Re: [bug #41758] VMS Make incorrectly reports archives support present.

[John E. Malmberg]

On 5/3/2014 4:30 AM, h.becker wrote:
> On 05/02/2014 03:17 PM, John E. Malmberg wrote:
> > Sounds reasonable.  I have not used image libraries in user mode code so
> > was not aware that they were also .olb.
>
> Hmm, the VMS linker uses imagelib.olb and you probably linked against
> shareable images from that library. So you may talk about something
> else. To me it seems convenient to use image libraries.

I meant explicit use, not the implicit use where the linker handles 
things automatically.

> Anyway, appended is a patch to the suggested patch for default.c. It
> implements checking for the VMS libraries within make, using make functions.
>
> Some notes:
>
> To me it looks OK to use this complex make function calls. Maybe it can
> be simplified. The other option is to write a new VMS specific function,
> like $(vms_create_library LIBRARY,MEMBER), which expands to the "if
> f$search ...".

That seems like a valid optimization.  Because process creation on VMS 
is more expensive than on Unix, anytime a process creation can be 
avoided can be a significant optimization.

> I'm not sure I understand why the previous code has the
> 'f$edit(...,"LOWERCASE")'. My change doesn't really LOWERCASE the
> suffix, but it works on all lower- and uppercase suffixes.

The lowercase was needed to do the compare.  I have switched to your 
functions and they are passing the test case after revising the image 
library to be an .OLB.

My local [.vms]library test script is now testing image libraries in 
addition to the other types as I have incorporated your examples into it.

I have the wildcard library processing coded and now all 10 
[.features]archive tests are usually passing on VMS, including the one 
that uses a VPATH.

A few of them are intermittently failing, so I am not posting the patch 
yet until I isolate it and come up with a fix.

The test script is passing about 3 times in a row, and then fails either 
because an object module is replaced when it should not be, or is not 
replaced when it should be.

It appears that there is a problem somewhere in comparing of the library 
insertion timestamp and the VMS object module timestamp.  Most of the 
time it works, but occasionally it fails.

I have instrumented the test makefile to log the timestamps that VMS 
sees, since that was trivial, and those timestamps are all as expected.

The next step is to have a test version of arscan.c start printing out 
the timestamps before and after the VMS local time to UNIX UTC 
conversion and build it as a standalone executable.

Unfortunately none of the built in make debug options will display the 
timestamps of the files as it considers the rules.

Regards,
-John