Re: [bug #42447] Fix exporting symbols for recursive make on VMS.

[John E. Malmberg]

On 6/6/2014 10:19 AM, h.becker wrote:
> On 06/06/2014 05:44 AM, John E. Malmberg wrote:
> > On 6/1/2014 3:22 PM, h.becker wrote:
> > > On 05/28/2014 07:00 AM, John Malmberg wrote:
>
> > And the pipe command may allow implementing I/O redirection with out a
> > command procedure if the lines are short enough.
>
> As I documented somewhere, I/O redirection is not really something make
> has to do. But it was already there, when I looked at the sources.
> Whether VMS pipe commands are the right solution for I/O redirection is
> another question: I don't see that the pipe command supports appending
> output. And I guess that pipe commands tend to be longer than simple
> commands.

The make self tests are expecting recipes to redirect output, so it is 
useful to have for VMS.

And someone needed in a VMS specific makefile to add it.

I think there is a way to use pipe commands to do appending of output.
I may have even used it in the patches for the make tests.  By moving 
the code into the vmsjobs.c or equivalent, it can remove VMS specific 
changes from the self tests.  I have to do some tests.

Right now, my focus is on getting all the tests to pass.

> > The current code appears to be always using the older limitation on the
> > command size when determining when to switch to command files, instead
> > of basing it on the version of VMS used for the compile.
>
> "for the compile"? Maybe I have a language problem, but I think it
> depends on the VMS version/environment where Gnu make runs.

A program built on a version of VMS that does not support the longer 
command lines would not be able to see the compile time symbols that the 
support was present on, so while it could build the support in, the code 
would look different.

A program built on a version of VMS that supports the longer command 
lines would not be able to run on the older versions of VMS, unless 
special (and mostly undocumented) build procedures are used.

And then you need the older version of VMS to validate that the build 
works.  Currently I am limited to VAX/VMS 8.3, Alpha 8.3/8.4, and 
Itanium 8.4.

So that is why I consider it a compile time option.

> > For exported symbol substitution, I would prefer it to be done when the
> > command to be sent to the child process is created.
>
> Does that mean re-implementing in make what DCL can do? As with I/O
> redirection, I don't think this is something make should do. Also, I'm
> not sure if that's what a user expects. For example, f$time(0) may be
> slightly off and f$getjpi(0,"pid") is simply wrong.

Lexical functions work in recipe commands now.  Symbol substitution does 
not.  I am just looking at simulating $xxx expansion at this point, done 
before the command is run.  GNU make is already supporting other bits of 
shell simulation on VMS, so the precedent of doing this type of thing 
has already been set.

I also plan to look at having a pool of subprocesses and sending 
commands to them over mailboxes instead of spawning a new process for 
each line of a recipe.  This should speed make, particularly on the 
older systems.

Regards,
-John