Re: transition periods

[Bruno Haible]

Paul Eggert wrote:
> if I see the warning I do the same thing that I 
> do if I see the error from the missing #include, so there's not much 
> point in having a transition period with the warning.

I see a value for the transition period in three cases:

1)
> Things can be different the transition involves other ports or rarely 
> used platforms. But this one involves all GNU/Linux platforms so I see 
> the #warning right away.

The package might use the getprogname module only on specific platforms.
Thus, even though our change in Gnulib affected all platforms, in the
specific package it may not be visible immediately.

2) The maintainer may have a fixed release date scheduled. He then
wants to upgrade to a newer Gnulib, to fix some issue, but without
getting dragged into other changes.

3) The package may have a master branch and a release branch. The maintainer
uses "diff -r" to compare the code in the two branches. The two branches
may use different versions of Gnulib (via git submodules). Extra changes
between the two branches are unwelcome, even small ones. Thus the maintainer
will not want to remove '#include "getprogname.h"' statements as long as
his release branch is active; this can be several months.

These are just the scenarios that come out of my experience as maintainer /
release manager. I'm sure distro packagers have their habits, and some of
them will also favour longer transition periods.

Bruno