Reproducibility, file names, and -frandom-seed

[Matt McCutchen <hashproduct at verizon dot net>]

Dear GCC developers,

I would like a compiler that produces identical object files when run
several times with identical source files and headers.

I noticed the documentation of -frandom-seed in the GCC man page and
found a simple way to make compiles reproducible while still reliably
matching object files with coverage data files: use a hash code of all
information affecting the object file (i.e. preprocessed source file,
compiler version, and compiler settings).  Alternatively, generate the
object file with a random seed of zero, hash it, and regenerate the
necessary parts using the hash as a new seed.  What do you think of
these techniques?

Furthermore, names of source files appear in object files, so if I
compile two identical source files foo.c and bar.c, I get two different
object files.  Names of source files are obviously useful to debuggers
looking for a program's source, but gcc 4.0.2 20051125 (Red Hat 4.0.2-8)
includes them even when -g is off.  Is there a good reason for this?

-- 
Matt McCutchen
hashproduct@verizon.net
http://hashproduct.metaesthetics.net/