Re: [Bug-gnulib] serial numbers in .m4 files: are they still useful?

[Paul Eggert]

address@hidden (Karl Berry) writes:

> To get around this, a datestamp (YYYYMMDD.HHMM or some such) could be
> used, updated automatically with time-stamp.el or similar.  This is done
> for texinfo.tex, mkinstalldirs, and many other scripts these days.

I'm not a big fan of datestamps like that, since too many non-Emacs
tools touch those files and they don't remember to maintain the
datestamps.  Also, like #serial, they introduce unimportant differences.

I agree that '$Id$' etc. is not the way to go.

Bruno Haible writes:

> For files which are frequently copied from one package to another, I find
> these serial numbers quite useful: For developers who don't know where
> the master source and CVS can be found, it easily answers the question
> which version of the .m4 files to prefer in a new release.

Unfortunately I've had quite a difference experience with that.  I've
often encountered situations where the serial numbers were wrong, and
gave the wrong indication of which file was newer.  For example, I
just encountered that situation with regex.m4 about an hour ago, with
a copy of coreutils and a copy of gnulib that I was messing with.

Since the serial numbers are often wrong, I find that I have to look
at the contents anyway.  So for me, the serial numbers are a
maintenance hassle with no redeeming benefit.

However, since other people find them useful let's keep them in, at
least unless someone can figure out a better way.