Re: Fwd: BUG report using autoconf…

[Eric Blake]

On 02/05/2018 04:49 PM, aotto wrote:
> > > You are correct that autoconf's detection of unexpanded macros is
> > > unaware of comment syntax.  Would you like to prepare a patch to the
> > > post-processing pass that looks for unexpanded macros to first filter
> > > out all comment lines?

Checking for unexpected names inside comments is different from checking 
for intentional output embedded in error messages, where the code has to 
work even without comments protecting the text.

> > >
> > the problem is not only the comment… today I had a "string" with 
> > macroname inside in the file
> > acinclude.m4… and "aclocal --force -I m4" run "forever" and eat all my 
> > memory… example:
> >
> > AC_DEFUN([SC_ENABLE_DEBUG], [
> >     AS_IF([test "x$config_site" = "xyes"], [
> >       AC_MSG_NOTICE(["ENABLE_DEBUG: using 'config.site' defaults"])
> > …
> >
> > if you switch ENABLE_DEBUG to SC_ENABLE_DEBUG… then you have a really 
> > BIG problem… :-)

That's why the manual suggests that you use a quadrigraph for cases like 
that, as in:

AC_MSG_NOTICE(["SC_@&address@hidden: ...."])

where the @&t@ quadrigraph expands to the empty string AFTER all 
checking for unexpanded macros has taken place, and its presence in the 
meantime is enough to prevent unintended infinite recursion of trying to 
expand the SC_ENABLE_DEBUG macro from its own output.

--
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org