Re: [PATCH] gnulib-tool: List modules separately, explicit vs dependencies.

[Bruno Haible]

Hi Thien-Thi,

> I saw the huge list of modules
> pulled in and wondered where to start cutting (via "--avoid").
> This change to gnulib-tool helps me to do the job more quickly
> by letting me ignore (for the purposes of experimental cutting)
> the explicitly specified modules.

If you just cut randomly, you will introduce bugs in your package,
for sure.

There are two reasonable uses of '--avoid':
  1) When the documentation of the modules mentions that a module
     fixes bugs on particular platforms, and you know that these
     platforms are not used by your users.
  2) When you provide fixes for the same bugs in a different way.

> +    # Separating explicit from implicit gives an idea of how bad your
> +    # particular case of creeping gnulibitis has developed.  --ttn

I don't think it is generally useful to repeat a program's input in
its output. The output should contain some level of information
that the user is not aware of.

Normally people are aware which gnulib modules they have specified.
If not, they can look it up in the gnulib-cache.m4 file.

For this reason, I would not like to apply your patch.

> +    sed -e 's|.*|/&/d|' "$tmp"/want > "$tmp"/want.sed
> +    echo '(dependencies)'
> +    echo "$modules" | sed -f "$tmp"/want.sed | sed -e 's/^/  /'

The autoconf documentation, section "Limitations of Usual Tools",
mentions that "HP-UX sed has a limit of 99 commands". But
one can have several module names specified to gnulib-tool.
Therefore, the more portable (and more efficient) way to produce
the difference set between two sets of module names is through
the 'join' program.

Bruno