bug#11485: In configure.in, gl_EARLY is too early?

bug#11485: In configure.in, gl_EARLY is too early?

[Glenn Morris]

Package: emacs
Version: 24.0.96
Severity: important


configure.in contains the equivalent of this (abridged):

  ## On several platforms:
  NON_GNU_CPP=<something>

  # Initialize gnulib right after verifying that the C compiler works.
  gl_EARLY

  [...]

  ## If no CPP was specified, do something for Sun compilers.
  if [...] && test x"$CPP" = x; then [do stuff]

  [...]

  ## If NON_GNU_CPP was set and CPP was not, set CPP to NON_GNU_CPP.
  if test "x$NON_GNU_CPP" != x && test x$GCC != xyes && test "x$CPP" = x
  then
    CPP="$NON_GNU_CPP"
  fi

  [...]

  AC_PROG_CPP


The CPP tests all predate the use of Gnulib.

What gl_EARLY does is opaque, but by inspection of configure one thing
it ends up doing is calling AC_PROG_CPP and so setting $CPP much earlier
than the rest of configure.in expects.

IIUC, this will prevent all the pre-existing code related to changing
CPP from having any effect; because it all tests if CPP was already set,
and if so does nothing. It was expecting that CPP could only have been
set by the user at this stage; however now gnulib is going to make it
always be set.


Moving gl_EARLY later in configure.in (to just before the check for
-Wdeclaration-after-statement) would presumably restore the behaviour of
Emacs 23 and earlier.

However, all previous pretesting for Emacs 24 has used the other
behaviour, so I do not know if it is too late to change this.
It affects the less common platforms, which might not have seen much
testing.

One example is using a Sun compiler on Solaris.
The intent of configure was to add -Xs to the CPP options.

bug#11485: In configure.in, gl_EARLY is too early?

[Andreas Schwab]

Isn't the CPP dance obsolete now that we no longer preprocess makefiles?

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

bug#11485: In configure.in, gl_EARLY is too early?

[Paul Eggert]

Andreas is right that CPP is no longer needed.
I have removed that stuff from the trunk in bzr 108252.
In the trunk, I also moved gl_EARLY to the place where
Glenn suggested, as that makes sense.

If gl_EARLY overrides the other CPP stuff
in emacs-24, that shouldn't harm emacs-24, since we don't
care what CPP is.

However, if gl_EARLY overrides the other CC stuff
that might be a problem.  This would affect two platforms:
Apple Darwin / Maco OS X (where emacs-24 currently
adds -I/sw/include and -L/sw/lib) and Iris 4D
(where it adds -D_LANGUAGE_C).  So it might make
sense in the emacs-24 trunk to move gl_EARLY as Glenn
suggests, if only because of these two issues.

bug#11485: In configure.in, gl_EARLY is too early?

[Glenn Morris]

Andreas Schwab wrote:

> Isn't the CPP dance obsolete now that we no longer preprocess makefiles?

Maybe. Doesn't it affect some of the later tests that configure runs?

bug#11485: In configure.in, gl_EARLY is too early?

[Glenn Morris]

>> Isn't the CPP dance obsolete now that we no longer preprocess makefiles?
>
> Maybe. Doesn't it affect some of the later tests that configure runs?

If it doesn't affect anything, why is configure testing for CPP at
all...