Re: [Nicolas Joly <address@hidden>] Re: CVS automake testsuite failures under Tru64 unix

[Akim Demaille]

|>>> "Akim" == Akim Demaille <address@hidden> writes:
| Akim> Also, the traces contain something extremely suspicous:
|
| Akim> m4trace:configure.in:1: -1- m4_define([AC_PACKAGE_TARNAME], 
[am--<garbage>]
|
| Akim> No doubt that the problem is here.  What the heck is this crap?
|
|I can reproduce something similar to an example Nicolas posted
|earlier, playing with sinclude and include:
|
|Running autoconf on 
|
|  AC_INIT([am__sinclude], [1.0])
|
|gives 
|
|  # Identity of this package.
|  PACKAGE_NAME='am__sinclude'
|  PACKAGE_TARNAME='am--'
|  PACKAGE_VERSION='1.0'
|  PACKAGE_STRING='am__sinclude 1.0'
|  PACKAGE_BUGREPORT=''
|
|but on
|
|  AC_INIT([am__include], [1.0])
|
|you get
|
|  # Identity of this package.
|  PACKAGE_NAME='am__include'
|  PACKAGE_TARNAME='am--include'
|  PACKAGE_VERSION='1.0'
|  PACKAGE_STRING='am__include 1.0'
|  PACKAGE_BUGREPORT=''
|
|Note the definition of PACKAGE_TARNAME in both cases.  

Heck, I never thought about this.


Well, I think it is high time to disable include and sinclude from
Autoconf.  That's too dangerous.  Paul, do we agree?  m4_include and
m4_sinclude are available since 2.50, and people are not supposed to
use include and sinclude.  Do your greps reveal some people still use
it?



As an aside.  I personnally consider it is a flawn in the specs of M4:
IMNSHO, something like

        patsubst([foo], [foo], [dnl])

should return a *quote* string, i.e., [dnl], not dnl.  That's the
bottom line of our problem.  I would really love to see a means to
educate GNU M4 in this regard.  I don't really care how: be it a new
series of macros, a new option, a new name, whatever.  I just want its
text processing macros to return a quoted string.