Re: Autoconf fails test 4 with m4-1.4.7a

[Stepan Kasal]

Hi,

On Tue, Oct 24, 2006 at 06:19:44PM +0200, Ralf Wildenhues wrote:
> With current Autoconf HEAD (and 2.60, FWIW), and M4 branch-1_4,
> I get a failure of Autoconf's test 4 on my GNU/Linux system.

I take the liberty to decipher Ralf's report a bit.  If we have a 5
line input file:

m4_define(`ONE', `TWO(`$1')')
m4_define(`TWO', `$1')
ONE(`a
b
c')

then the previous versions of m4, when tracing ONE and TWO, reported
ONE on line 5
TWO on line 5

It can be argued that the occurence of ONE starts on line 3, so it
would be better to report that line number.  But since TWO is
expanded ``inside'' the expansion of ONE, the ``more friendly'' line
number output would be:
ONE on line 3
TWO on line 3

It can be tricky to achieve that goal, but I believe it is possible.

But current m4 goes only half-way, saying:
ONE on line 3
TWO on line 5

I beleive that this is not worth the incompatibility hassle, and I
would humbly suggest backing out this change.  (Or completing it,
as sugested above.)

What do you think?
        Stepan