Re: Change of behavior: 3.80 vs 3.81beta3 WRT line continuation

[Steve James]

Paul, thanks for your reply on this.

I had seen the note in NEWS and I had re-read the latest make.info section 
Command Execution too. Honest! However, I still couldn't be sure if this was 
a bug. I suggest that the nature of this incompatibility is described in the 
manual (or at least the NEWS). I for one could do with some advice on what to 
do about all my make scripts that use the following idiom. I use this a lot.

ascript :
        @ echo -e \
                'line1\n\
                line2\n\
                line3' > $@

I also do this kind of thing a lot:
perl.out : x.pm
        @ perl -Mx -e '\
                $$this = q/that/; \
                do { \
                        something($$this); \
                } or $$not; \
                ' > $@

Can you recommend how to write these in a way that's compatible with both 3.80 
and 3.81? Perhaps I'd get more mileage with double-quotes? Or perhaps I 
should use 'define'?

Cheers,
Steve.


On Sunday 21 August 2005 23:28, Paul D. Smith wrote:
> %% Steve James <address@hidden> writes:
>
>   sj> AFAICT the behavior of line continuation escapes has changed
>   sj> within single-quoted strings. What's really confusing me is that
>   sj> this seems to be deliberate.
>
> Yes, you're right on all counts.  So.... what's confusing? :-)
>
>   sj> all:
>   sj>         @echo 'line1\
>   sj> line2'
>
>   sj> Version 3.80 prints this (as I would expect):
>   sj> line1line2
>
> You may expect that because that's what make has printed in the past,
> but that's not the correct result according to the POSIX definition of
> make.
>
> Please see the NEWS file, or the GNU make manual (the version provided
> with beta3 of course).
>
>   sj> But Version 3.80beta3 prints this:
>   sj> line1\
>   sj> line2
>
> This is correct.
>
>   sj> There's some background to this change concerning POSIX (Savannah
>   sj> bug # 1332).
>
> Right.  Did you look up that bug and the accompanying commentary on
> Savannah?
>
>   sj> See, entirely deliberate. What's more the regression tests in
>   sj> tests/scripts/misc/general3 verify this behavior. But this all
>   sj> looks broken to me.
>
> Why does it look broken?  Run the same command in a (POSIX) shell and
> you'll get the new behavior, not the old behavior:
>
>     $ echo 'line1\
>
>     > line2'
>
>     line1\
>     line2
>
> So, why is it broken that when make invokes a shell to run that same
> command you get the same result?
>
>   sj> So what's up? Is this
>   sj> 1. A new bug in the beta,
>
> No.
>
>   sj> 2. A deliberate non-backward compatibility
>
> Yes... it's a change that was made so that GNU make would conform to the
> POSIX specification for make.  Unfortunately, that involves an
> incompatibility.
>
> Again, the NEWS file has info about this and other user-visible changes;
> it should always be the first place you look when upgrading versions of
> GNU make.

-- 
All power corrupts, but we need electricity.