Re: VAX/VMS 7.3 and build from copy of master.

[John E. Malmberg]

On 2/20/2014 9:52 AM, h.becker wrote:

> 1. building from a git-cloned repository
>
> config_h_from_vms_template.com is a rather long name, no problem.
> However, there seems to be something similar for w32: "Windows32 SCM
> build preparation of ...". That file is named prepare_w32.bat. So I
> suggest to rename config_h_from_vms_template.com to prepare_vms.com. OK,
> the lowercase vms/w32 is not consistent with other file names which have
> VMS/W32, if that is important ...

Ironically the file case is significant if you are serving files via NFS 
to VAX/VMS.  When doing that, only lower case filenames show up 
correctly on the VAX.  The rest get NFS mangled.

I am ok with prepare_vms.com but prefer a name that describes the 
function, since a prepare_vms.com could mean doing many things, and this 
is a specialized command procedure.

> I would rather use configure.ac to determine the package and the version.

Now that I know where it that information is, I would agree.  I was 
searching for the word "version" in that file.  I did not try just 
reading it.

> Whether it is useful to check for NFS/ODS-2 encoded filenames, I don't
> know. But then, when a git clone is copied to an ODS-2 disk, somehow,
> how would one handle filenames with multiple dots, anyway. I suggest to
> simply require an ODS-5 disk for git cloned repositories and leave it to
> the user of old VMS systems to handle problems on other disks.

As I stated above, I am using a common NFS served disk for all my 
systems.  So for me to build on VAX, I have to deal with the NFS mangled 
names.

How a clone shows up on ODS-2 depends on what tool was used to unpack it.

> The user
> can change the symbol config_template, anyway. Maybe a comment should
> explain this, or maybe a P1 can be added so that a user can supply an
> ODS-2 name. To me it seems removing these special filenames makes things
> simpler. And, to make generating the config.h-vms more obvious I would
> use some "replace" commands. Again, I use a "new" feature: pipe, here to
> avoid creating a temporary file.

Some VMS utilities do not play well with random accessing files on an 
NFS volume, particularly before VMS 8.4.  So I use a more restricted set 
for building.  I did not create any temporary files in my scripts.

> Whether invoking this command procedure should go into makefile.com is
> another question. I think it isn't necessary. The prepare_w32.bat isn't
> run from any other .bat-file, as far as I can see.

I think it should go there, then we have the same build procedure on 
master as in a release tarball.  Eventually I hope to get to a single 
command file that will do the complete build and create a PCSI kit with 
no extra editing.

> 2. Compiling and linking guile
>
> Yes, after the release of 4.0 there was a change which added a call to
> guile_gmake_setup to main.c. I sent a patch to handle that for VMS, in
> last November. Obviously it didn't make it into the repository.
>
> But I don't see that you need gmk-default.h without having HAVE_GUILE
> defined.
>
> To me it seems enough to change makefile.com and makefile.vms as in the
> appended diff/patch and then there is no need for gmk_default_h.com.

I am not sure yet what the guile module can do on VMS, so I would prefer 
to leave generating the header file in place until that is determined.

> 3. Changes in config.h-vms.template
>
> Isn't it more obvious to do what decc$types.h does?
> #ifndef __CRTL_VER
> # define __CRTL_VER __VMS_VER
> #endif

It is also not a good idea.  The double underscore macros should only be 
redefined as a last resort.  Doing the above hack will also unexpectedly 
impact people doing cross version builds.

Are you subscribed to bugs-make or equivalent, or do you want me to 
continue to bcc you?

I will try to put together a revised patch this weekend.

Regards,
-John