Re: 63-rename-var-vars.patch

[Akim Demaille]

>>>>> "Tom" == Tom Tromey <address@hidden> writes:

>>>>> "Akim" == Akim Demaille <address@hidden> writes:
Akim> The aim of this patch and the next is to clarify the status of
Akim> variables: who owns them, and how their defined type (+=, := or
Akim> =) can change.

Tom> Why does it matter who owns them?

Because...

Akim> I fail to understand why the user is not allowed to += a
Akim> nonexistent variable.

Tom> Perhaps this was intended to catch user errors?

Tom> Actually in the automake I have we don't have this restriction.
Tom> We just have a check to make sure the user doesn't `+=' an
Tom> internally-defined variable.  This check was added because the
Tom> implementation didn't allow this to work.

of this.  To understand the write permissions which were coded, I had
to introduce the idea that the user has more rights than Automake.  In
particular Automake cannot override user variables with Automake
variables.  This can happen for instance when configure.ac is read,
producing user variables, not Automake variables.  Or something like
this, I don't remember well.

Also, Automake is not allowed to define twice a var, but the user may
redefine an AM var, hence I need to know who is the last to set a var.



Anyway, now from the point of view of functionalities, there should be
no difference between user and AM vars (now), so gratuitous
limitations should be removed.

What are the ideal rules we'd like to have?