[bind10-dev] Inconsistencies in configure behaviour

[Jeremy C. Reed]

On Sat, 4 Sep 2010, Michal 'vorner' Vaner wrote:

> I experimented a little with various parameters of configure and found these
> three problems:

> ? If it is run with --disable-static, the compilation fails with 
> unresolved externals. However, --disable-static is in the output of 
> --help, so it should be either removed from there or it should work.

For me the error is:

../../../../src/lib/cc/data.cc:909: fatal error: opening dependency file 
.deps/../../../../src/lib/cc/data.Tpo: No such file or directory
compilation terminated.

In addition, the custom feature

  --enable-static-link    build programs with static link [default=no]

which does AM_LDFLAGS = -static (to prevent linking with shared 
libraries) also seems redundant and opposite since it has:

  --enable-static[=PKGS]  build static libraries [default=yes]

by default.

> ? CXXFLAGS variable is ineffective. No matter what I set there, it is 
> overwritten with -g, however, CPPFLAGS work. Shouldn't it be that -g 
> is set only if there is nothing in the flags before, like a default 
> value?

Yes, it is overwritten. It causes me problems too. (I modified my 
configure.ac to set CXXFLAGS="$CXXFLAGS "...)

> ? Wiki describes 4 commands to create the configure script (aclocal, 
> automake, ?). However, they do not work for me, one file is missing 
> (ltmain.sh). Short look trough the Internet says I should use 
> ?autoreconf -i?, which works and is said to prevent future 
> incompatibilities such as this one. Does autoreconf work for other 
> people too? If yes, I think the wiki should be updated.

What wiki page(s)?  (Just in case I overlook when searching.)

I use "autoreconf --install" (also documented in the Guide).

> Should I create a ticket for that? Or try to fix it?

Yes, please create tickets for the two configure.ac issues.

> As a side note ? when we measure some kind of performance, do we use 
> -O2 or something? The default is no optimisations, but I think whole 
> STL's performance is based on assumption that the compiler does good 
> inlining.

For my benchmarks, I just use the default as configured by configure 
which is -O2 (for BIND 10 trunk). (This is the default in most autoconf 
"c.m4" macros.) If I should do benchmarks for different optimizations, 
please let me know.


  Jeremy C. Reed
  ISC