Re: [Simulavr-devel] Docu patches

[Theodore A. Roth]

On Fri, 4 Jan 2002, address@hidden wrote:


:)> test -z "$CONFIG_HEADERS" || echo timestamp > src/stamp-h
:)> 
:)> When you spec the extra-cmds, they just go after this and do essentially 
:)> the same thing.
:)No they doesn't!
:)I have added "echo timestamp > ${srcdir}/src/stamp-h.in 2> /dev/null", which
:)will create the stamp-h.in file in the real src directory. If you run 
configure
:)from an other build directory it is needed.

I'm configuring and building from a separate build directory only. Not 
from the my cvs sandbox.

I would think that the VPATH mechanism would handle the stamp-h in the 
build dir and the stamp-h.in in the source dir. The generated 
build/src/Makefile file already has rules in it to deal with in all the 
directories.

Also, ${srcdir}/src/stamp-h.in is not doing quite what you think it is. 
${srcdir} changes depending on which directory (in the build dirs) you are 
in. If you're in build-sim/src, then $srcdir will be 
../../simulavr-$ver/src and you would end up with 
../../simulavr-$ver/src/src/stamp-h.in. I think $srcdir should be replaced 
with $top_srcdir. (see "Preset Output Variables" section of autoconf 
manual)

One last note, the autoconf manual section for "Automatic Remaking" 
suggests adding 'echo timestamp > stamp-h' to the extra-cmds. Why are you 
using stamp-h.in?

:)
:)> The failure I was seeing was that autoheader was run without specifying
:)> the config dir, thus it couldn't find the acconfig.h file and a few things
:)> where'd added to the config.h file (namely defining UNSIGNED_8_BIT_T and
:)> friends). I beleive the real solution is to just move acconfig.h into the
:)> top level directory and remove the '--localdir=config' from the bootstrap
:)> script.
:)This might be a solution. But I didn't like to change this, to get it easier
:)approved by you:-)
:)
:)> I just verified that this indeed solves the failure I'm seeing. Here's a 
:)> sample of the failure output:
:)> 
:)> ../../simulavr/src/vdevs.h:46: bit-field `T' has invalid type
:)> ../../simulavr/src/vdevs.h:47: bit-field `I' has invalid type
:)> ../../simulavr/src/vdevs.h:71: field `parent' has incomplete type
:)> ../../simulavr/src/vdevs.h:73: confused by earlier errors, bailing out
:)> make[2]: *** [main.o] Error 1
:)I had an other error output. The reason was, that UNSIGNED_8_BIT_T
:)and friends was not declared.

That is the same failure. The UNSIGNED_*_BIT_T stuff is defined in the 
acconfig.h file which wasn't getting added to the config.h file by 
autoheader.

:)
:)> I was seeing this failure occur when I issued a make from within the src 
:)> dir after changing the configure.in file and running bootstrap but not 
:)> running configure again.
:)I have done (in an other build dir) configure and make. The make has
:)regenerated the config.h file (AFAIR), which was wrong, because of the
:)missing --localdir parameter to automake.

Try the 13-pre2 patch. If you can still get the build to fail, I'll add
the AC_OUTPUT extra-cmds back in. I was unable to get a build to fail
after moving the acconfig.h file.

I hope you don't feel I'm being stubborn here. I'm just trying to clarify
something I think neither of us completely understands. I don't want a
broken build system, but I also don't want to have to the build system be
any more complicated than it needs to be. It might already be too
complicated for all I know.

Please continue to tell me when you think I'm wrong (or even when I _am_
wrong), it forces me to think about things more carefully to try to
justify myself or see the evil of my ways. ;)


Ted