Re: Automatic set-serial?

[Peter Simons]

Hi Reuben,

 > A SET_SERIAL_PATH environment variable to set-serial.sh which, if
 > set, causes files of the form $SET_SERIAL_PATH/`basename $i` to be
 > read instead of $i. (And of course, $i is written to.)
 >
 > Any problems with this, or a better way?

I wonder whether the 'dist' hook is the best place to run set-serial.sh to
begin with. It feels like that stage is too late in the process. If we run
set-serial.sh in 'dist', then we ensure that #serial numbers are accurate in
release tarballs. What we'd really like to ensure, however, is that every
macro's #serial number is accurate in the Git repository.

The natural place to do that, IMHO, is Git's pre-commit hook. We could add a
hook that simply refuses any commits that modify a macro without bumping the
#serial number. This this way, the repository would be guaranteed to be
accurate at all times.

The disadvantage of relying on the commit hook is that administrators must
remember to activate that hook in their local check-outs before committing.
Personally, I feel that's a minor problem, because there's just a handful of
people who commit to the repository, and we all read this list and are aware
of the problem, but still the hook-solution doesn't seem to be entirely
bullet-proof.

A different approach would be to modify the build process, i.e. the
'maintainer-all' target. As of now, that target already ensures consistent
formatting of the macros. It could just as well guarantee accurate #serial
numbers, but to do that we would have to extend the Python code in macro.py.

Is there another solution that I overlooked?

Take care,
Peter