[Info-vax] gcc 4.9.1 and VMS

[Simon Clubley]

On 2014-09-06, John E. Malmberg <wb8tyw at qsl.network> wrote:
> On 9/6/2014 7:08 PM, Simon Clubley wrote:
>> I've just finished trying to build gcc 4.9.1 as a cross compiler
>> for a VMS Alpha target (using a Linux host) and I thought I would
>> let anyone interested know the results are no better than my previous
>> efforts with gcc 4.8.0.
>
> Have you tried building minimal C++ and Fortran Programs using the VMS 
> compilers and disassembling them to compare with what GCC is generating?
>
> On C++ your undefined symbol does not have a VMS decc$ or other prefix, 
> which may be all that is missing for it to be resolved.
>

The C++ cross compiler build is failing on Linux and the symbol is an
internal gcc symbol. A quick look around seems to reveal it's an actual
gcc bug. Fixing it should not be a major problem, but the far deeper
issues revealed means there's no point unless you can have confidence
the rest of the VMS specific code is in a state that's suitable for use.

The Fortran test programs are going down early on the image activation
sequence and fail in VMS debug before the program entry point is reached.
This may be a missing piece of code or may simply be an incorrect build
option (although I tried a range of them, including various exception
models, the last time around).

This brings me back to the size of the gcc code base, both in terms of
understanding _all_ the various bits of the code used for VMS and their
inter-relationships as well as the time it takes to rebuild gcc whenever
I change anything.

One of the problems here is that ACT use their own internal GCC tree for
their professional and GPL compiler products and every so often they
push bits of it into the public FSF tree. Hence, there's no confidence
on my part that all the required VMS bits to actually bring up a cross
compiler are present in the FSF tree.

In fact, if you remember last time around in 2013, they actually said
in the email discussions that not all the bits had been pushed into the
public FSF tree at that time but it's wasn't clear if this was required
core functionality or not.

Anyway, other hobbyist projects await me, so that's it for VMS for now.

Simon.

-- 
Simon Clubley, clubley at remove_me.eisner.decus.org-Earth.UFP
Microsoft: Bringing you 1980s technology to a 21st century world