Re: Aggregate GPRs

[Simon Wright <simon@pushface.org>]

Mark Lorenzen <mark.lorenzen@gmail.com> writes:

> On Sunday, August 30, 2015 at 7:24:01 PM UTC+2, Simon Wright wrote:
>> I see from the documentation[1] that, for projects built through the
>> aggregate project, package Builder is ignored (even if the aggregate
>> project doesn't have a package Builder?).
>> 
>> I've just found (with GPRBUILD GPL 2015) that the attributes Target and
>> Runtime are also ignored!
>
> I think this is reasonable. When aggregating several projects into an
> aggregate project, it is natural that they are build for the same
> target and same RTS. No?

No, I don't think so; what if you were building two sides of a
communications protocol to run on different hardware?

From my point of view, you can't use aggregate GPRs for cross-builds
unless you're prepared to go through hoops to set up a configuration
GPR. Which could *not* be done except on the user's compilation platform
[unless of course you know different], so I'd have to write a
complicated set of instructions for users rather than providing a set of
GPRs that could be used out of the box.

>> These features seem to me to reduce the usefulness of aggregate projects
>> to the point of uselessness. Does anyone have positive experiences with
>> them?
>
> Yes. We use an aggregate project to set an environment variable
> (LDSCRIPT) that is used by the linker to locate the linker script. The
> value is derived from a scenario variable and it would thus not be
> easy to set it in the shell before invoking GPS or command-line
> tools. The aggregate project is thus a simple wrapper around the
> "real" project.

Which is exactly what I wanted to do, but couldn't.