Re: How could I help now?

[Ramón García]

I have been doing some tests. I can confirm that if the call that
initializes automatic variables (call set_file_variables() in function
expand.c::reference_variable) is removed, the program behaves
correctly in the interesting cases. So, as far as this project is
concerned, removing the initialization of automatic variables before a
target::var expansion is safe.

>      1. Always returns the right value regardless of where/when it's
>         used.
>      2. Always returns the empty string until after all makefiles are
>         read in and make starts processing the rules; then always
>         returns the right value.
>      3. Returns the empty string up until target T is processed; then
>         returns the right value.
>      4. Always returns the empty string unless we're currently
>         processing target T (that is, $(T::@) cannot be used outside of
>         the recipe for T).

As far as the out of date determination is concerted both #2 and #3
satisfy requirements. I prefer 2 because the behaviour is simple for
the user. The behaviour of automatic variables defined if the target
is before the current one in the depth-first order is too complex. It
will cause suprises. But I am not completely sure.

Ramon