make/tests/work/FEATURES/ARCHIVES.mk
<skip>
FAILED (2/10 passed)

8 features/archive tests are failing for me because the archive support 
is incomplete.

> features.dir/comments ...................................
FEATURES/COMMENTS .................................... ok   (1 passed)

> features.dir/conditionals ...............................
FEATURES/CONDITIONALS ................................ ok   (4 passed)

> features.dir/default_names ..............................
FEATURES/DEFAULT_NAMES ............................... ok     (2 passed)

> OK, I can run one when I specify one, but it fails due to bash's ":":

It is not failing, the warning from DCL is written out, but does not 
otherwise affect the execution.  Rather than modify all the tests that 
were using ":" as null command, I changed the compare_output routine in 
the test harness to ignore it.

> $ make -f [.tests.work.variables]undefine.mk
> recursive simple recursive
> undefined recursive undefined
> $ @TEST_MAKE -keep "variables/undefine"

VARIABLES/UNDEFINE ......................................
run_command_with_output(/lcl_root/make/tests/work/VARIABLES/UNDEFINE.log,): 
make -f /lcl_root/make/tests/work/VARIABLES/UNDEFINE.mk
run_command_with_output returned 0.
vms status = 1
Comparing Output ........ DIFFERENT OUTPUT

<passing tests removed>
FAILED (3/4 passed)

> So it seems the tests still contain Unix/shell commands and not only VMS
> commands.

I only made changes needed to test the feature.  GNU Make does not care 
if a command is a foreign command or a native VMS command.

> $ make -f [.tests.work.variables]undefine.mk
> recursive simple recursive
> undefined recursive undefined

*** /lcl_root/make/tests/work/VARIABLES/UNDEFINE.base   Wed Feb  4 
23:01:01 2015
--- /lcl_root/make/tests/work/VARIABLES/UNDEFINE.log    Wed Feb  4 
23:01:01 2015
***************
*** 1,2 ****
  recursive simple recursive
! undefined undefined undefined
--- 1,5 ----
  recursive simple recursive
!
! undefined undefined recursive
!
! %DCL-W-NOCOMD, no command on line - reenter with alphabetic first 
character

> Or am I missing something?

The extra lines and the "%DCL-W-NOCOMD" lines are ignored by the test 
harness output compare.  If that compare fails, then the test harness 
issues a diff command on the raw output, not the cooked output used for 
grading the test, so every thing is present.

The actual failure appear to be that the second line should be 
"undefined undefined undefined" and on VMS is showing up as "undefined 
undefined recursive".

I have not fully investigated this issue, but it appears that this 
happens when make recurses or invokes itself, and on VMS, the second 
invocation of make is missing some information from the parent.  My 
guess is that information is normally passed by exported shell symbols.

Regards,
-John