Re: How to speed up 'automake'

[Jim Meyering]

On Tue, May 3, 2022 at 3:55 PM Nick Bowler <nbowler@draconx.ca> wrote:
>
> On 2022-05-02, Karl Berry <karl@freefriends.org> wrote:
> >     - @echo '# dummy' >$@-t && $(am__mv) $@-t $@
> >     + @: >>$@
> >
> > 1) does it actually speed anything up?
>
> The answer seems to be a resounding "yes".  I tried one of my packages
> on an old slow PC, and changing this one line in Makefile.in cuts almost
> 5 seconds off of the depfiles generation step in config.status.
>
> (All .deps directories manually deleted between runs as otherwise the
> rule commands will not be executed).
>
>   Before (x5):
>   % time config.status Makefile depfiles
>   real  0m15.320s
>   real  0m15.210s
>   real  0m15.210s
>   real  0m15.210s
>   real  0m15.220s
>
>   After (x5):
>   % time config.status Makefile depfiles
>   real  0m10.650s
>   real  0m10.550s
>   real  0m10.550s
>   real  0m10.550s
>   real  0m10.650s
>
> That 5 seconds is a relatively small part of total configure runtime but
> it is noticeable.
>
> So if make implementations have no problem including empty files (I tried
> a few and all seem OK with it) then it seems like a win.
>
> > 2) without the mv I fear we are no longer noticing write failure
>
> I think it's OK.  All shells that I know of set a failure status when
> redirection fails, at least for simple commands like that.

FYI, in general, the $@-t dance is not really about error detection so
much as persistent-error avoidance. For most uses of this idiom, we
must be careful that we don't end up with the target ($@) containing
the results of an incomplete or corrupted write **AND** an up-to-date
timestamp (the "persistent" part), so we update atomically only after
successful result creation into $@-t. However, in this case, we don't
care about the contents of that file: just the timestamp, so
eliminating the intermediate $@-t is a fine move.